Same program, rebuilt around user behaviour. Retention went from 3% to 23%

Context

GoodFlip's metabolic score, the number the whole app is built around, had become a one-time thing: people took it, got their number, and left. Metabolic Kickstarter was our reason for them to come back. Four weeks, seven days each, a short podcast and a piece of content every day. You take your score, you land in the program, you begin.

The bet we made was that the content was good enough to bring people back on its own. Delightful podcasts, well-written days, a real four-week arc. Weeks were locked, but days inside a week were open.

So people opened day 1, binged all seven days in one sitting, and never came back.

What v1 looked like

Before the numbers, see the flow itself: unlock Week 1, tap View Plan, and explore the week, all seven days open from the first minute.

Welcome to your

Unlock gems. Build habits. Earn rewards.

Your Kickstarter
Journey

1. 

   Week 1 – Foundation

   Recharge your body and restore your energy

   Unlock Now
2. 

   Week 2 – Energy

   Ignite your metabolism & burn fat effectively

   Unlock Now
3. 

   Week 3 – Balance

   Boost nutrients, sharpen focus

   Unlock Plan
4. 

   Week 4 – Strength

   Build habits that last

   Unlock Plan

Progress that pays off

Unlock exciting rewards each week of your Kickstarter journey

The 3%

Day 1 to day 7 retention was 3%. That's not a dip, that's a verdict. Of every 100 people who started day 1, 97 were gone before day 7.

The number stung, but we had shipped it, so we had the whole dataset. We knew exactly where people went.

Why it failed

Back at the whiteboard, we ran the teardown as ask 5x why: keep asking why people left until the comfortable answer, content quality, stopped surviving the question. It didn't survive because we tested it: solution interviews with a varied set of users, extreme users on both ends, people who binged the whole week and people who left on day 1, all asked one question: is this content worth coming back for? Most liked it as it was, a few had minor feedback we folded back in, and nobody wanted it rebuilt. The content was fine.

What was left was the real mistake, sitting in plain sight: a bet on behaviour, made without a model of behaviour. Hold v1's choices up against basic behavioural psychology and each one names its own failure:

• We designed for consumption. Retention needs participation.

  Content you can finish in one sitting has no reason to be daily. We had built a library and called it a program.

  the psychology
  People return to things they have acted on, not things they have read: commitment and consistency. v1 never asked the user to do anything.

• Nothing was left open.

  A user finished day 1 with everything they came for. No loose thread, so no reason for tomorrow to exist.

  the psychology
  The Zeigarnik effect: unfinished things stay on the mind, finished things get released. v1 ended every session finished.

• The reward was the content itself.

  And the content was free, immediate, and all available at once. The app taught people to binge and leave. It worked exactly as designed, which was the problem.

  the psychology
  The goal-gradient effect needs a visible goal you're closing in on. With everything free and open at once, there was nothing to approach.

So we went and asked actual coaches

We wanted Kickstarter to work like a coaching program, with Kaira, our in-app AI, as the coach. So instead of theorising about what a coach does, we ran explorative interviews with five working coaches: fitness, nutrition, and habit coaches, all running paid programs with real clients.

Their programs run on a loop so simple it's almost boring. In the morning, the coach tells you what to do today, and you commit to it. Through the day, you do it. At night, the coach asks whether you did.

That's the whole retention engine, and it has nothing to do with content quality. The client comes back at night because someone is going to ask. They come back tomorrow because tomorrow has an instruction. We weren't inventing a mechanism, we were benchmarking one from an adjacent field, one that already works, and that people already know how to follow.

The reframe, and the three moments

So the design question became a how might we: how might we make Kaira behave like a coach, so a user has a reason to come back tonight and again tomorrow?

That turned into three things we had to design, and only three:

1. 1

   Move the user from passive consumer to active participant

2. 2

   Create closure for today

3. 3

   Create anticipation for tomorrow

Each goal became one moment in the day, wrapped around the v1 content you already saw, which v2 ships word for word. Every screen below is the real product, running. Not a screenshot, not a redraw, the actual shipped Day 1. Play it.

The three moments, live

At the end of every morning's content, Kaira stops teaching and asks you to decide: how much protein will you add today? You pick an option and a written promise fills in with your choice. Try it below, pick an option and watch the promise build.

This is commitment and consistency. And the format matters: a goal paired with a concrete plan, an implementation intention, gets completed far more often than the wish alone.

At night Kaira checks in with a short sheet: did you keep the morning's promise? Answer honestly, and she closes the day and hands you a reward. Tap through it below, Yes, I've tried it, and follow it down.

Kaira AI

Hey 👋
Before we begin, I just want to check in with you.

 

Reflection works best when it's done after you've tried today's step, even in a small way. This is just about being honest with yourself.

✅ Yes, I've tried it

⏳ Not yet

The point is closure, a clear small win made visible. And the reward you just collected is a puzzle piece. It looks like a variable reward but deliberately isn't: the piece is guaranteed, the set is what's incomplete. Finish the days and watch it assemble, six in, the seventh pulls hardest. That's the Zeigarnik effect meeting the goal-gradient effect.

Days are locked now, so you can't binge the week the way v1 let you. But reach for tomorrow early and you don't hit a closed door: you get a live countdown to midnight and a glimpse of what's coming, priming the curiosity that brings you back.

Day 2 opens tomorrow

Tomorrow you will learn something unexpected

I have a simple hack for tomorrow that boosts metabolism by 30% in just one hour

Unlocks in

00hrs

00min

00sec

You have done enough for today. Let your body do its part. We'll continue tomorrow.

We'll send an app notification as soon as Day 2 opens up

Go to previous day

What we deliberately didn't do

Once you start reading behavioural psychology you find plenty that works on people rather than for them. In a health app, the levers you refuse to pull matter as much as the ones you ship, so we wrote the refusals down as design principles before v2 was designed. Every decision after had a line it could not cross.

• won't

  No variable rewards. We could have randomised the pieces, made some rare, added a slot-machine pull. It might have worked. Manufacturing compulsion in a health app is a line I didn't want to design past. A guaranteed reward with an incomplete set gets the return without the casino.

• won't

  No streak punishment. Miss a day and you keep your pieces. Loss aversion is the strongest lever in the box and the cruellest. People already quit health programs out of shame, and a broken streak is a shame machine.

• won't

  No new content. The temptation after a 3% is to blame the content and make more of it. The content was never the problem. We rebuilt the loop and left the library alone.

Impact

still active, by week's endw123%w217%w314%w412%

The week 1 curve is the headline, but the rest of that week's cohort data is where you can see the three moments actually doing their jobs:

• 2,140

  joined the Kickstarter

  the v2 cohort, across all four weeks

• 366

  kept a perfect streak

  checked in all 7 days of week 1, not one miss

• 261

  shared their puzzle

  over half of the 492 who finished week 1

The streaks are the part I watch. A perfect week means the user came back morning and night, seven days running, which is the coaching loop working, not the content:

What I learned

1. 1

   A good bet still needs a model. v1 wasn't lazy work. It was a well-crafted program built on an assumption about behaviour that nobody checked: that good content is a reason to come back. Delight is not a retention strategy. Name the model you're betting on before you ship, so you know what you got wrong when it fails.

2. 2

   The mechanism usually already exists outside the product. We didn't invent the commit-and-check-in loop, coaches have run it for decades. An hour with people who do the job for a living beat any amount of whiteboarding, and it's the second time on this team that the answer was sitting with a human expert rather than in a workshop.

3. 3

   Shipping the failure is what bought us the fix. v1 was in production with full data, so the teardown was based on what people actually did, not what we assumed went wrong. A prototype that never ships teaches you nothing. A 3% in production teaches you exactly where the hole is.

4. 4

   Know which levers you won't pull. Once you start reading behavioural psychology you find plenty that works on people rather than for them. In a health app, the mechanics you refuse to use matter as much as the ones you ship.

Credits

Two versions, one team. The 3% and the 23% belong to the same people.

• VB

  Vishal Bansal

  Design Manager

  Guidance the whole way through, from the v1 bet to the v2 reframe.

• DS

  Daksh Sharma

  Senior Product Designer

  Beautifully crafted v1 and its content.

• KP

  Khyati Patil

  Brand Communications Manager

  Kept the program's voice human across four weeks of daily content.

• AC

  Ankur Chatter

  Tech Lead

  Reviewed both builds and got each version safely into production.

• SV

  Shraddha Vasudeva

  SDE-II

  Paired on the front end, reviews and fixes across both versions.