Swipe right on the food you actually eat

Context

GoodFlip is a metabolic health platform. Programs for diabetes, PCOS, fatty liver and weight, with doctors, coaches, lab tests and devices behind them.

Food is where most of that work is won or lost. People handle the app fine. Dinner is the hard part.

So the same idea kept coming back in planning. GoodFlip should recommend recipes built around each person. Everyone agreed it sounded good. That was the whole problem.

Why plans get abandoned

Ask anyone who has tried to eat better and the story rhymes. A plan arrives. It opens with subtraction. Less rice, less sugar, less ghee.

Then it lists meals full of food they have never cooked. Quinoa on a Tuesday, in a house that eats khichdi. The plan is nutritionally correct and personally irrelevant, so it gets abandoned inside a week.

The person walks away from that believing they lack discipline. The plan never asked them a single question.

Why we tested it first

Building this into GoodFlip means touching the recipe engine, the plan screens, the content pipeline and the nutrition review. Months of work on a hunch.

So the brief I set was deliberately small. Build the simplest honest version, put it somewhere anyone can reach in one tap, and learn from what people do with it.

Keeping it outside the app was the point. Nothing to install, nothing to sign up for, and no risk to the product people already rely on.

Choose what you love, we'll cook up the rest

The whole experience is one gesture. An ingredient appears, you swipe right if you love it and left if you don’t. Five of those and you are done.

A set is never a random five. One ingredient is drawn from each food group, every time, so nobody gets a run of leaves or a run of meat.

• Dairy

  Paneer, curd, ghee

• Fruits

  Banana, guava, mango

• Grains and pulses

  Oats, rajma, ragi

• Non-veg

  Eggs, fish, chicken

• Vegetables

  Spinach, bhindi, brinjal

It’s a match

The wait shows a pan on the heat and says roughly how long it will take. Then two recipes arrive, under a line that says exactly what just happened. It’s a Match. These recipes contain the ingredients your heart said yes to.

Each recipe carries a match score, so you can see why it is on your list. A hundred percent means every ingredient came from the cards you loved. Calories and macros sit beside it, so the choice stays informed.

Open one and it reads like something a person would cook. Zucchini Egg Bhurji Twist, serves one, with the tadka as its own step. Save it, share it, or download it as a PDF for the kitchen.

Where the recipes come from

The recipes are generated, which is the part most likely to go wrong. Left loose, a generator will happily return something that sounds healthy and falls apart in a kitchen. Most of the work here went into narrowing what it is allowed to hand back.

Both halves of the swipe are sent, and the passes matter as much as the loves. A recipe built around something you rejected is worse than no recipe at all.

We ask for exactly two recipes, in a fixed shape, every time:

That last one turned out to be the useful constraint. Asking for a plain explanation of why a dish is healthy pulls the whole recipe towards food that can survive the question, and it gives the reader the reasoning rather than a claim.

Anything that comes back in the wrong shape is thrown away instead of shown. If the service is busy the app retries quietly, and only after three tries does it say plainly that it is overloaded.

The decisions

1. 1

   Borrow a gesture nobody has to learn. Everyone already knows what a swipe means, so the screen needs no instructions and no onboarding. Recognition over recall, doing a job a tutorial screen would have done badly.

2. 2

   Ask about ingredients. Rating a meal makes someone picture it first. Paneer or bananas land on an opinion they have held since childhood. Easier question, more honest answer.

3. 3

   Leave a door open at every dead end. Pass on all five and a fresh set appears. Love fewer than two and the recipes are held back rather than promised on thin evidence. Any swipe can be undone, so a wrong one costs nothing.

4. 4

   Earn the phone number. Nothing is asked at the door. You swipe, you get recipes, you open one. Only on a second round does it ask who you are, by which point it has given something first.

What we're watching

Every step is instrumented: the open, the first swipe, a set completed, a recipe opened, a recipe saved, a second round started.

People finishing a set and coming back for another round would mean personalisation has earned its place in the app. People leaving on the first card would mean months of engineering saved. The numbers are staying in house for now.

What I learned

1. 1

   A question about people is cheaper to answer than to argue. The experiment took weeks. Building it into the app on a hunch would have taken months, and we would still be guessing at the end of it.

2. 2

   Being asked is most of what personalisation feels like. The recipes matter. The moment that changes how someone sits up comes earlier, on the first card, when the product asks what they love and waits for an answer.

3. 3

   Make the research enjoyable and people finish it. The same preferences could have come from a form. Forms get abandoned halfway, and half-finished answers make for weak evidence.

On the second one, the lever has a name. Commitment and consistency is the same principle behind the daily promise in Metabolic Kickstarter, which is also where this experience hands people next.

Credits

A small experiment, and never a solo one.

• VB

  Vishal Bansal

  Design Manager

  Backed running this as an experiment, and steered it from the first sketch.

• KP

  Khyati Patil

  Brand Communications Manager

  Gave it a voice that sounds like a person, down to Swad be, Healthy be.