UX case study · Food & nutrition · Mobile AI
MealMap
From "what's for dinner?" to finding it on the shelf.
An AI grocery assistant connecting meal planning, pantry inventory and in-store navigation in one flow.
- Problem
- Meal planning, groceries, pantry tracking and recipes span 3–4 disconnected tools.
- My role
- UX researcher & designer in a team of three. Research, synthesis, six core flows and usability testing.
- Timeline
- 2.5 months
HCDE coursework, U-Michigan - Outcome so far
- A tested six-flow prototype. SUS rose from 65.8 to 79.5 after one iteration; in-store confirmation still needs work.
01 · Context & stakes
No single app does all of it.
One app to plan meals, another to navigate the store, a third for the pantry, a fourth for recipes — none of them talk to each other, and none adapt to accessibility needs. The fragmentation itself is the user's problem.
"How might we help adults plan meals and shop more easily using narrow AI — budget-aware lists, best-store routing, in-store item finding — while staying accessible for diverse needs?"
Design question02 · Role & boundaries
What I owned
I drove
- Designed the nine-participant interview protocol and ran four interviews
- Synthesized thematic analysis, journey mapping, task analysis and the competitive matrix
- Wireframed and prototyped all six core flows
- Designed and ran four of eleven think-aloud sessions with SEQ, SUS, NPS and DRT scoring
- Iterated on the testing findings
Collaborators & participants
This was a three-person HCDE coursework project at U-Michigan. The team conducted nine discovery interviews and eleven think-aloud sessions in total.
Interview participants were ages 24–64; the new participants in usability testing were ages 28–63.
Constraints
- A 2.5-month coursework timeline
- Six connected flows spanning planning, shopping and error handling
- Diverse dietary, accessibility and technology needs
- Prototype testing with a small sample, limiting statistical claims
03 · Research
What we heard when we actually listened.
9 interviews, ages 24-64, on shopping habits, technology use, pain points, and accessibility needs. Two quotes crystallized the problem space immediately.
"We just stopped [meal kit service] because we don't like making those decisions — what to make for dinner. So, even though [meal kit service] is probably more expensive, it's so easy just to not think about what's for dinner tonight."
Participant 7 - decision fatigue around meal planning"When they do a reset, I can't find anything... I have to look up at the aisles and see where things are listed."
Participant 2 - in-store disorientation after store resets
The journey map surfaced three recurring frictions: losing the written list, recipes scattered across tools, and locating items once inside the store.
Turning interviews into clear direction
Decision fatigue
The mental load of "what's for dinner" pushes users to costly meal-kit shortcuts.
No all-in-one app exists
Users juggle 3-4 disconnected tools for meals, groceries, pantry, and cooking.
Items are hard to find
In-store disorientation, especially after resets or in unfamiliar stores.
No competitor covers the full experience — a confirmed market gap, not an assumed one.
Task analysis showed where the tool handoffs break down — each handoff is a friction point.
04 · The solution
Six flows. One connected experience.
Onboarding captures dietary and accessibility needs. Meal suggestions connect to pantry inventory and grocery lists; in-store navigation carries the list into the shopping trip, with error handling for when the AI or connection fails.
Onboarding — dietary, accessibility & personalization needs first
Recipes — AI meal plans, imports, favorites
Food inventory — expiring soon, low stock, less waste
Grocery lists — "you already have milk"
In-store navigation — optimized route, substitutions
Error handling — AI fails gracefully; users are never stranded
Video walkthroughs, narrated with my design decisions
Onboarding
Grocery lists
In-store navigation
Also available: Recipes · Food inventory · Error handling
05 · Key decisions & trade-offs
What changed. And why.
Testing showed which flows needed another pass before high fidelity. Two navigation decisions show how the prototype changed in response.
Confirming items: from auto-confirmation to digital scratch-off
Tried → Learned → Revised
Result: users regained confidence in their purchasing ability — the mental model won.
Trade-off: Explicit checkoff gives shoppers control but adds an action to each item. The remaining confirmation steps need further testing.
Live maps: from 2D-only to a 3D option
Tried → Learned → Revised
Result: the 3D map matched mental models and reduced reported overwhelm. In-store confirmation remained the hardest unsolved problem — see Reflection.
Trade-off: Keeping 2D and adding a 3D option supports different map preferences, but changing the view alone did not resolve the item-confirmation flow.
"Don't trust the spec; always talk to users. The 'all-in-one app' problem was three separate problems hiding inside one design brief — only the interviews surfaced that."
What iteration taught usSee the sketches and wireframes behind the six flows
Hand sketches across all major flows, then straight to low-fidelity wireframes for usability testing — before any visual design was applied.
Hand sketches
Black & white wireframes
06 · Outcome & evidence
Better scores, with one clear limitation.
Twenty participants contributed across nine interviews and eleven think-alouds. Six core flows were wireframed, prototyped and tested; the scores below compare the two core tasks before and after one iteration cycle.
New participants, ages 28-63
Single Ease Question on both core tasks
System Usability Scale across sessions
My 4-dimension Likert scale for AI desirability, reliability & trust
Across all sessions + open AI-trust question
Sentiment shifted meaningfully toward positive after the first iteration round.
Post-iteration, both core tasks cleared acceptable SEQ scores — SUS, NPS, and DRT all improved.
Both core tasks reached acceptable SEQ thresholds, and sentiment shifted toward positive. These are results from a small prototype study; the NPS change is descriptive, with no statistical claim. In-store confirmation remains an unresolved usability problem.
07 · Reflection
Can shopping feel easier if confirming each item still takes too many steps?
The map improved, but in-store confirmation still has too many steps. Matching the store visually was useful; the next study needs to test the action of collecting an item.
Wins
AI-generated meals directly answered the decision fatigue behind Participant 7's meal-kit dependency — the clearest design-to-research link in the project. Frequently-bought-item reminders were the standout favorite: "helpful, not intrusive."
What I'd run next
In-store confirmation still has too many steps. Next test: three confirmation patterns (list-only, gesture-on-map, audio-confirm) in a within-subjects design — plus explicit post-task exploration prompts, since several participants never saw the broader app.