Case study · 04
Removing the drop-off wall from an intake
A 33-question mobile intake restructured into an adaptive, card-based flow that feeds a personalized meal-plan engine.
Context
The client, a health-tech company, turns metabolic measurements into personalized nutrition plans. Everything a breath test can’t measure — diet type, meal times, fasting, allergies, food likes and dislikes, coffee, alcohol, workout fuelling, supplements — comes from a questionnaire in the nutrition tab of its mobile app. Every answer maps to a field the meal-plan engine needs, so if people abandon it or answer inaccurately, the plan is wrong or never generated.
The problem
The first version collected the right data but fought the user the whole way. It showed 33 questions with a raw counter, took an estimated 10–14 minutes, and asked them at the moment motivation is highest — then spent that motivation badly. The middle of the flow, a run of near-identical food screens, was identified as the main drop-off point.
- A wall of 5–8 near-identical food-category screens in a row — the flow’s biggest drop-off point.
- Clinical jargon (“FODMAP”, diet labels people misread) that led to random answers and silently corrupted the plan.
- Rigid inputs that ignored what the flow already knew — a vegan still walked through animal-protein screens, black-coffee drinkers scanned a full list of additions.
Approach
Start from where people failed
We organised the redesign around named failure points, not screens: the food-screen wall, the terms people didn’t understand, the questions that didn’t fit real routines. Each change in the new flow traces back to one of them.
Restructure into four named sections
Thirty-three numbered questions became four sections — eating pattern, food preferences, eating habits, workout fuels — with section-based progress instead of “Q14 of 33”. The flow opens with easy, identity-affirming questions to build momentum before the harder ones.
Branch, merge and filter
Conditional logic gives every user the shortest valid path: non-drinkers never see alcohol questions, non-exercisers skip workout fuelling, fasting windows appear only for people who fast. Double questions were merged, and options are filtered to what the user actually eats, so the path ranges from about 18 to 25 questions.
Write it down for the experts
Alongside ~50 screens, we wrote a 15-page spec for the client’s nutrition team: exact wording, options, logic and the reason behind every question, plus what changed, what was removed and a verification checklist. It let the experts confirm clinical completeness line by line while design owned the experience.
Key decisions
Neutral by default
The 5–8 food screens collapsed into one screen of category cards. Every item starts neutral; people only mark strong feelings — prefer or exclude — and can open a food to set individual varieties. It trades some granularity for completion: the engine can plan around neutral answers, not around people who quit.
Collapse, don’t hide
Diet type personalises the rest of the flow, but irrelevant categories collapse instead of disappearing, so a vegetarian who sometimes eats fish can still make an exception. The flow respects what people tell it without locking them in.
Plain language, and friction only where it pays
Clinical terms became everyday descriptions (“hard-to-digest fruits” instead of FODMAP), every diet option got a one-line explanation, and the most common coffee answer became a single tap. The one place we added friction on purpose: time pickers have no pre-filled defaults, so people don’t accept a guess as their real routine.
Outcome
The new intake is an adaptive, four-section flow that sets a time expectation up front, rewards progress with short insights between sections and hands off visibly to the meal plan being generated. The timing and question counts are design-stage estimates from the spec, not post-launch measurements. What is fixed by design: every backend field was kept, so the nutrition team lost no data and the core API needed no changes.
Client names and screens are kept confidential; illustrations are abstract recreations.
What we learned
In a long form, perceived length matters as much as real length — named sections and a shorter adaptive path change how the same questions feel. And in health products, plain language and clinical precision aren’t a trade-off; with the experts reviewing every line, we could have both.