1. Running 41km for the first time (of course)

2. Getting people through the charity checkout flow

I’m raising money for the ME association - a cause that affects my mum, brother, godmother, best friend and hundreds of thousands in the UK. What better charity to run for! I’m so pumped.

But a few weeks in, the messages started coming in:

I just tried to donate, will try again

The page was too hard, will try again later.

I filled everything in… was expecting to enter card details, but it didn’t work. I’ll transfer directly.

I thought: what is going on?!

These are super high intent users; they’ve already decided to donate. Friends, family, colleagues - there’s a lot of rapport there.

When this sort of customer base drops off, there’s a big issue. It’s never a messaging or motivation problem, it’s about momentum and friction.

So, to diagnose what’s actually going wrong, it I did what I always do: a UX analysis 🤌

First, let’s take a look how big this thing is.

The platform behind £34 million

The London Marathon broke two world records last year: the world’s largest ever marathon (over 56,000 participants finishing), and raising a record £87.3 million for charities. £34 million of which, came through their partner platform: Enthuse.

I also have to use Enthuse for my fundraising.

And this isn’t a niche tool.

As well as nailing the biggest race in the world, the platform is used by 1 in 5 of the UK’s largest charities.

The stakes are high.

As a system handling huge volumes of money, small inefficiencies compound. A small % drop off on a page means huge dollar value drop-offs in donation amount.

Which is precisely why this 13-click payment flow is baffling.

Let’s take a look step by step.

From landing page to payment

From fundraise page to successful payment, there ‘six’ stages signaled to the user:

In reality there’s eight steps in the flow:

• Fundraising landing page

• Select the amount

• Platform fees

• Create account

• Add your name and message

• Agree to Gift Aid

• Payment details

• Payment confirmation

So what should be a single action (donate) becomes a series of micro decisions, increasing cognitive load and decision fatigue.

The average adult makes 35,000 decisions every day. Hence, every additional step Enthuse puts us through is a chance for users to reassess: “do I still want to do this?”.

And it’s not just the number of steps.

Alongside the length of the flow, the second thing that confused me was the visual clutter.

On mobile, safe to say it is busy.

Many people raising money will post on WhatsApp, Instagram, Facebook - so mobile is where the majority of donators will be.

Users rely on visual hierarchy to decide where to focus, and here that hierarchy is competing with… itself.

Zooming into the start of the journey, there are three big things that slow me down on these first two pages:

• Excess clutter: the landing page has two titles, an image banner and avatar picture, share buttons and cookies banner, leaving a small slice for the person’s content.

• Excess questions: the first question isn’t “how much would you like to give?” but “how often?”. This introduces a second decision early - risky given when users face multiple decisions at once, they tend to think ‘I’ll do this later’.

• Inconsistent UI: the hierarchy and buttons change size, shape and location on each page, meaning it is hard to flick through at speed. There’s no mental model for a page here.

Even before entering a ££ amount, there’s friction.

But my people are keen to donate, so they power through.

Before long, the confusion starts to build again.

Dark UX to absorb platform fees

On the donation amount page, I keep the default to ‘one off’. Then I have to scroll down to choose the amount.

Once I’ve typed in my amount, an orange module pops up which automatically adds 30% to your amount for the platform.

Dragging the smiley face left, the minimum it feels like you can add is 10%, until you see the hyperlinked ‘other amount’.

Clicking ‘other amount’ brings up a wordy page describing how the extra helps.

Within this flow, it takes me two to three steps to understand what this mystery charge is.

Unexpected costs can trigger loss aversion, where users become more sensitive to what they might lose than what they intend to give. Even when the fee is optional or justified, the timing creates friction by breaking the user’s mental model of a simple donation.

Nevertheless, I power on….

Next, I have:

• Log in

• Write message

• Add gift aid

Which, in the copy looks like another fee:

Followed by a wall of opt-ins:

At this point, the user is presented with a dense set of options. Cognitive load is high. It feels easier to say no rather than yes. Which is bad for the charity.

After the platform fee fiasco, I don’t trust it - but I should, as it’s just gift aid.

For the charity, gift aid adds 25% on donations made by taxpayers, at no additional cost to the payer. For a £100 donation, the charity receives £125. Though the ‘no extra cost’ on the page is hidden.

What could have worked better here is a set of simple questions:

• Are you a UX tax payer?

• If yes…

• Add gift aid to increase your donation by 25% for free?

• Great! Before we continue, confirm that: [two extra questions checking it’s not from a third party or raffle)

With nicer copy, I’d have felt less likely to be lazy and just select ‘no’.

Paying feeling old-school

After opting out of marketing, I finally crawl through to payment.

And I’m instantly deflated: there’s no apple pay.

No one link, either.

So I physically get up to find my wallet. The biggest type of friction - physical friction.

After manually typing in my details (which I haven’t done in ages) and accepting the two-step verification, I’m through. With…no confirmation?

To check the sum I’ve donated, I have to tap the ‘Show donation details’.

Very anticlimactic, I expected more of a pat on the back for that effort.

For a platform that has tens of millions passing through it I’m impressed people made it through at all. Especially people like my grandmother, my godmother and more.

Across the flow there is:

• Poor mobile optimisation

• Visual overwhelm and noise

• Excess steps

• Excess scrolling

• Excess friction

Almost to the limit of feeling buggy.

But with 4000 charities using this page for 20,000 events. I doubt this is an accident.

Let’s look at why.

Why charities add friction

None of this is random.

When you look closely, the friction isn’t just poor UX. It’s doing a few jobs for the charity:

• Adding default fees in there in case people don’t realize they’re giving more

• Asking for log-in in a way that makes it feel necessary, but in reality it isn’t

• Slowing users down just enough to increase contributions, but not enough to fully exit the flow.

When you’re adding 25% here and 30% there, the upside is so large that the friction has to be enormous to cancel it out.

Even if my godmother and family friends text me that they’re struggling, Enthuse is still probably making more from this.

This flow sits in a sleazy middle ground between optimization and obstruction.

If this flow was truly built for users in mind, it would be simpler:

• Enter amount

• Apple pay / one link

Boom. Done

As a fundraiser and giver, that’s what I want. Fewer steps. Clearer choices. Less thinking.

But probably… less money captured along the way.

So the real question isn’t “how did this happen?”

It’s “who is this optimised for?”

It’s not optimized for the donor trying to give £20 as quickly as possible.

It is for Enthuse: trying to earn 30% in platform fees

It is for the charity: trying to get 25% more from Gift aid. And some contacts they can warm up later in the year.

And when you frame it like that, the 13 clicks start to make a lot more sense. Even if they still feel incredibly annoying.

If you’d like to try the flow out yourself, here’s the fundraise page - at the time of writing this I’m 20% off my target!!

That came after finishing HubSpot’s onboarding and being left with a long list of setup tasks. It felt like I had to put a lot of manual work into legacy CRM systems before I could use them.

But going back to it last week, I realized I was looking at the wrong thing. I was focusing on the onboarding, UI, what the screens looked like.

I wasn’t looking at how you’re actually expected to interact with the product.

When I onboarded with Attio last week, I landed on a completely different home screen to what I’m used to from most CRMs.

It felt different straight away.

At first, it’s a cold start:

Analyzing data

We are analyzing your email and calendar events…

Then it finishes loading, and I get an aha! moment.

The end of the classic CRM home screen 👋

When my new Attio home screen populates, it’s clear that this is a drastically different format than any CRM has had before.

On my new, populated homepage:

• ‘Ask Attio’ takes the centre stage - a conversational way to work with your CRM

• There’s very little content there, different from the normal overwhelm of CRMs

• There’s a default prompt options ‘prep for next meeting’ if I’m stuck

• My next meeting sits directly below chat, with just enough UI to see who’s coming and grab the link

• I can act on it immediately, either by chatting with it or prepping, through two subtle icon buttons

So I try it. I click ‘Prep for next meeting’.

What’s refreshing is that the replies aren’t generic, ChatGPT-style responses. Instead they’re grounded in customer language that the target persona would use day-to-day:

Put together a solid meeting brief

Now let me look up the people records for the key participants to get their roles.

Where we left off

What’s interesting is that instead of leaning too heavily on words, Attio has opted for a mix of UI and button-led design to nudge users in the right direction.

ChatGPT and Claude’s responses are 95% text, whereas with Attio there’s always something to click: a button, or a shortcut.

And if you go further into the Ask Attio experience, at certain moments you’ll be asked if you want Attio to do the work for you:

• Update a CRM record (for instance, after a call)

• Create new records

• Create tasks with due dates, assignees, and linked records

When Attio does these things for you, it’s leaning into a more agentic direction (completing tasks, not just giving information).

But, this only shows up later, once Ask Attio has built trust in the basics: that it understands your context, pulls in the right information, preps you properly.

As a user, I can see that trust being built in the small details like the ‘thinking’ UI:

Let me search for some records in your CRM. I’ll look across both People and Companies.

Let me gather information about the meeting participants

Now let me search for any emails

I have enough information to put together the briefing

With each prompt, I:

• See the full thought process

• Have to wait for it to run through all it’s steps

• Am told where it’s searched and why

• I’m given the count of the number of results

• I can see the tool’s assessment of whether it has enough information for me

This is thelabour illusion effect at work 🧠 When you can see the work being done, the system feels more reliable.

This trust building is especially important when you want to be automating more of a customer’s tasks in the future. Instead of throwing users straight in the deep end of automation, it’s built up slowly over time.

I come away from my first experience with Ask Attio so intrigued.

CRMs are already packed with stuff, how do you add more without adding to the mess?

The design debt of CRM tools

What struck me actually wasn’t the UI details of Ask Attio, it was that launching this sort of feature is really hard to do in a CRM.

CRMs have classically been some of the busiest interfaces.

They try to structure messy, human relationships into:

• fields

• stages

• pipelines

• objects

You end up with so much structure, parallel navigations, endless customization, visual onslaught.

Old CRMs face a huge design challenge: making something complex feel simple again.

And not just making it simpler: making it smarter.

In the pre-Ask Attio model, the entry point is the structure.

You go straight into a:

• Contact list

• Pipeline view

• Company record

The person is still piecing together the information themselves, and analyzing where a deal is at.

In the new model, the entry point is the question: you start with what you’re trying to do, and the CRM then pulls across your entire context to deliver that for you.

If you want to go deeper, the detail is still there (you can still go into the contact or the company). It’s just not where you start.

The challenge: work tools follow mental models 🧠

The problem is that people have been using clunky CRMs for decades.

You don’t open one expecting to type a question; you expect to click around, find a record, update a field.

That’s the mental model 🧠 You already know how it’s going to work before you even touch it.

When the stakes are high (revenue, income, million-dollar-deals), the appetite to try new things is lower. Customers are more likely to go for similar-looking tools.

That’s exactly what Attio did; they started with the expected model.

A CRM that looks like a CRM.

Familiar enough, but also clean, minimal and simple.

That was the ‘in’: a cleaner CRM that removes a lot of the manual work, and they’ve grown quickly off the back of it.

Last year Attio was one of the fastest growing startups in Europe:

• Grew annual revenue 4X

• Raised a $52m Series B

• Reached over 5000 customers including the likes of Wispr Flow and Granola.

By starting with a familiar face, Attio built trust as a CRM.

Which buys them brand, customers and momentum to then pave the way for a smarter type of CRM - one where you go to ask questions, not just update fields.

In the old model:

• The entry point is the detail

• The user keeps everything in order

• Data is entered, structured and maintained manually

• Admin work is created

• Entire roles exist to maintain the system

In the new model:

• The user starts with a question, not a field

• Interpretation and analysis becomes the starting point

• Responsibility moves from the user maintaining the system to the system doing the work

• The system pulls context together for you, but you can dive into the details if you wish

Seeing this, my previous way of thinking about AI product design doesn’t quite hold up.

What smarter CRMs tell us about designing AI products

Two weeks ago, I wrote about four ways AI products are being designed:

It made it feel like you had to pick one direction.

But looking at Ask Attio, that’s not quite what’s happening.

Ask Attio combines chat, UI, and actions in one experience.

Designing AI products feels less like a matrix, and more like a timeline:

1. Start with what people know - a familiar UI, something tangible to trust and understand

2. Layer in AI - features, background work, assisting features but not taking centre stage

3. Introduce chat - a simple way to interact

4. Allow the system to act on your behalf - consent to actions taken for you

Ask Attio feels like stages 3 and 4 of this timeline, but stages 1 and 2 are still there in the contact view, company view and the classic parts of a CRM that exist in the navigation.

The most interesting part in my view is how UI is layered into the chat.

Ask Attio doesn’t just have natural language UX (words), it has buttons, default prompts, UI modules.

Instead of leaving you with a blank input, it gives you ready-made options.

This is so important because when a new mental model is introduced, especially one that moves away from familiar systems, the product needs to guide the user more actively so they don’t get stuck, they activate and they get to the aha! moment.

The big product question from a becomes:

How do you introduce a new way of interacting without overwhelming people?

The answer?

Ask Attio keeps familiar mental models in place and changes the starting point. It steps back and focuses on the customer jobs: closing deals.

Then works out how the information and context in the CRM can support that better.

And builds the interface around that, without getting stuck in the decades of CRM design debt that’s come before it.

In partnership with Attio. Thanks for the early sneak peek into the feature 🤟🏻

At the time, I had just started at Fyxer. Six months later, I’ve been sitting with the idea a bit longer and safe to say there’s more to say here.

In the original article, I argued that many of the product design levers we’re used to either don’t exist anymore or matter far less.

The AI’s output itself has become the main product. UI now acts as a set of nudges to get customers to use it properly, or to build trust.

At first glance this can feel like there is less to design 🎻

But in practice, the design choices get much more challenging.

Most AI products are trying to do two things at the same time:

1. Build trust, that the product can get your work done, and

2. Remove friction to getting said work done.

Do you see the issue here?

If you remove most of the human effort, the user has to trust the system a lot more to let it do the work.

But if you build trust by letting the user fiddle with every part of the UI, then the effort required from the human increases. Have you really removed the labour?

AI product design isn’t really about UI anymore. It’s about deciding how much work the user should do, and how much trust they’re willing to place in the tech.

Instead of starting with design - colors, buttons, layout - designers now have to take another step back.

And ask who your customer is, how much guidance they need, and how they should interact with AI.

It’s easier to explain when you think of two main axes:

• How much work the user does
• How much of the system’s work is visible

This then pops out…. 🥁 🥁 🥁

The four strategies for designing AI products

Most AI products today sit somewhere between four broad approaches. Each one represents a different balance between how much work the user does and how much they trust the system to do work for them.

Let’s go through 1 by 1:

Top left: AI-assisted features

What these are: AI is embedded inside an interface. Customers use AI it through buttons, menus, templates, or predefined actions. The system strongly guides what you can do next - it is essentially a range of templates and presets.

✅ Pros: Predictable and ‘safe’ for the user, relatively low cognitive load.

😶‍🌫️ Cons: Less flexible use of AI - the system can only do what the interface allows.

🙋🏻‍♀️ Target audiences: non-technical users, general population.

❓ Examples: AI features inside SaaS tools, Grammarly suggestions, Canva Magic Write / Magic Design

Top right: Chat interfaces

What these are: hardly any UI, often just a prompt box. Users interact with the system through language and back-and-forth conversation. Instead of clicking buttons, the user describes what they want and refines it through chatting.

✅ Pros: Extremely flexible. Users can ask for almost anything.

😶‍🌫️ Cons: Work happens behind the scenes, meaning outputs can be harder to trust. Without good prompting, can get inaccurate answers and lack of a paper trail.

🙋🏻‍♀️ Target audiences: General population. Chat is now a universal, familiar interface - nearly 1bn people now use ChatGPT.

❓ Examples: ChatGPT, Claude, Perplexity chats.

Bottom left: Automations

What these are: AI is used inside workflows that run through step-by-step. The user defines triggers and processes, the system then performs tasks automatically but within a structured process that the person can see.

✅ Pros: Transparent and trustworthy. Users can see what the system is doing step-by-step.

😶‍🌫️ Cons: Often requires more setup. Users have to spend time configuring workflows before the automation actually saves time.

🙋🏻‍♀️ Target audiences: Operators, power users, people comfortable setting up systems.

❓ Examples: Zapier AI workflows, Make, Retool workflows.

Bottom right: Agents

What these are: The system performs tasks end-to-end on behalf of the user. Instead of interacting step-by-step, the user delegates goals or outcomes and the AI handles the process. Users can often see the steps so they can check the work.

✅ Pros: Removes the most manual work. The system can plan, execute, and iterate on tasks independently.

😶‍🌫️ Cons: Requires extremely high trust. Users often feel uncomfortable if they cannot see or control what the system is doing. Ofc, still needs QA.

🧑‍🚀 Target audiences: Technical users, early adopters, people comfortable delegating large chunks of their work to AI.

❓ Examples: Cursor, Claude Code, Codex, Lindy, Lovable.

Underneath all four of these approaches are the same two design decisions: how much work the user does, and how much of the machine’s work we show them.

Both of which are product decisions you can influence.

The first decision: UI vs interaction

On one end, you can build products that rely heavily on visible things: buttons, menus, flows, carousels, images. The system is heavy-handed in what it guides you to do in the product.

At the other end of the spectrum, your AI products can be minimal, devoid of UI - just a prompt box. The user interacts with the system through language and back-and-forth conversation.

This gives maximum flexibility. But it pushes more thinking onto the user, who now has to figure out what to ask and how to ask it.

So the first design decision becomes:

How much structure should the interface provide vs how much freedom should the interaction allow?

Are you designing a guided path, or handing users a blank page?

The second decision: visible work vs hidden work

This design decision is about trust.

AI systems can either show their work, or hide it. If everything is hidden, the system simply produces an output. This can feel magical, but it also makes the result harder to ‘get’ or believe.

If everything is visible, users can see every step the system takes. This can build trust, but it also creates mental overload. It defeats the point if you take more time reading through every step the AI took…

So the second design decision becomes:

How much of the machine’s work should the user see?

Think of it like an iceberg:

This ‘visible’ work can come out in thinking UI, like with Lovable:

And Deepseek:

Ultimately, these steps are only for the user. The system doesn’t need to show it, but it does to build trust.

If you show too much, it’s overwhelming. Too little, and it’s untrustworthy. That’s the key design challenge: what’s too much? Or too little?

That’s what coding agents excel at: just the right amount of ‘work’ shown.

Take Claude Code, which has:

• Task steps

• Thinking UI

• Project plans

• Questions to manage up to the user

These are all things we can play with to build trust over time.

But they don’t tell us where a product should sit in the quadrant.

So, where does product go from here?

AI product design is no longer about arranging UI components, the challenge is bigger than that.

Designers need to become more strategic and ask bigger, scarier questions about how people interact with technology.

After coming to terms with the fact that ‘product’ now lives in this weird space between UI and language - where trust and labour pull against each other - I feel like the only guiding light through these decisions is coming back to the customer 🔦

(As always).

Who is your customer?

To what extent do they:

• Try new tools and explore unfamiliar interfaces

• Spend time curating and configuring their experience

• Want something that works out of the box

• Trust AI systems to act on their behalf

• Learn to lean into things over time?

Different answers to these questions push your product toward different parts of the quadrant. Some customers want control, others like delegation, others want conversation as their ‘mode’.

The real job of product design in the age of AI is figuring out how much work the user should do, and how much they’re willing to trust the system to do for them.

These are my thoughts for now.

Who knows, in six months they might change again 🤷‍♀️

What do you think?

A surreal moment.

Is this real work?

Yes, yes it is.

Let’s rewind.

Every product team thinks they’re customer obsessed.

In practice, most research is something done in individual projects and presented back in bullets.

The problem is not that the insights are wrong - it’s that they’re second hand; there’s a gap between the builders and the problem they’re building for.

Shock: intuition doesn’t come from reading. It comes from doing.

When a builder does the research themselves, they don’t inherit an opinion. They’re seeing the problem first hand and forming their own judgement.

Short term, it may feel new, uncomfortable, time-consuming. Long term it means the whole team can make better decisions.

If it’s so good, why doesn’t everyone do it?

Time.

And access.

So, three weeks ago, we decided to try something new.

We started experimenting with watch parties in the office.

What’s a watch party?

It’s where a group of builders sit in a room watching customers.

Companies like Wise, Google Ventures, Skyscanner and others run watch parties to help get stakeholders closer to customers and scale UX research.

For Fyxer, that means eight or so product engineers gather in our biggest meeting room to watch a recording of someone’s screen as they work, sometimes with snacks 🍿 Our product marketing manager sourced recordings from the team and customers, so we have MP4s ready to play.

We occasionally pause to chat about what we see, and we go round at the end to share one friction we noticed at the end.

It was an idea from our CTO to do these twice a week, along with daily ‘product shadowing’ sessions.

Shadowing is where a handful of engineers literally stand behind people as they work. Luckily for us, we have colleagues who are our target customers (so we can explain why it’s not meant to be creepy). For other companies these might be trickier to organize.

In both shadowing and watch parties, we’re looking for friction in peoples’ workflow, and any current behaviors and coping mechanisms to these frictions. ‘Friction’ comes out in many ways:

• Repetitive or repeat tasks

• Context switching

• Excess scrolling

• Excess clicking or highlighting

• Delays or pauses

• Missteps and mistakes

• Number of windows open, tabs, screens

• Time to complete tasks

• Number of steps or destinations in a flow

When shadowing you can ask in the moment why someone is acting the way they are. With watching screen recording, we can’t ask ‘why’.

Observation shows you the “what” with no fluff or lies. Interviews help you understand the “why”. Experiments prove whether any of that actually matters for your commercial metrics.

Hence why no single method is enough.

Where watch parties sit within the broader insights stack

Let’s be clear: watch parties aren’t a replacement for interviews or AB testing.

They’re not analytics. They’re not customer calls.

They sit in the juicy space in between.

They’re for identifying new problems to solve, or ways that your current product isn’t hitting the mark. Because you’re observing behaviour directly, you remove some of the distortion that comes from memory and self report in interviews.

Watching someone struggle with something is very powerful too - it inspires ideas, creates urgency and energy within the team.

But it doesn’t automatically mean the problem is worth solving.

Maybe the person you’re watching enjoys that friction. Maybe the 100-tab Chrome browser gives them a dopamine hit when they close them all 🫠

As a product or growth person, you need to separate:

• Problem: is there one?

• Demand: is there interest in solving the problem?

• Impact: does solving it meaningfully move business metrics?

Watch parties help with the first one (sometimes the second). The rest of your testing work can answer the rest.

Here’s a full summary of where watching users sits within other testing methods:

The simplest way I think about it:

Watch parties make everyone in the room uncomfortable at the same time. The rest of the challenge involves working out whether that discomfort deserves to be solved.

To end: the three reasons you should do watch parties

Even though they’re not perfect (but realistically, what is?) here are three main reasons watch parties are becoming a team staple for us:

Reason 1) Problem alignment

When the whole team has watched the same awkward pause, the same excessive tab switching, the same slightly awks workaround, our conversations in the office change from ‘is there a problem to solve’ to ‘how can we solve this?’ and ‘will this have an impact’?

Meaning we’re one step closer to an impactful product.

Reason 2) Operational ease

They’re so scalable and easy on the product ops 😮‍💨

These sessions don’t require 6 hours of my time for 12 separate user research calls where I add an engineer into each.

You can run them twice a week for 45 minutes each, and add as many people to the meeting as you want (until someone’s too far away from the screen to see).

And finally, 3) Distributing intuition

When we’ve collectively watched tens of hours of a customer’s workflow, we’re closer to how they actually work. We start to recognize patterns without needing them explained. We don’t need a deck and wasted time communicating this.

Short term, we have fewer assumptions and more evidence.

Long term, we’re scaling product sense within the team 🤘

What next?

Now you know what snack selection to bring - and how to run them - have a go, let me know how it goes.

Why does Claude feel like a barista whereas ChatGPT feel like the ice king?

Same product. Completely different first impression.

There difference is black and white (or black and orange):

• Warm vs. efficient

• Human vs. mechanical

• Conversational vs. clinical

What’s impressive is there isn’t a huge amount of UI to play with here.

This classic chat screen has been designed, copied and optimized to death over the last few years. The structure is almost identical everywhere: input box, centered headline, black or white minimal background.

Same layout. Same components.

So the differentiation is not the structure. It’s the details.

Let’s break it down.

With Claude:

• The headline is in a Serif font = instantly more soft and human.

• The greeting is large, while the input prompt is understated.

• “Coffee and Claude time?” creates the mood before asking for a task.

• The copy humanizes by using first person ‘how can I help you today?’

• The logo mark sits inline with the headline, warming the page with the orange.

• The two upgrade CTAs are visible but softened, meaning they feel low pressure rather than sale-y.

• The overall vibe is about emotion and connection, not transaction - which feels inviting and fun.

Whereas with ChatGPT:

• The sans serif type feels clean and efficient.

• Black and white UI feels no-nonsense and minimal.

• “What can I help with?” jumps straight to the task. No mood setting, just utility.

• The copy is brief, neutral and functional ‘ask anything’ - sounds like a system not a person.

• The input box is a larger part of the the screen compared to the header, showing that function comes before introductions.

• Action buttons are central (Attach, Search, Create image, Voice) showing off the capabilities.

• The overall vibe is transactional, not relational. It feels ready to get sh*t done.

This UI is optimized for output, not atmosphere.

Now compare that to warm, fluffy Claude.

A vibe which continues through the paid experience:

Happy Friday, rosie

Good morning, rosie

Not just simple name personalization, ‘rosie’, it acknowledges time of day, day of week, where you are in your rhythm.

It makes me feel like Claude is aware and considered, instead of static.

Just like a warm, friendly barista.

On designing tone, not just UI

Whether you have a huge amount of UI in your product or not, what matters is in the details. When deciding what your first moment looks like, consider:

• What’s the first sentence your product says?

• Does it set mood, or just state function?

• Are you building a relationship, or just enabling a task?

• Where does emotion sit, above the input or below it?

• Does your typography and color palette signal warmth or precision? Does that match your brand?

• Are you designing for habit, or just usage?

What your product says first matters. The relationship starts earlier than you think.

Ready to get organized?

Hell yes I am. My slack is a mess.

My sidebar is like a stock market floor, different bits screaming for my attention. Channels are either too alive or too dead. I know I’ve missed some things but I have no idea where they are.

Instead of clicking it, I paused.

How did I get here?

I use Slack every day. I onboarded years ago. I know the basics, I’m pretty hot with the search bar 😏

So why do I feel so chaotic?

I’ll tell you why.

The dark side to productivity tools

Productivity tools have a weird side effect: they create more work.

Take Notion. How many half-built systems do you have? Dead tables? Orphaned pages?

Customization = choices. And with choices comes human chaos:

• Notion gives you infinite structure.

• Slack gives you infinite channels.

• Airtable/Sheets gives you infinite databases.

Customers end up creating admin debt.

And contrary to popular belief a great onboarding rarely solves that.

Better onboarding isn’t the answer

Onboarding is often seen as the ‘one shot’ you have to activate users.

The thinking goes: customers arrive, their brains are like goldfish, you either win them or you lose them in that moment.

After the first 24 hours, a user has gone from newbie-learning mode to get-stuff-done-mode. If they haven’t grasped features that help them use the product well, they’re unlikely to retain as well.

By the time users actually see the mess they’ve created, onboarding is long over.

This creates a churn risk, and a value proposition risk. It’s easier for new startups to come in and clean up the mess:

Tired of [insert old solution]? Try [better solution].

Products need to intercept that negative adoption spiral, before customers start looking elsewhere.

How?

The answer is:

• Intercept users at the right time

• Convince them to try out the feature

• Allow them to adopt the feature with ease

Something which Slack does regularly.

So, let’s see how it does it.

Step 1: Intercept users

The best way to get in front of users is to be in front of their eyeballs. Sounds obvious, but the amount of modules that are hidden in the UI….

Slack goes for the top spot on their sidebar to introduce me to their VIP feature.

What I love about this example is the copy. I don’t get shamed for not trying a feature, Slack simply says:

Lots of notifications?

Yes, so many

You’re so popular

Oh why thank you.

And then the call to action: Choose your VIPs so you don’t miss the important stuff.

Notice how this module is:

• Framed positively

• Feels like a level-up vs. a shame ‘you haven’t tried this’

• Makes me feel seen and important

• Has a ‘later’ button, which is a low friction exit (not a hard ‘no’)

• Feels light and easy

What this means is that I’m softened as I read it, and I’m more likely to click.

A second place to intercept is in the middle of a flow.

Take this module, which asks me to use Slack’s grouping feature as I join a new channel.

Slack has picked this moment because I’m in a flow and this feels relevant. Every time I add a new channel, the risk of noise becomes greater. So I should tidy.

The same goes for this hook: Scrolling for something you might have missed?

Jumping in at the right moment means the issue is top of mind.

But intercepting is only half the battle.

Now you have to make me care.

Step 2: Convince

We are creatures of habit, and months or years into someone’s usage means they can be resistant to create new habits. You have to convince them.

Slack does this by:

• Making the copy feel personal (even if it isn’t)

• Styling out the ask

• Making it feel easy

Take this module. It feels like Slack has done work for me by analyzing my channels and identifying I need to tidy them up.

This is the reciprocity effect (I do work for you, so you do work for me).

Sneaky.

It’s also amusing: the emoji’s make it look like my fairy godmother has appeared to tidy me up for the ball ✨✨

On the topic of fun, another way to convince is to gamify.

In another session, Slack gives a three-option question at the bottom of my nav bar:

Want a productivity boost?

What’s satisfying about the design here is:

• The headline is prominent and clearly framed as a question

• Icons support each option, making the list scannable

• Button-like containers clearly show interactivity

The copy is all action-oriented ‘organize’, ‘send’, and feels inviting versus demanding. There’s a hint of a task list here, making me feel like I want to try them all and get that dopamine hit.

What’s key is to rotate your modules - some are tactics that work great the first time, but become noise with time. A gamified quiz question gets boring if overused.

But even if I click, the job isn’t done.

That’s where adoption comes in.

Step 3: Adopt

Once you’ve managed to convince the user to click on a new feature, you need to make it incredibly easy to follow through.

Slack does this by bringing the settings forward into the module, and keeping flows short. Take this example feature:

• Slack introduced me to a new feature ‘dense mode’ through a popup

• I clicked through the carousel telling me what it was

• At the end of the carousel, I was offered to change the setting then and there

The friction to try it out was minimal, meaning it took me a few seconds to give it a spin.

Notice how this UI gives me:

• Two options

• Clarity on what’s the default vs not

• Save button as the only next step

The only challenge here is that I can’t see my view change as the pop up takes up 30% of the screen.

Stepping back from the individual modules, there’s something bigger going on here.

Takeaways on post-onboarding adoption

This isn’t about tooltips, it’s about preventing churn.

It’s about those moments, months down the line, when doubt creeps in. When you wonder whether the product still works for you.

My favorite thing across all these examples is the positioning. Slack doesn’t tell you you’ve failed, it’s assuming that confusion is inevitable.

Slack knows I’m overwhelmed.

As a result, it can honestly design for my second wave of learning, noise reduction and focus.

If you’re designing for feature adoption, here’s six things to try from the Slack example:

1. First, try the obvious: put something right in their line of sight. But if you do, make sure the copy actually speaks to them vs. promotes your own agenda

2. Second, try a prompt in a journey, ideally where the pain is felt (so it is top of mind)

3. Mix up the pattern. If every nudge looks like every other module, people stop seeing them.

4. Write in a way that reflects how your user sees themselves. Slack doesn’t talk to me like I’m incompetent. It talks to me like I’m busy and important and just need a bit of help filtering the noise.

5. Do some of the work for the user. Don’t just say “you could organize this.” Show them what needs organizing. Make a recommendation. Reduce the effort before they’ve even clicked.

6. Make the onward journey incredibly easy, so that the user can try it with ease

Seen other good post-onboarding nudges? Let me know if the comments ⌨️ ⌨️ ⌨️

Till next week 🫶

I’ve just hit six months full time, and in the last few weeks we’ve launched our rebrand, released a huge industry report (The Admin Burden Index) and spent a lot of time exploring what 2026 needs to look like.

And to top it off, I’ve stepped into a new role as Head of Product and UX 🫨 🫨 🫨 WHAT.

It’s a role I’m so, so excited about, and one that feels like a natural next step as we head into a new phase for the product.

What I didn’t expect was the knock-on effect: a career epiphany that helped me connect the dots on my last ten years of work.

And it came from one conversation that completely changed how I think about my skills.

Find what you’re excellent at, not just ‘quite’ good at.

It’s rare that career chats are transformational, but this one really was.

I was co-working with my friend Daphne who shared a framework with me from The Big Leap by Gay Hendricks.

In our careers, we often know what we dislike, and we know what we love (most of the time), but the middle? That’s where you end up taking on things that don’t give you energy.

When that happens, you stall in what’s called ‘the Zone of Excellence’. The key takeaway, in brief, is if you can shift into the ‘Zone of Genius’, work stops feeling like work.

There are two axes to this idea:

1. Passion: how motivated does this work make you? How fast does time go? How energized do you feel? How aligned are you?

2. Skill: what are you uniquely good at? What can you pick up quickly and execute better than everyone else in the room? What’s your leverage?

Each of the four areas are zones you can find yourself in throughout your career:

1. The Zone of Incompetence

Low skill, low energy

Work that drains you and slows everything down. This work feels heavy and frustrating. You’re not great at it, progress is slow, and it takes more effort than it should. If you stay here too long, it eats away at your confidence and momentum.

2. The Zone of Competence

Moderate skill, mixed energy

Work you can do well, but others can do just as well. You’re capable and reliable, but you’re not uniquely valuable. Some of this work can be enjoyable, but it rarely feels like the best use of your time. This is often a default zone, not a destination.

3. The Zone of Excellence

High skill, low energy

Work you’re known for, but that doesn’t light you up.

You do this better than most people and often get praised for it. It looks impressive from the outside but it doesn’t feel deeply satisfying.

4. The Zone of Genius

High skill, high energy

Work that fits how your brain works. You are unusually good at this, it feels natural, and time passes quickly when you’re doing it. The impact is high and the work gives you energy rather than taking it. This is where you should spend as much time as possible.

I’ve found myself in all four zones over the past ten years 🫠

At first glance, you might think the zone of incompetence is the biggest risk.

And, from personal experience, it is bad.

Very bad.

I was once in a role where I wasn’t a fit. They hired the wrong person, and I had joined the wrong place. I needed to be highly technical and excellent at delivery without asking too many questions. I needed to play games, stay under the radar, get on the right side of people without being too noisy.

Safe to say, I didn’t last long.

In a weird way, that’s why ‘the Zone of Incompetence’ is not the worst zone. There are more triggers for you to leave this zone.

You either leave, get fired, don’t make it past probation. Whatever it is, this is ultimately better for you in the long run (whilst very painful in the short run, I know).

The most dangerous zone for your career is the one you can sit in for ages.

The danger of being excellent.

Stalling.

That is the ‘Zone of Excellence’.

Why?

Because it is okay. And that is exactly why it is risky.

You can stay there forever. It feels good enough. There is nothing obviously wrong. People want you to stay, because you are doing well.

There’s no external triggers for you to leave this zone. And that’s why it’s called the Danger Zone. Moving from the danger zone, to the zone of genius is the hardest move to make in your career, but the most worthwhile.

For me, a sequence of five moments led me to move between these two zones in the past half year:

1. Wearing so many hats at Fyxer, bouncing around all parts of the business (lifecycle CRM, product marketing, lead generation and GTM, SEO, user research)

2. An audience question at a talk I did at Trainline’s HQ: ‘what’s your super power?’. which made me really think. I answered ‘enthusiasm and diagramming’ jokingly, but the last part had something to it…

3. A chat with my wise friend Daphne, where she shared the zones framework

4. A 1-1 with our CEO, where he shared why me moving into my new role made sense for the business

5. A virtual coffee with a new friend in UX, where we went deeper on the idea that we get put in boxes in our career, and as soon as you get out your box, you can become excellent

From these five moments, I came to realize my value is:

• Seeing the shape of a system before it is fully formed

• Turning ambiguity into something others can understand and act on

• Bridging product, growth, UX, brand, and strategy

• Making complex ideas feel simple

• Creating clarity that speeds people up

• Bringing people along (though still working on this one)

In a nutshell, I am good at turning messy problems into clear systems that people can build from.

Ten years later, I now know my zones of genius. Feels… late? But it makes so much sense - always drawing on the walls in my childhood, how I studied for school through mind maps, why some early jobs felt boring, why freelancing works so well for me, why I love this newsletter and drawing my deep dives.

And it made sense why I felt the urge to move on from past roles where I was either incompetent, competent or excellent.

With this new clarity, there’s only one final puzzle piece I’ve yet to mention.

The final piece: work-role-zone fit.

The missing piece here is where you are working.

What do they need? If it’s not your zone of genius, there’s likely not a workplace-career fit long term. You’ll likely be unable to move into your genius, as staying in your excellence is what’s best for the business.

That’s also OK: when I worked with a career coach in 2024, the biggest thing that remains with me is the notion of walking and running in your career. Different seasons call for different paces. Right now, I’m running.

When you’re moving house, raising kids, caring for family, have a health bump or just need stability, a different zone can be exactly what you need. You might choose to walk.

If a role is 70 percent working for you, that can be enough. Most important is just to be intentional about why you’re there.

I’m also very aware that I didn’t get here alone. People saw things in me before I could articulate them myself, had that space for me to grow and moved me into it. So thank you to those people.

Now, onwards! 🐎 I’m excited to see where it goes.

Yours,

Rosie, Head of product & UX woop!

P.S. This week why not try it out? I hereby set you a task for the very first time in this newsletter 😈

Map your zones:

• What zone are you in now?

• What skills sit in each zone?

If it’s useful, literally map it out (here’s a Figma template for you to duplicate), and let me know how it goes in the comments 💬

P.S.S. Shout out to the (perhaps zero) people who got the Pride and Prejudice reference

The first is excitement.

One 20-minute conversation and hundreds of ideas, followups, snippets to share 🤤

The second is regret.

Regret that I didn’t capture all of it.

I draw what I can in my notepad. I record the call. I write a quick debrief in Slack. And still, the deepest insights 💨 poof 💨 leave my head before I’ve formed the thought.

That used to be the norm.

Then I found Dovetail.

Dovetail feels like it grew under the radar for years since it started in 2017, largely because the founders wanted a lifestyle business in the early days.

Despite the chill start the tool caught traction quickly. In 2019 with just three employees, they had 100s of customers including big fish like Shopify, BCG and Harvard.

At this point they changed track, deciding to accept VC funding. In 2020, Dovetail raised a $4M seed round followed by a $63M Series A in 2022.

Now, the tool has 100,000 users across 4,000 brands including the likes of Netflix, Shopify, and Canva (their customers page is bursting with logos.) On their site, Dovetail claims to

• Save 30% of team capacity

• Give back 30hrs in a week per user

• See a 2.3X return on investment

For $20 a month, that’s nae too bad.

To test it properly, I had to throw my real work at it.

The moment my research pain got more painful.

Last week, I shared how I recently unlocked AI prototyping as a skill. But with that, the hardest part of research moved downstream: getting from customer calls to insight. Prototyping unlocked = more chaos.

Rewind to December. I had the prototype and tens of calls in the diary. I ended up doing 29 interviews, iterating the prototype on the fly. When it came time to share insights with the team, I downloaded the videos and cut them up in Canva.

And it was so painful.

I’ve never wanted to smash my laptop more.

At that same moment, I was chatting to a friend Daphne, and she said, ‘you know Rosie, there’s a product for that. It’s called Dovetail.’

From landing page to first project

I’ve never run faster to a landing page.

From the homepage, I went straight into signup. Next, join team.

There was an old team already there - I went in, had a look, saw it was dead and so created my own.

The workspace step was great for a few reasons:

• Calm visual hierarchy (good spacing, simple colors), kept me in a low-stress cognitive state

• Personalizing ‘Rosie’s Fyxer workspace’ to add a sense of ownership and remove a step for me

• Adding in a trust step with data storage drop down

• Copy that feels so easy to understand it’s like soft butter

After this screen, I’m snapped back to reality with the question:

What do you want to explore first?

Uhhh, not sure?

I didn’t know what a channel or project was. I scanned this module, and didn’t get it. I knew I needed to upload calls so I picked the one with the face 🫠

I then made it into the product’s insights section, where I found an onboarding carousel.

Compared to the lightweight workspace setup, this step felt visually dense, with too little separation between the title, imagery, and supporting copy.

What was smart, was the only pop of color being the AI chat, which subtly draws attention and nudges us towards it without an explicit prompt.

When I finished clicking through this carousel, I was in.

I uploaded my most recent interviews and Dovetail got to work.

Magic moments: tagging, highlights & AI chat

My time to magic moment was longer than most products, given the need to import data.

But once I imported and clicked on a call, it was immediate what the value was: automatic highlights and tagging.

Dovetail pulls out the strongest quotes and prompts you to tag them.

It starts with a few defaults like positive and negative, but I quickly created my own based on what I was analysing: core feature themes, bugs, prototype reactions, general cheerleader quotes. I ended up with around 12 themes.

On the next interview, Dovetail reused those tags automatically. I just scanned through and accepted or declined them.

That alone helped me slow down and comb through each recording. But the bigger magic moment came next: highlights.

Each tag generates its own highlight reel, pulling together all the relevant quotes into a single video of customer snippets. You just paste the link and share it with the team.

This one feature alone saved me a full day in the week before Christmas.

On top of that, the AI chat lets me query anything across interviews, or zoom into one, and turn it into a full-page written and video insight with a click.

That’s it I’m sold.

I was ready to pay.

I was ready to invite my colleagues.

But would they let me?

Monetization friction: just take my money

My first week of use spanned my 7-day free trial and led me up to Christmas eve.

When I came back in January, I found my workspace locked. However, I’d logged into the wrong workspace (unsure how), saw none of my work and panicked for two hours that I’d lost my new tags. I then paid for the wrong workspace.

Luckily support sorted me out.

At that point, I needed to share my work with my colleague but he couldn’t get in.

I knew I needed to pay for him but payment was hard.

Why?

Typically in invite flows one of two things happens:

• You invite, and it’s automatically added to your bill

• You try to invite, and you get hit with a paywall

Dovetail does neither. If you haven’t paid for a seat, the invite button vanishes.

Literally vanishes 😂

If I want to add a person to my team, I have to go to:

• Settings > Billing > Modify Plan > use a dropdown to toggle from N to N+1 users > tick T&Cs > pay > update plan

And then the invite button reappears. Then I:

• click invite > type email > click send

That’s 10 clicks when I just want to invite my colleague to my workspace 😭

The excess friction stopped me from moving fast and was deeply frustrating in the moment. Also confusing, why gate invites?

Perhaps they’d had complaints people were auto added to plans. Perhaps it’s always been that way.

As a growth person, it just screams for an easier invite flow, where payment is either part of it, or the user accepts with one click that they’re going to be added to the plan.

Once the admin friction passed, a different kind of problem appeared.

Lastly, new serenity turns to new chaos

Due to my speed, my workspace got messy quickly.

Two UX people told me not to worry:

Everyone’s Dovetail is messy, you’re not alone

I now have 30 unorganized interviews, in one project, 20+ tags (and counting), 9 messy insights and growing.

And a whole load of UX admin.

In order to make this a place all the team can come to, it needs a spring clean four weeks after it’s inception. So that I can be somewhere not just me and the design lead (who gets it) can hang out.

The UX research bottleneck has got further down the road, now we’re at scaling user research.

As a user, I’m wondering:

• Have I saved more time than I’ve created? (probably yes)

• Have I missed some best practice here? (definitely yes)

• Am I getting the most out of this tool (100% no)

• How should I have set this up for scaling? (we’ll have to find out)

• Is it still worth it? (hell yes).

Zooming out from the user lens to a product lens, a few patterns stood out.

Three takeaways from Dovetail’s early experience

1. Fast activation creates onboarding debt

A quick onboarding helps time to value. But without understanding (or resources) users can build up onboarding debt with your product. The hard part of user research moved from being ‘do it well, quickly’ to ‘help scale this process with my team’. With that, mess just shows up later as organization, maintenance, and clean-up. Keeping things usable over time becomes the real job of the tool.

Between week two and month two, consider:

• Suggesting best practice for account set up (1-2 key settings)

• Email onboarding with ‘power user’ next level skills

• Surfacing webinars for activation

• Offering 1-many onboarding calls

• Giving me a mini course (and an accreditation?)

Without these, I asked the support chat - which is a high-touch flow that isn’t ideal for their team.

2. Balance between product-led growth and billing

If a user wants to invite someone, and they can’t easily see how to do that, we have a problem.

Hiding the invite button until someone upgrades ruins momentum at exactly the wrong moment. I wanted to move fast, share work, and bring someone in. Instead, I had to hunt through settings, billing, plan changes, and confirmations just to get an invite button back.

PLG works best when the product leads and billing follows. When billing leads, you interrupt flow at the moment of highest value.

3. Scaling research needs different support than starting it

Great tools help one person get to value fast (which is hard in itself). Fewer help teams get to value fast.

After the initial setup, the challenge is no longer activation of a solo user. It’s managing multiple projects, overlapping tags, and half-finished insights that pile up quickly. As soon as the workspace gets messy, it’s less likely the power user will share it, and become unusable for the less research-inclined team mates.

My space is embarassing.

Clean workspace = pride in the work = more word of mouth and PLG within a team.

The recipe for team growth is cleanliness (some invite buttons, pretty please).

Curious to know, have you tried Dovetail?

That’s it for today’s newsletter, see you next week!

Lovable was founded in 2023, but last year it exploded, going from 0 to $100M ARR in eight months, then to $200M in the next four.

How-to articles sprouted everywhere, saying “if you aren’t prototyping with AI you’re doing it wrong.”

Uh oh.

Turns out I was doing product wrong for a year then…

I knew I should be trying AI prototyping but felt I didn’t have time - I was busy and working too fast to try something new. The things I was working on were quick tests, not big risky assumptions.

That was until two months ago.

Prototyping used to mean a slow, heavyweight Figma processes dependent on skills I don’t have.

First, you need a UI design with the right layers and frames, then you need to add connections and interactions - one by one. This can easily take over a day (especially if it’s your first time).

In one role, I suggested prototyping a product bet. The response from a seasoned designer was: “that takes too long.”

So, for the past few years I found a workaround.

Something fast, scrappy, and good enough.

Google Slides 😆

It is not interactive but it let me test ideas with real users on the same day. I stuck with that approach until it stopped me learning.

In November, I had a customer insight at Fyxer that I wanted to validate. But the underlying problem was technically hard.

I started the same way: take screenshots, add fake text, cover up some bits (ta-da you have a slide prototype). After showing it to eight users I had learned:

• Customers understanding of the current UX writing

• What’s unclear, what adds value, what doesn’t

• What would they remove or add

• Their gut reaction to a pre-conceived version 2, and a version 3

• What the words mean to them

That got me a long way. But then I hit a wall.

A static prototype won’t show you flows, curiosity, or whether someone will try a feature on their own.

I thought: It’s time

I knew I needed something interactive.

I typed ‘Lovable’ into Google.

Here’s what happened.

Time-to-prototype: 10 minutes 🏎️

Time to first action was quick. Very quick.

On Lovable’s homepage, I got to the AI chat and started typing.

I always start with zero-shot prompting. What this means is I don’t say how to perform the task; there’s instructions without providing any examples of desired input/output pairs. I like testing how good the first output can be without much direction.

In this case, here’s my first prompt (misspellings and all):

I need an email prototype inteface on desktop, it needs to look halfway between gmail and outlook

I need 5 labels categories and 15-20 test emails

The categories are:

(I list out 1-5)

The 15 emails are common emails for someone working in [industries]. I list out examples.

In the main inbox, I don’t want to see marketing. That gets sorted away. In the sent folder there needs to be some sent/actioned mail.

it needs to be able to be interacted with. Click into emails. Read them

After which, I was sent through a lightweight onboarding before I could see my new baby:

The onboarding didn’t feel like too much friction: it was short and stayed out of the way. Account creation happened in an overlay, not a separate flow, so there was no sense of “starting over” or being '“taken away” from the flow.

When I landed in the workspace, it looked almost identical to the marketing site. Same black GenAI chat over a pink gradient, creating a new mental model for what my building space looks like. They’d added a sidebar and some personalization:

Build something lovable

Became

Let’s build something, Rosie

My prompt was already there - nothing to retype or set up. I was one click away, all I had to do was hit the little arrow and generate.

And it really was just that easy.

As I waited for Lovable to spit out my new baby, it told me it’s process:

Here it’s building trust with me:

• Tell me the process

• Share your logic so I can check it

• Take me through the underlying assumptions

• Don’t generate too fast else I won’t trust the quality

With the ‘thinking’ UI, I could see what it was doing, which helped me build confidence that what I was going to see was backed up by some design thinking. It was almost like a mini PRD (product requirement doc) with: features, design inspiration & design approach.

After a few minutes, my first version appeared:

End-to-end, I went from landing page to first prototype in as little as six clicks.

What surprised me most was not how cool the feeling of building was, but actually how well it took feedback.

I followed up 16 prompts to edit, like:

• Can you remove ‘pro mail’ remove any detail like settings notifications. Make it super simple, It needs to look really lo fi, any more lo fi you can make it (remove the blue and make it grey)

• The labels are squished

• Bring the label colors back, with need to have the numbers and full stop

• Can you put the colors on the right label folders too but subtly, like they would look in gmail

• Can you add 10 more sample emails that a real estate or recruitment person might receive

• The labels are not working - when i click on the folders they should then show in the inbox

These aimed to remove distraction but keep the essentials, and make sure the basic interaction worked.

Of all of my 16 edits, Lovable managed to make the exact edit with one instruction. Like when you have a very intelligent intern who just ‘gets’ it.

The only area where I needed to keep chasing was color. Where in the end, one of our product engineers recommended that I give it an actual hex color code:

Can you make the to respond label dot this color : #FFA1A8

This process took around ten minutes.

In terms of product activation, in ten minutes I had:

• Gone from marketing site to fully working product

• Taken the first action

• Reached my first magic moment (first generation)

• Reached my second magic moment (how quick to take in feedback the prototype was)

I was in a place where I was happy to share this with an interviewee. Which was great because I had five lined up that very same day.

Publishing to production = easy peasy 🫛

Very quickly I was on to magic moment number three: deployment.

I assumed this step was where things would slow down. I’m used to hearing of PR reviews, merging branches, staging environments, production rollouts. Which all scream: this takes time.

Whereas in Lovable, I hit “publish”, the big blue button top right.

And that was pretty much it. I saw the little ‘live’ pill confirming my baby had in fact been shown to the world for the first time.

I copied my link. Pasted it into a new tab and voila, my living breathing prototype was there.

Having started this task at 8.29am, I had a working prototype before 9am. It was easier than creating a slide, and 100 times more ‘real’.

Which then brought a the next challenge.

The hard part: a moving bottleneck 🍾

With this new prototyping-as-fast-as-light skill unlocked, the building of the prototype suddenly stops being the hard part.

The real challenge is now downstream: assumption testing. Asking the right questions, silence at the right moments, and knowing what to notice versus ignore in what you’re seeing.

Even with a great prototype, if you say the wrong thing you can lead people and collect the wrong data.

With the first few prototype interviews with my new Lovable baby, I start testing like this:

1. Frame it loosely: “I want to show you a rough, interactive mockup. It’s not a design, just a way of exploring an idea. I’m going to paste a link in the call chat, and I’d like you to click that and share your screen”

2. Then wait. Watch what they do first (if anything). Where do they click. Where do they hesitate. Where does the cursor hover. Count to ten before saying anything.

3. Gut check: “Can you talk me through what’s going through your head?”

4. Action: “What would you do next if this showed up for you, if anything?”

5. Note down their questions: What are they asking? What distracts them? What stops them from trusting or acting?

6. Finally, look for gaps: “What feels missing or risky here?” And, “If you didn’t have this, what would you do instead right now?”

After the first two or three interviews, I made edits to remove distractions in the prototype and improve the examples. As soon as you remove distractions, you start to hear the real insights.

After five of the same type of person, you typically stop hearing net new insights.

That can take longer if you’re iterating the prototype a lot, or testing across multiple industries or ICPs. As a rule of thumb, you’re aiming for five relevant, similar customers. Even with screening, you’ll lose a couple, so it often means speaking to around seven to get five good ones.

So, after edits, iterations, and a second prototype on a new topic, two months later I’ve:

• Done 60 + customers interviews

• Across 2 slide prototypes and 2 lovable prototypes

• Tested 6 assumptions

• Made edits on the go, between calls

And created a new workflow that’s easier, more valid and more accurate than my previous one.

What surprised me most wasn’t the tool itself, but how much it changed the way I work. Less time messing around with the ‘what’ I was going to show, more time learning and talking to customer.

Now that’s a win.

To conclude: the next challenges of testing

The moment my google slides were no longer fit for purpose was the moment I had to level up. But realistically, a lot of people won’t be there yet.

I worry that framing AI prototyping as a marker of a “good” product person risks shaming people into doing research they don’t actually need. Not every problem calls for a prototype, and not every prototype needs to exist.

But, if you:

• Have a problem is technically hard, politically sensitive, or expensive to reverse

• Have no behavioral or quantitative data

• Hear stakeholders debating hypotheticals over and over

• Have multiple viable routes and want a quick gauge on which one could land best

• Want to reduce risk before committing to build

• Are building a product vision that needs to get people on the same page

Then I’d highly recommend spending 10 minutes on Lovable.

And when you do, there’s three things to bear in mind when you start:

Lesson 1: Get something in front of people ASAP.

Imperfect is fine. What matters is getting something interactive in front of people early, before you’ve talked yourself into or out of an idea. Don’t spend ages on all the details, as you never know what people think or focus on before you start.

You can edit and change as you go, and once you’ve reached an MVP prototype that’s:

• Not distracting

• Tests your core assumption

• Has relevant defaults for your ICP

Then you need five customers to test your hypothesis.

Lesson 2: Write your assumptions down before you test

Fast prototyping removes friction but it also makes it easier to test the wrong thing. Have a hypothesis with the underlying assumptions written down in Notion, Linear, Slack (whatever) before you start. Something like:

What’s the big bet?

What’s the hypothesis?

What are all the risky assumptions?

For example: Search

• What’s the big bet? People want to find what they need by searching, rather than navigating through menus.

• What’s the hypothesis? If we add a fast, global search, people will complete tasks quicker and feel less frustrated. This behaviour change will impact commercial metric (retention).

• What are all the risky assumptions?

  • People think “search” before they think “browse”

  • They know what to search for

  • Results are meaningful without extra context

  • Search is faster than clicking through the UI

  • Existing navigation isn’t already good enough

I’d recommend adding what information would falsify the assumption. I normally do this in a table and use ChatGPT as a sparring partner. For example, if the big bet was AI chat:

Lesson 3: Spend more time on your interview script than on your prototype.

With faster prototyping, the quality of what you learn depends much less on the artefact itself and much more on how you run the conversation around it.

If your interview plan is weak, you will just get to the wrong conclusion faster. The same prototype can give you completely different outcomes depending on:

• How much you explain upfront

• Whether you interrupt hesitation

• Whether you lead people toward an answer

• What you treat as useful insights vs. ignore

Spend some time sparring with ChatGPT, ask for feedback on your script:

• What are the best things to ask here?

• Am I leading anywhere?

• What am I missing that I should be looking for?

I often ask: “Imagine you’re an expert UX researcher coaching me” to start off the prompt.

If you’re unsure, test on a colleague first or a partner. Practice makes perfect.

And, that’s it!

Let me know how you get on, I’d love to know what you’re researching and how if you want to share.

Thanks for reading this week’s Growth Dives 🫶 Next week I’m going to run through the next bottleneck: interview capture and analysis, with a new shiny tool I’m using called Dovetail.

Then, I iterated to this:

And now, we’re on Substack:

Seeing these side by side is a good reminder that a year is a long time. I’m much further along than I was this time last year. New brand, double the audience, new job, new routine.

Which is why retros are so important to work out what to carry into the next year.

So, let’s continue the tradition with a 2025 look-back, starting with metrics.

By the Numbers

This year, we’ve hit 2,000 subscribers 🎉

I saw wild growth in the first three months of this newsletter, getting to 1000 subscribers in 100 days since launch. 2025 has grown steadily after that (with one big bump in Jan).

In 2025, on the newsletter:

• 2X subscriber growth

• 50 newsletters published

• First sponsor 🎉

• 217 unsubscribes (all-time) which is ~8.9% overall, and around 0.3% per issue

And on LinkedIn:

• 15,028 followers, up 89% year-on-year (up from 55% growth last year)

• 815,811 impressions (down 22% compared to last year’s 1 million)

• My top post of all time this year with 250k impressions

• 327,816 people reached

You often see people say “we grew to 10k subscribers in 90 days!!” and while that’s tempting to chase, the real win for me is consistency and enjoying the process.

Across those bullet points, 50 newsletters published is the real winner for me as it shows consistency. The sheer volume of work is visible on my Figma board.

And the final metric: Hikes done 🥾

I’m at 238 kilometers hiked this year, up 45% year on year.

Now that’s a win.

Best posts:

Despite not hitting 1 million impressions, 2025 delivered my best-performing post ever. Here are the top three LinkedIn posts this year:

• I ignored Duolingo for 3 months. Here’s how they got me back - and why I (neeeeearly) stayed. (254k impressions, 1149 likes, 131 comments, 42 reposts)

• When someone lands on your site, every extra word, button, or menu is a cognitive tax. (56k impressions, 170 likes, 41 comments, 4 reposts)

• I turned off my WiFi to try and break Slack. Instead, I found one of the best bits of copy I’ve seen. (47k impressions, 215 likes, 28 comments, 0 reposts)

My biggest learning here is you never know what’s going to work, so try out a lot of things and repeat what catches on. Last year, my top posts were all Duolingo, so it’s nice to see more of a spread this year 😅

On the newsletter side, the move to Substack means the data isn’t directly comparable. Still, if I had to guess, the top five newsletters this year were:

• Is this the new UX? A look into natural language UX (NLX), the user experience of language-based AI products (66.5% open rate, 7.9% click rate)

• The complete guide to test methods A macro look at discovery and demand testing methods (66.6% open rate, 8.2% click rate)

• What typeform gets right (and Mailchimp misses) Why the end of a PLG flow is one of the most important parts (66.9% open rate, 3.7% click rate)

• 3 excellent B2B SaaS paywalls from Slack, Notion & Canva Copy, UI and personalisation tactics on paywalls (70.4% open rate, 6.7% click rate)

• Time to Magic moment: Claude, ChatGPT & Perplexity How to activate AI products (67.2% open rate, 6.4% click rate)

Three learnings from 2025

Writing every week, sometimes into what feels like the abyss, is not always easy. So instead of pretending it’s been smooth sailing, here are the three lessons I’m carrying into 2026.

Lesson 1: I should have moved platforms earlier

It’s a classic case of shoulda, woulda, coulda.

If I’m honest, I’d probably be slightly further along if I’d picked the right platform three years ago when I started writing, or even one year ago when I launched the newsletter.

That said, the most important thing is starting. Pick something and begin. That’s what I did, and I’d still recommend it.

So far, I’m really liking Substack. It feels like a mix of social network and writing platform, which kills two birds with one stone.

The downside is my open rate dropped from around 60% to about 40%. That’s likely because Substack uses a different sending domain that email providers don’t recognize yet.

So, I’m going to do a big cull of non-openers like I did on Kit last year:

I’d rather keep the list smaller and highly engaged than inflated with inactive readers. That probably means saying goodbye to the 2,000 subscriber mark for now, but hello to healthier open rates.

If you want to help, replying to this email or adding me to your contacts tells Gmail, Apple Mail, and others that I’m not a bot.

Lesson 2: LinkedIn slowed this year, mainly due to… life

In growth, we expect everything to go up and to the right.

This year, my LinkedIn activity went the other way. The main reason is simple: I started a job.

Posting on LinkedIn already feels a bit toe-curling. I got over that earlier on, but it turns out it feels more daunting when you’re working in-house. There’s a new fear unlocked that colleagues see your posts and disagree, or see you differently. Which is irrational, but it’s there.

That liiiittle bit of extra friction has got me second guessing what to write, not having time to post and just generally going more ‘safe’ with posts.

As a result, I’ve fallen off the LinkedIn bandwagon. But that’s ok, something tells me as soon as I work it into my routine, I’ll get back to it.

Lesson 3: I love presenting!

In Q4, I spoke three times:

• The three rungs of retention at Now, Next, Product hosted by Trainline in London

• Retention webinar with the CMO circle and Customer.io

• Presented at Fyxer on product direction & user research

What these reminded me is that I love presenting.

I kinda forgot this in 2025: I optimized for time at home writing vs being on stage.

I think both go hand in hand, as I’ve realized that when you present, you have to really make what you’re presenting make sense. There’s an extra need to communicate well, get your thoughts in order and bring people on the journey with you.

So, in 2026 I’d like to do more 🥂

So, the plan for 2026

More of the same. Woo!

By that, I mean continuing to focus on what’s worked best:

• Actionable, annotated deep dives

• On cool and interesting products

• Shared here in my newsletter

So you can learn with me, get inspired and take ideas back to your team.

Specifically, that looks like:

• Continuing to publish weekly-to-fortnightly

• Using Substack more intentionally as both a writing home and a place to connect

• Getting back into a regular LinkedIn rhythm that feels sustainable

• Saying yes to more talks, workshops, and presentations when they align with my writing

If you’ve been reading, replying, or sharing posts with friends, thank you! The feedback you share is what drives me and energizes me.

To the long-time readers, thanks for sticking around. To the newer subscribers, welcome. There are plenty more growth dives to come.

Here’s to another year of learning, writing and product-ing. Happy Christmas to those who celebrate, and Happy New Year!

Turns out I accepted a default I didn’t realize was another paid tier.

My mishap came down to Uber’s default choices, personalization flow, and language. With a pre-selected tier and some option shuffling, I picked the wrong one.

Let’s break it down and see how, in four steps, Uber increased my order value by 15%.

First, micro personalization flow that feels harmless

The first thing I remember, was seeing a small module before payment: Trip Preferences.

‘Ooh this is new’ I thought.

I head for ‘temperature’ first - I’m always too hot. And, I get carsick - especially in hybrid or electric cars. With most Uber drivers in London choosing a Toyota Prius or Tesla Model 3, I’m always in for a queasy ride 🥴

So this option was music to my ears. A cool car = less risky for sick-y.

Next, I flick through Conversation, from No preference through Quiet preferred to Happy to chat. I decide to go back to Quiet preferred - it’d been a long day.

Notice how the options are single-select instead of a slider, keeping cognitive load low. The button also has a timer animation which forces me to select before it auto confirms the page, adding urgency 🧠

At this point, I’m thinking Uber has a new micro personalization feature for normal uber rides.

Little did I know I just hadn’t paid enough attention.

What’s interesting is the UX writing on the screen. On one hand, the copy is super conversational: ‘Happy to chat’, ‘cool’. Meaning it relaxes you and draws you in.

On the other, it’s very non-committal. Notice how the the sub copy under Trip Preferences says:

Your driver will be notified and may be able to accommodate your preferences

This design is made for scanning, not reading. It’s easy to miss.

Though, even if I did read it fully, I wouldn’t have cared, as I didn’t realize I was paying more for this feature 💀

The drive: a queasy ride 🚗

So, the car arrives, I get in.

It’s a normal Uber. The seat in front of me is pushed forward to give my tiny legs more room (I’m 5’3).

The driver is lovely - chill vibes, speaks when he needs to explain a diversion. Otherwise, a relatively normal ride.

He’s quiet, it’s not hot in there, but I wouldn’t say it’s cool.

I have to open the window to get cold air on my forehead. The queasiness comes in.

40 minutes later, I get out the car. I survived.

A perfectly normal ride for me, nothing special. And that’s what is odd.

The realization hits: I paid more for the same ride

While writing this article, I look at the screenshots in more detail. I ask ChatGPT if Uber always had this ‘lil personalization flow and I just missed it. To which ChatGPT promptly told me I booked a comfort ride.

Wait, what?

I check my bill and it’s right. I’ve paid for my first ever comfort ride. Thinking back, I’m stumped. What extra did I get? It felt like a tier that’s marketed different but the product experience (my ride) is the same.

I picked the wrong tier. How? I thought.

The answer isn’t hidden in the price or the ride itself. It’s in the defaults. Specifically, how Uber sets them, moves them, and asks you to confirm them.

Three UX psychology principles are doing the work here:

1. Mental models (the patterns we learn and expect when using a product repeatedly)

2. Default effect (how the top or pre-selected option guides choice)

3. Choice architecture (how badges and hierarchy draw attention)

On mental models, we’re used to certain layouts and UI. In terms of Uber’s pricing page, UberX and Uber Pet and my go-tos. One is normally up top, the other you have to scroll for.

However, comfort has taken the #1 slot.

Not only that, but each option moves around. Uber X is sometimes position at 2, sometimes 3 and sometimes off the page. As a result, it’s harder to find the cheaper option, and Uber has broken my mental model of how to use their pricing page.

This instability stops me from using my memory pattern: tap, tap, pay. Causing user error.

As for the choice architecture, UberX is the most subtle tier. The others have pills for:

• Faster ⚡️

• ⭐️ Good deal

Meaning that we’re nudged away from the cheaper tier, towards the more expensive ones.

The last thing here is not about what’s there, but about what’s missing:

• No clear moment of confirmation

• The end experience does not dramatically improve, so it’s hard to notice

• The experience also does not fail hard enough to trigger regret or complaints - it really feels like user error (I felt like such a noob)

What’s impressive here is that many monetization tactics rely on outrage or friction. This one relies on mild satisfaction, invisible changes and forgettability. The experience failed softly, which is why it worked.

This only works because I use Uber a lot. Familiarity lowers my guard and stops me reading every last detail on a screen.

In conclusion: paywall tactics don’t have to be in-your-face

I didn’t need an Uber Comfort ride. But hey, I got one.

What’s wild is how gentle the nudges on the paywall were. Uber didn’t force me to upgrade, it just rearranged the environment I was used to, pre-selected something new and kept quiet about it.

It did it so subtly that my tired brain on a Thursday night missed all the cues. The word ‘comfort’ on the CTA, the fact I went through a comfort personalization flow. By the time I noticed, the ride was over and the decision was already behind me 🎻

This shows subtle UX writing, design choices and psychology can trick people if they (like me) don’t pay attention or use a product regularly.

Familiarity makes you fast, and speed makes you stop questioning what’s already selected.

Turns out I’m 24.

That’s the angle for the 11th year of Spotify Wrapped: your age in music.

But it’s not just that.

There’s parties, sprints and quizzes this year.

According to the team “the experience is more captivating, layered, and revealing than ever before.” It has a total of 6 fan favorite features and 10 new ones, bringing the total moments to 16.

What matters isn’t how many moments there are, it’s whether they feel simple, meaningful, and share-worthy.

The timing matters, too. Since September this year, artists are leaving the ‘garbage hole’ platform, including major label Massive Attack, given CEO Daniel Ek’s investment in AI military tech.

And after 2024’s lukewarm reception, Spotify couldn’t afford another year that felt flat.

Last year I wrote a piece on Spotify Wrapped 2024, and safe to say it felt like a flop -described “a homework project that was turned in late”. It was stark difference to the creative, fun years before. In 2021 to 2023 we had: Genre Burgers, Myers-Briggs-come-tarot-card Listening Personality and Audio Auras:

2023: Genre Burger

2022: Listening Personality

2021: Audio aura

Whilst last year felt flat and surface-level in the execution, it did impact metrics in a positive way.

The first 12 hours of Wrapped 2024 were the ‘biggest they’ve seen’, up 26% compared to the same time period in 2023. On socials, they’re seeing a ‘record number of shares’. A few years before, in 2021 the campaign drove 60 million shares and saw 120 million users engaging with the feature in 2022 (28% of their total MAU at the time).

It’s too early for 2025 stats, so instead I’m judging 2025 Wrapped on the experience: how it opens, how it rewards you, and how it nudges you to share. We’ll run through:

• The opening: the UX to wrapped and first look

• Design: why 2025 feels deeper than 2024

• Motion: the experience that keeps you scrolling

• Interaction: the mini-games turning a recap into a product

To kick off: first impressions.

First impressions: Wrapped goes vertical

For the past few years, Wrapped has been story-style. Tap, tap, tap through the sections.

This year it’s a continuous vertical scroll, which feels more TikTok feed than Instagram story.

Minor detail, you might think. But it makes all the difference.

It feels more like a never-ending arcade: bright & colourful, with so many fun games. Less orderly, more chaotic, and easier to get lost in (in a good way).

Each scroll drops you deeper into the race-car-meets-circus aesthetic.

Once you notice the scroll, the next thing that hits is the design direction.

Second impressions: Texture you can almost feel

The design is so layered you can almost feel it through the screen. There’s something about the colours, patterns and shapes that gives it depth.

Compared to 2024, there’s such a stark difference - one that I can’t quite put my finger on.

Visual design isn’t my area so I did a bit of research here into what is actually creating that depth. A few things stood out.

2024:

• Lots of neon gradients and glowy shapes, but not much variation beyond that

• Most screens follow the same structure: text centered & decoration around it

• Feels clean, but a bit flat and same-y

2025:

• More patterns (stripes, spirals, dots, optical tricks)

• A mix of line styles: wavy scribbles alongside sharp, straight geometry

• A blend of shapes & sizes: tiny dots next to huge circles and full-page patterns

• Both soft and bold colours. High-contrast black and white with bold accent colours (red, green, purple).

• Mix of art styles: geometric, hand-drawn illustrations, sticker-like icons that create a layered feel

• More negative space and more variation in composition, less ‘same card again’

As a result, 2024 feels like a set of nightclub templates whereas 2025 feels like a whole world you’re moving through.

And, when you layer in the animations, this world is taken to the next level.

Third: Motion design with personality

The animations do three jobs: pull you forward, add depth and turn separate screens into one continuous flow.

Notice in this first screen that the pacing goes from intense moving, full-screen patterns to calmer, spacious screens. This is called ‘breathing’, and is the same technique as using whitespace: to bring attention to the things that come next and avoid overwhelm.

There’s also parallax working here. Different layers move at different speeds, which makes the visuals feel almost 3D.

The best animation design is the artist sprints.

They’re kinda ridiculous in concept, but also a really creative way to show change through time. My iphone vibrates when the artists start too, like the bang of a gun at the start line.

The same goes for the files section, towards the end of my Wrapped. I can flick horizontally through files that give me a snapshot of times in the year.

The little file waving and pointing arrows inviting me to tap on my report. For me to then see a tidal wave of folders. In each there’s a little nugget:

• Biggest music listening day

• Most diverse day

• Most nostalgic day

It feels like a music photo album. My “most nostalgic day” was exactly that, a friend had just re-introduced me to Fela Kuti, and Wrapped reminded me of that moment.

It also allows me to scroll horizontally, breaking the vertical scroll pattern, and making it feel like a hidden treat.

Yummy.

Fourth: interaction design

Unlike last year, you can interact with your Wrapped. And I mean more than just a tap or a scroll.

I actually missed this the first time, but when I went back to my top song, it turns out it is a live quiz. I selected one and was met with a red cross told: failure builds character.

Very funny.

There’s also wrapped parties where you can compare your Wraps live with others. In this way, Wrapped feels more competitive. It also moves wrapped from being single player to multi-player, adding a bit of gamification with a leaderboard.

There’s a tiny network effect here: the more friends participate, the better the feature gets. This is such an interesting one I think it needs a whole launch of its own, else it flies under the radar.

Finally, Spotify has made it easier to re-enter Wrapped. From the Wrapped tab, I can jump straight to specific sections instead of scrolling all the way through again. It turns Wrapped into something you revisit, not just something you look at once then leave.

What’s the impact of these small interactions? Wrapped stops being something you watch and becomes something you do.

The quiz adds play, the party adds people, and the navigation adds replay.

The culmination of all these this is that the wrapped feels stickier, more shareable, and more likely to pull you back in after it ends.

To conclude, is this the best wrapped?

It’s very hard to compare Wrapped year-to-year. Expectations get higher, the path is well-trodden and there’s a pressure to bring in fresh concepts each time.

Last year Spotify laid off a lot of the team, and as a result last year was a particular low. That context makes it easier for this one to stand out.

However, if we’re talking about the most detailed, most engaging, highest complexity, more interactive Wrapped, then in my view 2025 wins.

On concept, I’m less convinced.

“Music age” is an incredible hook, but it doesn’t stand out as much against the sheer mass of other features. The team went big this year, and I wonder if the core theme could have been pulled through better.

In any case if you’re building your own annual recap, here’s three lessons to take away:

1. Make it a world, not a slideshow

2. Depth isn’t more stats, it’s better pacing, better interaction, and a clearer story

3. Design is what turns data into a moment people want to share.

Side note: I personally loved writing this deep dive. I’ve not gone so detailed on interaction design or visual styles before - if you’re someone with expertise in that space, would love to know any extra insights you see ✨

Awkward.

Then you have to make the decision: do you cut off the call?

You probably should.

Speaking to the wrong people leads to misleading data, wasted time and the wrong product direction.

Talk only to power users and you’ll build for edge cases. Talk to people outside your target customers and you’ll build things your core users won’t want, use or pay for, which means no real impact on revenue or retention.

It happens all the time.

In a survey of researchers, recruiting the wrong people was the most common mistake with 17.6% of researchers having recruited the wrong people.

Personally, of the 20 or so calls I’ve had in the past four months, I’d say a good 10% - 15% of them weren’t the ideal target customer.

Why?

Well, I went for speed. And volume.

People worry that screening or strict criteria will mean your sample is too low. Or you’ll stall time-to-live for your research. In reality, 2 great calls is better than 6 unfocused ones, as the extra noise only creates confusion.

To get great calls in the diary, you need to screen.

And, a few weeks ago I came across one of the most detailed screening flows I’ve ever seen. It was in Figma, and moved through four parts:

1. In-product mini survey

2. CTA to join research program

3. Screening survey in a browser

4. Invite to further research (42 days later)

So, let’s break down that flow step by step, and show how you can screen customers more effectively, whilst also still getting enough calls in the diary.

1) First, a one-question survey

There I was in FigJam mapping out a retention flow.

As I was adding another post-it, I saw a little pop up bottom right of the screen.

It’s subtle, probably around 15% of the screen and blends in nicely to my workspace in terms of color. The module says:

👋 Hi there. Which of the following use cases apply to how you typically use FigJam?

Followed by eight options:

• Analyzing experience gaps and opportunities

• Facilitating meetings and workshops

• Gathering context

• Mapping systems, flows, or experiences

• Designing and iterating

• Planning projects

• Brainstorming ideas

• Other

As a fast follow, I’m asked a second question:

Which use case would you say is the primary way you use FigJam?

Followed by the same eight options, in a different order.

Single select is always more challenging for people, hence why this opens with a multi select. And, because I’ve already started the survey with the easier question, I feel like I want to continue due to the sunk cost effect 🧠.

What’s good about this module is that:

• The questions are easy to understand in a heartbeat (no re-reading)

• The options are concise (max five works, minimum two)

• It’s embedded in the experience (I’m literally mid-task, so the answer is fresh in my mind)

• It’s visible but not in-the-way

It also feels slightly gamified, I don’t know what’s coming next, so it feels fun.

Carrying on, I’m asked three more questions:

• How effective or ineffective is FigJam in helping you explore and visualize ideas or systems?

• When using FigJam, do you tend to work alone or collaboratively with others?

• When using FigJam collaboratively, do you tend to work more live, in real time or asynchronously?

What Figma has done here is:

• Check the core job I use the tool for (visualization)

• Rank that core job out of 5

• Put me into a persona of: solo versus collaborative user, or sync vs. async user

Expertly done.

However there’s one catch.

I answered with the middle option for the last two questions. Why? Well I use FigJam both alone and with others, sometimes sync, sometimes async. For me, it’s less of a yes or no, more of a timeline.

My workflow starts with solo async, builds to sync collaboration work, and ends again with solo planning work.

So, whilst this survey gathers data, it might not be able to screen me properly given how complex workflows are.

I end on a thank-you screen that invites me to join their research panel.

I feel ✨ special ✨

So, I hit Join program, and I’m taken to an 11-part screening survey so Figma can find out exactly who I am. And whether I matter to them.

2) Figma’s screening survey

Clicking the blue button takes me out to a page where I enter my name and email.

Next I see a series of 11 questions that start by asking for age and language requirements, followed by my role, company and use of AI at work.

The first few questions are standard, but these ones caught my eye:

• How involved are you in the decision making process at your organization when it comes to the evaluation/purchase of any product design or development tools?

• Does your design team standardize their work with a shared “design system”? If so, what is your role in contributing to the design system?

• Do you currently use AI or machine learning at work?

With this question set, Figma are able to pick out:

• Organizations with more mature design practices (like design systems)

• People who use AI today or are open to adopting it

• People whose role or seniority allows them to contribute to tooling and spend decisions

And they’re able to weed out:

• Solopreneurs or individual users who may not represent team workflows

• People who have no interest in AI or are unlikely to adopt future features

• Teams that are too early in their design maturity

• People who sit far from design or engineering and would give less relevant input

Added in with the workflow questions from the original survey (solo vs collaborative, sync vs async, visualizing vs facilitating vs brainstorming), Figma has a huge range of segmentations they can use across lots of research questions.

At this point I was curious what they were researching.

** cough cough ** AI ** cough**

And that curiosity was answered 42 days later when I was called into service.

3) The call up email from Figma

Over a month later, I receive an email with the subject:

Talk to us about Figma AI experiences!

The email asked for a 60-minute research call, which I didn’t feel like I had time for without some sort of compensation (though I was curious to see what questions they’d ask, or if I’d see a prototype 👀).

What I suspect here is that their screening system has been used. The copy states:

“We’re looking for input from users like you”

I wasn’t chosen at random.

I was selected because my answers matched the exact profile they were researching. I work collaboratively, with a team, in a product role, at a company that already uses AI, and I make decisions on tools.

My guess is that they were exploring AI features with potential monetization implications or a direction that could prompt switching from another tool. Hence why decision makers need to be involved.

What also stood out was the timing.

I received the invite 42 days after filling in the survey, yet that didn’t feel like it messed with momentum. Once you have a bank of pre-qualified and engaged people, research becomes a pull system rather than a push system. You bring in the right people when you actually need them.

But what if you don’t have enough users?

For this to work, you need enough participants. As of March this year, Figma has over 13 million monthly active users, so there’s no issue in recruiting there.

Most early stage startups and some scale-ups might not have the numbers to make such an extensive screening system work.

If you’re unsure, some quick maths helps look at how much friction you can add to screen people before you get diminishing returns.

I’ve seen call booking rate from emails at around 0.5%-2% from a user research email. Working backwards, if you want 5 calls booked, you’d need to email 1111 people, assuming:

• 2% email click rate (levers: number of links/buttons, size of buttons, copy, offer, sender)

• 30% meeting page booking rate (levers: number of slots, time zone match, copy on page, profile pic)

• 25% drop off rate (levers: email reminders, length of call)

With these numbers, adding friction feels risky.

But arguably, talking to the wrong users is riskier.

If you need a lighter approach, there are simpler screening flows that still work, such as:

• Email → Typeform → Calendly (only if they meet criteria)

• Email → Calendly with a couple of qualifying questions

• Email → Calendly with no questions, but very clear copy about who you want to speak to

Ideally, cohorting who you send the email to will help to narrow down too. I’ve also seen copy that makes people feel special work really well, things like “we’re inviting you to our advisory group”, “you’re one of 100 who has been selected for our user advisory board”. This can increase conversion to call, as well as offers ($30 for 30mins, for instance).

So it is possible to do this without a huge audience. It just takes more care in how you design the flow, the copy, and the ask.

In conclusion: save your time and screen

Talking to the wrong users is like building the perfect chart from the wrong dataset. No matter how well you run the interview, the output tells you nothing.

Screening is crucial to get the right folks on a call and make better product decisions. Figma shows the extent to which you can screen, and the detailed segmentation you can get out of it if you ask the right questions.

Most of us don’t have 13 million users. But even asking one to three focused questions and targeting carefully with clear copy can reduce the risk of a poor call.

Here are the top 5 takeaways from the Figma flow:

• Keep questions short and light (no double-takes or re-reading)

• Keep the options multi choice where possible (instead of free text)

• Mix behavioral personas with demographic ones

• Know your buckets beforehand and choose questions that tease them out

• Offering no compensation for a 60 min call is risky, however if people feel special, they might be tempted. We all love an insider look.

Notice anything else? Curious to hear.

Nearly 1 billion runs were recorded on Strava in 2024.

And it’s a market that’s growing — according to Strava, running was the ‘fastest growing sport globally’ with ‘Gen Z turning to running to create community and connection’.

Which is the winning strategy for coaching app Runna.

If you run, you’ve probably heard of Runna. The consumer app is all over Instagram, TikTok, and Strava, with 48,000+ ratings at 4.9 stars on the app store. The Reddit community alone now has around 47,000 members.

Officially launching in March 2022, Runna reached 7,000 paying subscribers a year later and has since helped millions of runners complete their first race.

In April 2025, Runna was acquired by Stava for an unknown amount. Though estimates are around £150 million, based on a 30X return to early crowdfunding investors, who invested when it was valued at £5 million (30 X £5M = £150M).

That’s when I decided to see what all the hype was about.

The acquisition caught my eye, and so did the fact that my friends had just signed me up for a half-marathon 🫠 As a Strava paying subscriber, I got the email about a Strava × Runna bundle, it made sense to give it a go.

And this is that story: my journey through Runna’s incredible onboarding.

We’ll see three big lessons across the onboarding flow, covering:

• Why most product marketing tests fail to have an impact, but Runna’s packs a punch

• Where user-generated content and community can come into onboarding

• How having a 12-minute, 30-step onboarding isn’t a bad thing

By the end we’ll see that there’s nothing flashy about Runna’s onboarding. It just makes you believe you can actually do it.

Let’s jump straight into lesson one, starting from the first thing we see: a welcome screen carousel.

Lesson 1) Show your product clearly

Most apps try to look impressive. Runna just tries to make sense.

When I open the app on my phone for the first time, I see an onboarding carousel which looks slightly old-school, but I love it.

The classic saying is that ‘no one reads an onboarding carousel’.

Which, in my experience is half true.

Studies have shown users don’t have time to read: most people read 28% of the words on a page; 20% is more likely.

However, I’ve added a carousel before a paywall before and it improved conversion to trial. So there’s something to be said for adding extra context for customers.

In my view, Runna’s carousel works here. It does three things really well:

• It runs through screens like an Instagram story (no tapping required)

• There’s very few words (max 9 in each headline)

• Imagery does the heavy lifting

The Instagram-style story creates a variable reward 🧠 as you don’t know what’s coming next. It also follows a common consumer app mental model which works well in this context.

What stands out here is the imagery. From the second carousel screenshot, it’s clear to me what a ‘personalized running plan’ might look like.

As someone who gets injured all the time and has no real game plan for running, this structure, clarity and format made me sigh with relief.

It created a mini-magic-moment for me early on.

On the fourth screenshot, my running injuries came to life:

Runna has used the most common running injury (Achilles Tendinitis) as an example, with micro copy:

• Get back to running quickly and carefully

• Returning to running post-injury

They know that this is one of the biggest fears, frustrations, or anxieties that holds people back from running. Whether it’s past injuries or future injuries, with these images, Runna is signalling:

We’ve got you

Does this impact conversion? Hard to say. My gut says no.

In my experience, I’ve rarely seen adding imagery into onboarding improve conversion.

That’s because it’s really hard to do well. You need to have excellent image designers and really think about what will make people have a lightbulb moment. Imagery needs to be:

• Easy to scan, easy to see

• Use specific, real examples

• Be able to be understood in a moment’s glance

That’s a tough order.

Lesson 2) Bring your marketing inside your product

Runna’s onboarding starts long before download. Its community and content seep straight into the product journey.

The first obvious thing I noticed is the authentic lifestyle imagery and social proof throughout the flow.

After the carousel, I see a load of smiling women and men I want to be friends with sitting on a park bench.

Further into onboarding, after I select the run I’m doing, I see 230 other little avatars ready to run with me.

There are little signals of value add here too:

• Date

• Location

• Run type = flat (thank god)

Before this onboarding flow, I hadn’t checked the run route for the run I’d booked.

With these details (flat ‘elevation of 34 meters’ & ‘230 others running’), the app has calmed some anxieties around being fit enough for hills and showed me other’s are doing it too. Quietly reassuring me:

You got this

And it continues…

At the end of onboarding, I see more runners 👀

Notice how the human imagery varies:

• Full screen vs. small avatars vs. videos

• Stylised shoots, vs ‘taken on the run’ vibes

• Background gradient fade or spotlight image

What’s curious is that these touch points are littered throughout. There are 30 screens I see in onboarding before I see my plan on the homepage.

Too many to show…

Here’s a summary (as I know the above image is too small 🥴):

In terms of where the imagery sits, there’s imagery beginning, middle (almost) and end of onboarding. With a big cluster near the paywall.

What I see here: a product direction that knows community is important. And an onboarding funnel that matches the acquisition funnel (influencer marketing, user-generated content in ads, reddit communities).

Whilst hard to execute, I’ve heard of experiments showing that matching ad creative and copy at the top of the funnel to the product journey, from onboarding to homepage, can improve conversion to paid and retention.

I’m yet to test such a full-funnel play myself, one day perhaps.

Lesson 3) Design for confidence, not speed

Everyone has fears.

My fears are around running (and the dark).

The key doubts on my mind coming into this were:

• Can I run this run I’ve booked?

• Will my old injuries resurface?

• Will I get new injuries?

• Will I get faster? Can I run under 2 hours?

• Can I fit this into my work week?

• Will this app work with my Garmin and Strava set up?

By the end of my 12-minute-long onboarding, I’m clear on all of the above. And honestly, I felt pretty relaxed about it all.

The hard part of testing onboarding is balancing two things:

1. The macro steps (getting you set up, called ‘set up tasks’)

2. The micro details that aren’t crucial, but subconsciously push people along.

On 1. it’s incredibly hard to prioritise what to put in onboarding. As a rule of thumb, if it positively impacts retention (i.e. when people take [product action] we see they retain better) then AB test it in onboarding to see if it’s a causal relationship, not just a correlation.

On 2, it’s even harder to make decisions of what small details to add. What micro copy will calm users, what images can help.

Individual, isolated tests rarely reach significance on these things. And if you’re not close to the customer, you can end up just adding words, steps and noise that get in the way.

For Runna, the macro steps are:

• Set a goal

• Pick a race

• Self select your running ability

• What days would you like to run?

• Predict your running time

• Pick km/mi

From the copy on these steps, my anxieties around fitting the plan into my week, running a good pace and ‘will I be ready’ for my half marathon are covered.

Then it’s the micro details that add confidence. If I pick a 5k time that’s too fast, it tells me the app isn’t built for me.

Which reinforces that I am in the right place, because I cannot run that fast (but I hope to one day).

The post-onboarding plan introduction takes the level of confidence one step further, with:

• Estimated race times

• A guide to the four run types (with a coach video explainer for each)

• The first of those workouts reflected in my plan (starting with an easy run, yay)

The estimated times show me what I can achieve with and without the app:

• Without Runna: 1hr 59

• With Runna: 8 minutes faster (1hr 51)

Again imagery being used here to show value at a glance.

Lastly, the copy on ‘Types of run’ guides is just so simple.

‘The purpose of these runs is to recover and freshen up’

No jargon, no running-speak. None of that elite language that puts people off.

Just plain old english. And again, none of it feels optimised. It’s just plain and simple.

Bish bash bosh.

The summary: confidence is what Runna is optimizing for.

Thinking back over this flow, the funny thing is that it isn’t particularly sexy. Runna doesn’t try to impress me.

Instead, every choice - from unhurried pacing to clear copy - signals that this app is serious about getting me to my goal.

It feels like the restraint on product sexiness or pazazz, and the focus on plain and simple, is what earns my trust here.

The irony about this UX analysis is that this onboarding doesn’t just onboard you into an app. It onboards you into believing you can actually run the race. Which feels more important than the product UI.

That’s what great onboarding does; it earns belief over attention. Runna proves that calm, clear, confident design can still move people - arguably better than a flashy funnel ever could.

What’s the secret behind a tool used by OpenAI, Ramp and over other 15,000 companies?

Apparently not running any AB tests.

🫨🫨🫨

Enter: Linear.

Linear is a project and issue tracking tool for product and engineering teams, started in 2019 by three founders.

It’s like if Notion and Jira had a baby: it’s minimal and design-first (like Notion), but does the heavier product and engineering workflows people use Jira for.

In June earlier this year, Linear announced a $82M Series C at a $1.25B valuation. Which, if the valuation is at a 20x multiple of annual recurring revenue, puts their revenue at around $62.5 million a year.

In their fundraise announcement, Co-Founder and CEO Karri Saarine wrote a piece about their product culture and journey so far called Six years, our way. He shared:

We’ve built Linear without running a single A/B test and without chasing metrics or looking to data for answers.

This feels crazy to me; that they’ve gone 6 years without running an AB test.

They also have no metrics-based goals, just one North Star as a company-level metric goal.

Karri goes on to explain:

“I believe that building companies is a kind of craft. You’re creating something one of a kind. It’s shaped by a very specific group of people, in a very specific moment in time. You can never really replicate that or manufacture it again. And for most founders, it will be the only company you’ll ever build. That makes how you build just as important as what you build. Therefore, the way you craft should be important to you, and not just a means to an end.”

What I find most interesting is how Linear has been able to take over from Jira for certain customer segments.

Linear is preferred by fast-growing startups for how light and simple it is, while classically Jira is seen to be better in larger, enterprise organizations (though this might be changing).

Switching to a new workflow tool is painful (as work tools are so embedded they often become part of peoples’ CVs).

Which is exactly why these customers moving is worth paying attention to.

According to one review on reddit, Linear is “not a magic bullet” instead “it’s just consistently better in design and performance”.

All of that made me curious to see whether the product actually feels as good as people say.

With this mystique of product craftsmanship, strong growth and good reviews on reddit, I was keen to give it a try.

So, when I joined the Fyxer workspace on Linear of course I screenshotted the whole journey.

And the best place to see whether that craft shows up is the onboarding: the first impression.

Three things jumped out immediately from the onboarding flow:

• Frictionless team sign up: no invites required

• Aha moments: how fast you hit your first real action

• The simplicity: how uncluttered each screen is and how naturally they guide you forward

• The language: how soft and gentle it feels, almost like a verbal hug

So let’s start with the first thing that surprised me: the lack of invite triggers.

First surprise: I never got an invite to Linear.

As I was pasting screenshots for this deep dive into Figma, I was searching and searching through my gmail to find the invite email for Linear.

I checked my deleted folder. Nothing. I couldn’t find it.

So, I thought:

‘I must have deleted it and cleared my bin, oh well’.

Until, I noticed this screen at the end of onboarding:

Tell your team

Anyone with an @fyxer.com email can now sign in.

What this means is that anyone in our business can join without an invite. Expansion within the team is therefore frictionless and relies on word-of-mouth rather than invite triggers.

As teams share links to tickets and builds, the rest of the team will naturally join.

With this flow, there’s no:

‘Can someone add me to the team’

‘Who has invite permissions?’

‘I still don’t have access, can someone help?’

Once teammates join, Linear will bill based on team signups from the same domain. $10 per user per month for basic, or $16 per user per month for Business.

For this to work, people need to use the product, so that Linear links start appearing in the wild across Slack / in messages. Then these projects and conversations will naturally pull team members into Linear (and increase the bill).

That expansion and adoption cycle relies on one thing: how quickly someone gets value.

Second surprise: the onboarding was…unconventional

Most onboarding start with account set-up:

• Sign up

• Name

• Role

• Invite teammates

• Connect integrations

Not Linear.

There’s six main actions in onboarding after signup:

• Sign up

• Dark or light mode

• Keyboard shortcuts

• Join team

• Github explainer

• Invite team

• Subscribe to emails.

After the welcome screen I was asked about light or dark mode immediately.

Then, I was asked to use a keyboard shortcut to open the command menu.

Only then was I asked to join my team, invite more people and check out the github integration.

But giving me some choices early on, Linear is subtly showing me:

• I’m in control

• This will feel like a tool designed for me

• The experience matters as much as the features

The theme choice changes the mood of the whole flow. And most engineers live in dark mode, so letting them switch immediately makes everything feel calmer.

Zooming out, this approach fits their “craft over data” philosophy. A data driven team might not have put dark/light mode first, because (I suspect):

• It doesn’t drive activation

• It doesn’t correlate to retention

• It doesn’t expand revenue

But a craft-led team might put it first because:

• It creates emotional resonance

• It shows taste ’n’ style

• It gives you a sense of agency

• It’s all about feeling over the features

In the midst of this onboarding flow, I completed a tiiiiny tutorial: keyboard shortcuts. Specifically the shortcut to bring up the command menu:

⌘ + K

The visual was so clear: large buttons that look like my mac keyboard, soft language.

It’s the friendliest way to teach a shortcut: one thing to do, calm visuals and big keyboard-style buttons that make it feel like muscle memory.

What popped up next wasn’t as simple, but the language was still soft:

• Let’s try it out

• Type a command

After my foray into shortcuts, I was congratulated:

But also reminded of the shortcut, to make it into a habit. And told what the next most important action is.

Next, I’m asked what team I want to join, with the colleague counts and avatars nearby (a nice little bit of social proof and bandwagon effect 🧠).

Something to note here: the UX is so clean and delineated between joining an organization vs. creating a new organization.

how many of us have accidentally created a duplicate company when signing up to something?

‍ 🙋🏻‍♀️ 🙋🏻‍♀️ 🙋🏻‍♀️

After invites is the GitHub integration page:

Here Linear is saying:

• “We’ll link the PR to the Linear issue for you”

• “We’ll sync the details in both tools”

• “We’ll automatically update the status in Linear based on what’s happening in GitHub”

As a user, I see: less manual work, more syncing.

After all of this, something else started to stand out to me across these screens.

Linear wasn’t showing me much visually at all.

It was… talking to me instead.

Third surprise: words are preferred over imagery

Most of the time, I’d say “use imagery to subtly communicate how things work to customers”, or “show customers value as soon as possible in imagery”.

Not in Linear’s onboarding flow.

Instead of showing you the product, they lean entirely on calm, simple copy to guide you through each step.

On the invite screen, the copy is:

Linear is meant to be used with your team. Invite some co-workers to test it out with them.

On the ‘subscribe to updates’, the copy again leans toward the simple:

Linear is constantly evolving. Subscribe to learn about changes.

And, on the last screen of onboarding, again it’s so simple:

You’re good to go

Go ahead and explore the app

When you’re ready, create your first issue by pressing c

There’s no confetti. No celebration. No ‘omg yay you’re in!’.

Just calm energy. Each screen speaks in short, steady sentences that feel minimalist (just like the design) and confident.

Part of me makes it feel like the start of a mystery movie, where you’re itching for it to get to the real story.

The minimal flow feels like it trusts that you’ll understand when you get inside.

There are other benefits to this simplicity:

• Short sentences reduce cognitive load

• Calm copy reduces emotional friction

• No imagery reduces distraction

• A confident tone signals a mature product

• Simple instructions feel less demanding

• You get into the product faster

Again, craft over design.

In conclusion: a quiet, calm onboarding

Stepping back, how do I feel at the end of this flow?

I feel calm. And, in the moment I didn’t actually notice it much — which is probably the point. The flow doesn’t push, celebrate or demand much from you.

Linear’s onboarding isn’t trying to be impressive or clever. It doesn’t rely on animations, product tours or growth tricks, but instead it shows you that it’s a tool designed for you (as an engineer or product person).

Because of that, it still manages to reduce friction, speed up time-to-value and make team adoption feel natural. You move through the flow without ever feeling lost or overloaded, and before you know it, you’re in the product. Yipee.

For an onboarding with no AB testing, it’s impressive. It shows what can happen when you design the whole thing end to end with a clear understanding of the user, rather than optimizing each screen in isolation, or jamming people through steps to hit metrics.

Will I never run an AB test again?

No.

But will I think a little differently now?

Quite possibly yes.

What this really shows is that there are many ways to build things — and many ways to execute well.

To my newbie delight, the article took off with 9,000 views and 4,000 reads.

More than that, the feeling of getting something coherent on paper and freeing up brain space for deeper thinking was so, so satisfying.

So, I kept writing.

A few months later I had two viral articles in a row.

Fast forward 2.5 years, 87 articles, 720,800 total views, 16,000 followers, $11,000+ in earnings, and I am done with Medium.

From a product perspective, this is a really interesting change because the switching cost is so high (having created a writing routine and spent 300+ hours on the platform).

To understand how I reached this point, we’ll look at the product choices that shaped the experience over time. We’ll see that three things have an outsized impact:

• Monetization strategy: how remuneration works (and doesn’t) on medium, and how it incentivizes the wrong things

• Trust & safety: the impact of bad actors on the writer experience

• Autonomy and reactance: how Medium’s locking down of audience information was the last straw

Each of these has a direct impact on one thing above everything else: writer satisfaction.

And writer satisfaction is the foundation of any writing platform.

Why writer satisfaction matters

Writing platforms are networks. What this means is that each new writer or reader has the potential to improve the experience for everyone else, creating a positive feedback system.

According to Andrew Chen’s book ‘The Cold Start Problem’, there are two sides to every network:

• The ‘hard’ side of the network: a smaller group of people that supply the value. It’s called the hard side as they’re generally harder to get into your product.

• The ‘soft’ side of the network: the rest of the people that are there to consume the value that the hard side provides. Creating a value cycle between the two sides.

Some examples:

• Uber: hard side = the drivers. Soft = the passengers

• Patreon: hard side = content creators. Soft side = the viewers.

• Tinder: hard side = the most desirable profiles. Soft side = the masses matching with them.

• Vinted: hard side = clothes sellers. Soft side = buyers.

• Airbnb: hard side = hosts. Soft side = guests.

• Udemy: hard side = the teachers and course creators. Soft side = students

• Reddit: hard side = posters and moderators. Soft side = readers and lurkers.

For writing platforms like Medium, Substack, Beehive and ConvertKit, the hard side of the network is the writers, the soft side of the network is the readers.

What most people focus on with networks is how hard it is to get them started (hence the name of Andrew’s book: ‘the cold start problem’).

But once a network reaches scale, the priority becomes keeping the hard side engaged, because they are the group that makes everything else work.

When Uber drivers are unhappy, they drive for another product. Once that happens, there’s no cars available. Once there’s no cars available, the passengers follow suit, jumping to another solution versus standing on the pavement. The core user experience then tanks, and retention falls as a result.

For Medium, the challenge is not scale.

There are over 100,000 writers, 2.4 million stories published a month and over 1 million paying subscribers as of 2024.

At their $50 a year pricing, this puts them at around $50 million annual recurring revenue. If the valuation is 20X that, it puts it at around $1 billion valuation.

Which means the problem is not the size of the network.

It’s something else.

The first problem: a monetization strategy that ignores reader-writer relationships

Medium’s monetization model is built on one subscription being shared between every writer on the platform.

Each month, Medium takes the total subscription revenue and divides it through a distribution model based on read time, engagement points, reads from external sources, and a boost bonus chosen by editors.

In reality, few medium writers can explain exactly how the payouts work. Even people who have been on the platform for years are still in the dark:

For the small group of writers who do earn, the income is almost impossible to predict from month to month.

After writing articles consistently over 2.5 years, I’ve seen:

• Average earnings: $138.83 per article

• Highest: $1,433.00 (I cancelled my Calm membership. Here’s all the emails they sent me)

• Lowest: $0.10

• Bottom 25% of articles: $12.37

• Top 75% of articles: $169.88

A few strong articles lift the entire average, and everything else sits far below it.

On a month-by-month earnings basis, it’s a similar trend.

• Average month: $358.74

• High: $1,250 (Oct 2024 and Aug 2023)

• Low: $8.56 (three months around the $10 mark)

What I see here: A few high-earning spikes, a cluster around the mid-range, and a long tail of low-earning months.

The real issue is not the amounts, but the unpredictability.

If earning is your goal, this bumpyness pushes writers to cross post elsewhere to spread risk. Virality is the only reliable path to earn on Medium — and that is impossible to produce consistently.

Zooming out reveals a much bigger flaw too.

Medium’s monetization model keeps readers and writers separate. If a reader loves a specific writer, there’s no guarantee their subscription supports their work.

Instead of rewarding relationships or loyalty, the pricing model rewards reach, length and hooks. What you end up with is content that reflects this pricing model:

• Click bait

• Exaggerated stories

The lower-quality content has been noted by both readers and writers alike, with one subscriber writing on reddit:

Is anyone else disappointed by how many clickbait stories Medium promotes?

I’ve been a paying Medium member for years now, and recently I keep seeing clickbait stories on my home feed. They often have very attractive titles like: « I made xxx$ with ChatGPT and Gumroad », «I published my first book using AI and it’s selling every day »

The top comment crystallizes the writer sentiment in a nutshell:

The core issue is that people are not subscribing to you. They are subscribing to Medium, and Medium decides where their subscription money goes.

This shows a lack of model-market fit: where the monetization model doesn’t fit the way that people use the product, creating a ceiling for growth.

It also means the lifetime value of a paying Medium subscriber is capped. There’s one subscription per reader. On Substack or Patreon, a reader/consumer can subscribe to ten creators, or a hundred if they want to. That means LTV on those platforms could be $1000s per year.

The multi-subscription model scales much better, and also creates competition between writers.

Writers know they have to consistently produce great content to convert their readers, else they’ll go to a competitor publication. The inputs and outputs are clear. The incentive is aligned with the work, and the monetization strategy, not with chasing an algorithm for reach.

Even if earnings had been predictable, another issue was already eroding the experience behind the scenes.

Problem number 2: Bad actors and bots

I know many people who don’t write to earn, they write for the enjoyment.
Part of that enjoyment is replying to comments from engaged readers. It’s what keeps the value loop spinning.

As a writer, I loved spending time replying to comments.

Then the comments changed.

I saw more spam, bots, schemes, catfishing and fake profiles. There was a point where I was reporting comments more than replying.

At the height, I had a 70:30 bot to real-comment ratio.

This is the kind of thing that kept showing up:

I’m not alone here either. The bots are all over medium.

Because of this extra mental load, I stopped replying to comments. Instead of being a place that gives me energy, the comment section was admin.

I’ve seen this happen before with social networks. I once worked for a social network app, and we ran the Product-Market Fit survey to work out where we could improve retention.

For the people who were on-the-fence about the product, the two big things they disliked were:

• Bullying and bad actors*

• Notification spam

*‘Bad actors’ is an umbrella term used in product to describe accounts that breach the guidelines, they are often made for malicious purposes (spreading misinformation, fraud, or abuse).

At the beginning, teams worry about being able to scale enough to get value in the network. They open the gates and make it easy to get in and get up to speed. But if your trust and safety team doesn’t scale at the same rate, bad actors will start to eat away at your retention. Because ultimately, they erode trust.

To solve this, you need people whose whole job is to improve how safe the product feels. That involves working on community guidelines, content moderation, user authentication flows, reporting flows, detection algorithms, banning flows, customer support and more.

Bad actors are one of the biggest risks to retention for social products.

It’s the reason I don’t sell on Facebook marketplace anymore. Too many schemes. Too much reporting.

But, with low pay, unclear earnings payouts, and a load more admin to do reporting bots, I still wasn’t willing to switch.

Until…

The final straw: No audience ownership

Having spent 2.5 years growing an audience to 16,000, I went in to look at all my followers and email subscribers the other day.

I tried to download the 1,300 email subscribers. I opened the CSV and saw 25 names….

Turns out I can only download the people from before April 2025. After some searching, I found that Medium had removed the ability to download your email subscribers earlier this year.

Then it hit me: my audience is stuck on Medium.

If someone subscribes by email, I can’t contact them directly. The only way to reach them is by posting publicly on Medium and checking a box ‘email to my subscribers’.

If they’re followers, I don’t know what they see. I don’t know how often my work appears for them, or whether it appears at all: they might receive a push notification, they might see me in their feed. Maybe not.

After all that time and effort, I realized that the audience I had built wasn’t mine. It belonged to Medium.

As one redditor put it:

Hidden subscriber lists taught me: Audience access > Audience size.

And that was the moment it all fell into place in my mind (and why this article took so long to write).

I realized wasn’t just at the whim of the platform for payouts. I was at the whim of the platform for audience access 😭

This was the final straw because there’s no future where I can write with ease, own my audience, and be rewarded fairly for it on Medium.

To conclude: Medium’s product strategy has lost sight of its users

Ultimately. Medium feels like it has lost Product-Market Fit. Writers are losing value from the platform, and readers struggle to find content they actually want to read.

It comes down to a poor monetization strategy that shaped the wrong incentives, which breaks the value loop between readers and writers.

The interaction between readers and writers is what keeps a writing platform going. With Medium, there are no features designed to strengthen the relationship between both sides of the network. As a writer, I have few tools to engage my audience, and as a reader I can’t directly support the authors I like.

What’s left is the black box of Medium’s remuneration model and their suggested reading algorithm, both of which have lost the trust of the end user.

So what’s next?

For the last year, I’ve been building an owned newsletter audience on ConvertKit, whilst also posting on Medium. Funnily enough I wanted maximum flexibility with my audience and the platform (perhaps as a subconscious reaction to Medium…).

Now I need the middle: enough automation that I don’t have to think about templates (like I do on Kit), but enough autonomy that I can tailor, create and give things to my audience if I want to (unlike Medium).

So, goodbye ConvertKit and Medium, hello Substack 👋