Up
00%
Words
Share
X
,
LinkedIn
Cover
There is a screen inside an app called BitePal, buried about halfway through its onboarding, where the app asks you to name a raccoon.
The raccoon: 61 screens
There is a screen inside an app called BitePal, buried about halfway through its onboarding, where the app asks you to name a raccoon.
The raccoon is small and cartoon and it lives, for the moment, inside a bin. Someone drew the raccoon. Someone decided the bin was funny. You have not given the app your email. You have not paid for anything. You have not been shown, yet, what the app actually does. But the raccoon is there, and it needs a name, and the app will wait.
I named mine Caramel.
BitePal, if you haven’t heard of it, is an AI calorie-counter. If you count every screen between tapping the icon and arriving inside the product, you will count 61.
That is almost two and a half times the overall average, and roughly four times what many onboarding experts would call acceptable. And yet users do walk through all 61, and then pay. BitePal did $400k in its first year.
Everyone agrees, broadly, that onboarding should be short. Nielsen Norman Group recommends skipping it when you can. Growth marketers treat every extra screen as a tax.
So I went to Mobbin — where every onboarding flow from 986 apps is documented, screen by screen — and I looked at the numbers.
The average app runs 25 onboarding screens. The range is 3 to 187.
A whole category of apps was running long. Lemonade runs 143 screens. Monzo, 114. Noom, 113. Philips Hue, ninety-nine. Hinge, ninety-five. Duolingo, seventy-six. These are all well-regarded apps and brands.
At the other end: Google News, two screens. Procreate Pocket, two. Midjourney, three. Microsoft Copilot, four.
So the collective wisdom was right about some apps, and wrong about others.
How many onboarding screens? 25 on average
25 screens. That’s important to know, because most conversations about onboarding don't include a number.
It’s a mean across 986 apps on Mobbin, counting every screen. Even the permission prompts, even the small illustrated pages most users don’t register as screens at all.
Since an average flattens, the better view is by category.
Finance averages 40 screens. Health and Fitness, 36. Education and Lifestyle, 29. On the other end, Maps and Navigation at 13; Graphics and Design at 14.
The long ones need to confirm something before they can begin — a regulatory check, a personal fact. The short ones don’t: you walked in already knowing what you came for.
Some of these categories have figured out what their onboarding is for. Others are still arguing about it.
AI is the clearest case. Almost every AI product lives between 10 and 25 screens. Prompt, generation, user typing within a minute of opening the icon. The category has settled on get to value fast, and most of its apps have landed on the same number of screens to get there.
Finance is all over the place, at 3 to 187 screens. Cleo at 47 and Lemonade at 143 are both tagged Finance, but are not, in any meaningful sense, running the same kind of flow. One helps you manage your spend. The other figures out what to charge you.
25 screens means nothing until you know what kind of app is running the flow.
Platform matters less. Web averages 21 screens, iOS 27 — mostly because mobile apps ask for location, notifications, camera, tracking. Each one adds a screen count.
In every discussion of long onboarding flows, someone in the comments writes: the best onboarding is the one with a skip button.
Not making this up – here's a comment on the video this piece is built on.
It gets upvoted. It sounds right. And it is right, for an app where the user already knows what they came to do. For an app that needs to know who you are before it can help you, the skip button skips the part that makes the product work.
Onboarding length by category
What sets onboarding length, more than anything, is what you must give the app before it can do its job.
Midjourney is three screens. You sign in, you read a line of house rules, and you arrive at a prompt box. The prompt box is the product. You came knowing what you wanted to make. Asking anything else would be wasting your time.
Noom is 113. The app is going to tell you how to eat. Before it can, it needs to know what you weigh and what you want to weigh, how you’ve tried to get there before, how you speak to yourself when it isn’t going well, and whether you believe you can change. I wouldn’t trust a plan from a stranger who doesn’t know any of that.
At the short end, the pattern holds. Midjourney, 3. Google News, 2. Procreate Pocket, 2. Microsoft Copilot, 4. Amazon Shopping, 5. You already have what the app needs — a picture, a search, a purchase — and the onboarding is just where you hand it over.
The long end is doing something else.
Philips Hue runs 99 screens because smart-home setup means pairing with hardware. Hinge, 95, because matching means knowing your taste. Duolingo, 76, because teaching starts with placing you.
Outside Finance, the longest flows are apps that need to know who you are: not your name, but what you want, what you like, what you won’t do.
The first five screens
When the app is going to ask that much — ninety screens, a hundred screens — what does the first one say?
Most people treat onboarding as a preface. The best apps don’t.
Someone opening Midjourney wants to make a picture. Someone opening Noom is doing something harder: they have decided to try to lose weight.
Onboarding is the product’s opening line. Get it wrong and you’re five screens of forms deep, annoyed and losing interest, when you first needed to feel that the app understood why you were there.
The tone is set by screen 2 or 3, and it is almost never revised. You can feel it by the fourth or fifth screen of any good one. Duolingo’s owl shows up and you know, already, that this is going to be the kind of app that cheers when you get a word right. Noom asks you what kind of person you are before it asks you anything else. Headspace opens with a breath. BitePal opens, a little improbably, with a raccoon.
The apps that get it wrong show you just as fast. The screens are not coordinated. The first question is a form field. The second is a permission request. The third is a discount code you did not ask for. Nobody thought about the flow, and by screen five you know it.
Why long flows feel short: Duolingo, Noom
Walk through BitePal’s 61 screens and you notice what they are not: a form. The progress bar appears early and never leaves. You can see that you are a quarter of the way through, then half, then three-quarters.
A form feels long. A conversation doesn’t, even when it takes as long.
In Noom, each question is a slider between two statements.
The first asks what you want. You can either reach a lower weight, or change your lifestyle to a healthier one. You drag a slider toward the one you agree with more. There are four positions: Agree most at each end, and Agree a little more on each side. No typing, no open-ended boxes. Just two statements and a slider.
This goes on for ten questions. Some ask whether you need a program to reach your weight goals, or whether you can do it on your own. The questions are not factual. They are asking who you are, not what you weigh, and the slider gives you room to say you are mostly this kind of person, which feels more honest than radio buttons would.
At the end of the ten questions, the app pauses. Analyzing your profile…appears on a soft green screen with a puzzle piece. When it finishes, it tells you who you are. You are a Structure Seeker. You’re ready to get with the program. Hungry for innovation and invigoration — you will succeed in a new, promising environment.
The sentence has the precision and emotional register of a horoscope reading, and it made me feel seen even when I knew what it was doing.
The screens that follow are positioning: Noom is based on real psychology, you can still eat your favorite foods, nothing is off limits, you will not be punished for messing up.
Somewhere in there, Noom asks you a trivia question. When is the best time to weigh yourself? You pick first thing in the morning, before you’ve eaten. The next screen says You’re right! with a citation to the International Journal of Behavioral Nutrition and Physical Activity. I felt smart, which was a different kind of flattery from the horoscope.
None of this is necessary to reach the product. It felt vaguely soothing, like a first therapy session.
Duolingo does this too. By screen 8, you’re picking a language. By screen 12, you’re in a tiny first lesson. You get a word right, a small animation plays, the owl shows up. You’re already learning, and the remaining 60-plus screens don’t feel like onboarding anymore.
Headspace uses silence. You spend a screen breathing before you are ever shown a feature. By the time it asks about your sleep, your stress, what brought you here, you have already meditated for 20 seconds.
BitePal does it too. Each screen is small, and each one adds a piece to something that feels, by the end, like it was built for you. The raccoon is part of this. You named the raccoon; it is in your plan; leaving means leaving Caramel in the bin.
It still works on you even when you see what they’re doing. A long onboarding done well is a conversation you want to finish. A short one does the same work in less time: Headspace in 21 screens, Calm in 14.
And then there are the apps that skip the conversation entirely. No questions, no profile, no plan. Just the product, dropped in your lap. Without a conversation, the app is telling you: I did not want to know you.
Onboarding paywall placement
By screen 72, something odd happens. Seven progress bars, stacked, filling in sequence. Based on your answers… Preparing your profile… Calibrating your program… You have been in this app for 8 minutes. You have told it things you are not sure a calorie tracker needed to know, and now it is pretending to think.
Then the paywall hits.
The Structure Seeker is on the page. The program is calibrating. $209 for the year.
Of the 986 apps, 217 put up a paywall during onboarding. Of those, 155 — about 71% — ask for money without first asking you anything about yourself. Only 62 do both: personalize, then paywall. Those 62 run the longest flows on average — 42 screens.
Those 62 apps — about 6% of the total — are doing something different.
It goes roughly like this: ask about the person. Open a conversation where each answer visibly shapes what comes next, and build something from what they tell you. Show them what you built. Then say: here is the plan, and this is what it costs.
The other 71% do this: here is the product. It has features and three tiers. One is on sale. Please enter your card.
If you’re handed a paywall on screen five, you’ve invested nothing. You’re being pitched, and we all know a pitch when we see one.
But if you’ve spent 8 minutes telling an app what you weigh and how you speak to yourself when it isn’t going well, you’re in a different emotional state from when you downloaded it. The paywall, when it comes, is asking if you're prepared to pay for the thing you just spent 8 minutes building.
Paywall-only apps come in at 31 screens. Personalize-only, at 35. The ones that do both, 42.
BitePal belongs to the 6%. Noom, too. Calm. All trial-gated — card up front, free for a week, billed on day eight. What you get in return for ninety screens is a plan with your name on it and an offer that feels tailored to you.
Turns out length doesn’t kill the sale, if the length is doing something.
Not everyone buys it, though. The same pattern, a personalized quiz that never mentioned a price was coming, gets called manipulative. Loudly. (There’s a real comment on Mobbin’s video about this, 28 likes and 5 replies. They really hate it.) If you don’t know there’s a paywall at the end of 70 screens, the whole thing feels like a trap. The pattern works. Just make sure they know what’s coming.
Then there’s Lemonade. Not every app that personalizes before the paywall is building a plan.
Lemonade runs 143 screens. Tap through the first 15 and what you find isn’t a compliance document. It’s a chatbot. Maya asks about your building. You answer. Maya tells you she’s analyzing your data. Connecting to municipal databases. Connecting to environmental databases. A screen labeled Analysis Complete shows up with star ratings: Building Age, four stars; Distance From Coast, four; Fire Hazard Proximity, three. Your answers, dressed up in stars.
What comes next is a price.
The flow does everything the good ones do, except what’s being built is a quote instead of a plan.
My own read: it’s clever and manipulative. It definitely works.
Five app onboarding best practices
Noom’s flow worked for a reason. So did Duolingo’s, and BitePal’s. The same five things keep showing up.
The earliest is the progress bar. It shows up on screen 2 or 3 — not at screen 20, but right away. A progress bar is a promise: there is an end to this, and here is how far in we are.
Then, questions that feel personal: Noom’s binary sliders, Hinge’s prompts, BitePal’s plate-shaped dietary questions. Each asks you something and offers room to answer honestly. That small difference, repeated across a dozen screens, is what creates a sense of connection.
After the questions, a mirror that reflects what you said. The app, having listened, tells you who you are. Structure Seeker. Careful Beginner. Adventurous Saver. Like an MBTI result: imprecise, unfalsifiable, and yet I eat it up. Every. Single. Time.
Somewhere in the middle, a moment where you try the product. Before the paywall, before the plan is revealed, the app lets you have a small taste of the product itself. Duolingo’s first lesson, on screen 12. Headspace’s 20-second breath. BitePal’s raccoon.
Hinge does this at screen 15. The app asks you to write your profile answers — Typical Sunday, My best Dad Joke, You should leave a comment if — and suddenly you are being funny for a stranger. The onboarding is 95 screens long, and by screen 15, you are already creating the most dateable version of yourself.
Then, late in the flow, a plan built from your answers. The app shows you what it assembled. Here is your program. Here is what it will focus on. Here is when we think you will see results.
The apps that get it wrong have the screens, but the screens don’t have a purpose.
How long should onboarding be?
Midjourney runs 3 screens. Noom runs 113. Both are doing it right.
Across 986 apps, the patterns that make a long flow feel like a conversation are the less common ones. Personalization sits at 23%. Paywalls, 22%. Tutorials, 8%. Most apps are still just verifying identity and getting out of the way.
By category, Health and Fitness and Education sit near the top of both lists. They paywall because they run on subscription, and they personalize because the product doesn’t work without knowing who you are. The asking and the selling have become the same flow.
Personalization splits along the same lines.
The conventional wisdom of less friction, more conversions is right for apps where you already know what you came to do. Midjourney at 3 screens isn’t cutting corners.
But the apps running 80, 100, 113 screens aren’t ignoring that advice. They’re running long because the product doesn’t work without what you tell it first.
Caramel
Caramel is still on my phone. I have not opened BitePal in weeks. I do not know whether I lost weight, or whether the app did what it said it would do.
But somewhere in the 61 screens, a designer put a raccoon inside a functional flow. A small decision that suggested someone had cared about something not-entirely-functional. This is the kind of thing that gets flagged in review as unnecessary, or shall we cut for brevity, or scope creep. The small ornaments and strange little touches that someone fought for, and someone else quietly deleted.
Maybe they do not move a metric.
What I know is that I still think about Caramel. Sixty-one screens, and the one I remember is the one that had nothing to do with onboarding.
App onboarding statistics
Data from Mobbin. All flows captured on iOS. Revenue figures: BitePal reported roughly $400K in its first year; Noom subscription was $209/year at time of writing.
apps studied
986
onboarding flows
1,460
screens on average
25
longest (Noom)
113
of paywalling apps don't personalize first
71%
personalize then paywall
6%