Case Study · Food & Nutrition · AI

MealMap

Using AI to take the guesswork out of grocery shopping and meal planning — learning your preferences, dietary needs, and budget to plan, shop, and navigate for you.

Role UX Researcher & Designer (team of 3)

What I did Designed the 9-participant interview protocol and ran 4 sessions · synthesized thematic analysis · created journey map, task analysis, and competitive matrix · wireframed and prototyped all 6 core flows · designed and ran 4 of the 11 think-aloud usability sessions with mixed quant scoring (SEQ/SUS/NPS/DRT) · iterated based on findings

Duration 2.5 months · Context: HCDE coursework, U-Michigan

Read the study Skip to final design ↓

🧠

AI-Powered

Learns your preferences

Research

User interviews

TL;DR

The grocery-app market is fragmented across 3–4 disconnected tools. I interviewed 9 users (ages 24–64), identified 5 friction themes, and designed a single AI-powered concept covering meal planning through in-store navigation. After one iteration cycle, both core tasks reached acceptable SEQ thresholds, SUS improved, and sentiment shifted meaningfully positive. The remaining hard problem: in-store map confirmation interactions — still being iterated.

+13.7

SUS points

+0.7

SEQ Meal Planning

+1.9

In-Store Navigation

Think Aloud Usability Sessions

The Problem

No single app does all of it.

The grocery and meal-planning app landscape is deeply fragmented: one app to plan meals, another to navigate the store, a third to manage the pantry, a fourth to look up recipes. None of them talk to each other, and none of them adapt to your personal needs.

MealMap was designed to change that — a single, AI-powered experience that covers the entire journey from "what's for dinner tonight" to finding it on the shelf, while staying accessible to users with diverse needs.

No grocery shopping app creates meals, remembers when your favorites are running out, manages your pantry, allows you to import recipes from anywhere, helps you navigate to your items in the store, and adapts to user accessibility needs — all in one app.

Design Question

How might we help adults plan meals and shop more easily by using narrow AI to auto-build budget-aware lists, find the best store(s) and prices, plan multi-stop routes, and guide in-store item finding — while staying accessible for diverse needs?

01 — Discovery

What we heard when we actually listened.

9 individuals — ages 24 to 64 — were interviewed on their grocery shopping habits, technology usage, pain points, and accessibility needs. Two participant quotes crystallized the core problem space immediately.

User Interviews

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

User Journey Map

The journey map surfaced recurring friction points: losing a written grocery list, recipes scattered across multiple locations, and struggling to locate items once inside the store.

02 — Define

Turning interviews into clear direction.

Thematic analysis of the 9 user interviews produced 5 clear themes. Competitive analysis confirmed no existing app addressed all of them — revealing a genuine market gap.

Thematic Analysis — 5 Themes

🍽️ Planning & Meal Decisions 🗺️ Store Navigation 📱 Digital Tool Adoption 😤 Pain Points & Barriers 💰 Substitutions & Budgeting

Core Gaps & Problems Identified

01Meal planning and decision fatigue — the mental load of "what's for dinner" pushes users to costly meal-kit shortcuts
02No "all-in-one" app exists — users manage meals, groceries, pantry, and cooking across 3–4 disconnected tools
03Trouble locating items in stores — especially after resets or in unfamiliar locations

Competitive Analysis

No existing app addresses all parts of the grocery shopping experience — confirming a clear design opportunity.

Task Analysis

Task analysis mapped the full flow from meal planning to cooking — revealing where handoffs between tools break down and create friction.

03 — Ideate

From blank paper to tested wireframes.

Ideation began with hand sketches across all major flows, then rapidly progressed to low-fidelity wireframes for usability testing — before any visual design was applied.

Hand Sketches — Early Ideation

Home & Dashboard

Recipes & Meals

Grocery List

In-Store Navigation

B&W Wireframes

Home Dashboard

AI Meal Suggestion

Grocery List

In-Store Navigation

Design Iterations

What changed. And why.

Testing showed which flows needed another pass before high-fidelity design. These two small but powerful changes had the biggest impact on the final product.

Confirming items in cart

From auto-confirmation to digital scratch-off

Before

Problem identified

Users were confused by the auto-confirmation that they had added an item to their basket once they tapped on the live map for new directions.

After

Iteration to include check boxes to match user mental models

Design decision

By adding a simple checkbox, users were able to regain a sense of control and internal confirmation: "I got my item." It mirrored the familiar action of scratching an item off a handwritten list.

Result

Users gained confidence in their purchasing ability.

Live Maps

From manual list-building to a smarter handoff

Before

Problem identified

While users liked the 2D maps, several participants reported grocery shopping can be overwhelming, especially when the store is busy, so a map that didn't match their mental model (i.e. high shelves, narrow aisles) added to cognitive load.

After

Design decision

I added a 3D map option to mimic the real feel of the store and match user expectations.

Result

3D map matched user mental models and reduced reported overwhelm. In-store confirmation interactions remained the hardest unsolved problem — see Reflection.

What iteration taught us

Mental models matter

If you design for all ages and all abilities, you will develop a more well-rounded and well-received product sooner.

Specs can deceive

Don't trust the spec; always talk to users. The 'all-in-one app' problem I designed for was three separate problems hiding inside one design brief. Only the interviews surfaced that.

The little things make a big difference

I didn't expect a simple button here or there to create real change in users' minds, but those little details created real results.

Final Design

Six screens. One complete experience.

The final solution unifies onboarding, recipes, food inventory, grocery lists, in-store navigation, and error handling into a single AI-powered app — each screen built around accessibility and reducing friction.

Onboarding

3 sections: Dietary Needs, Accessibility Needs, and Personalization Preferences
Ensures AI tailoring is user-need-first from day one
Multiple filtering inputs per section

Recipes

AI-generated meal plans and single meals
Upload from outside sources, write your own, or browse recent & favorited

Food Inventory

Quick view of 'Expiring Soon' and 'Low Stock' items
Categorized by pantry, refrigerator, and freezer
Helps users manage food and prevent waste

Grocery Lists

AI-generated, from previous lists, or write your own
AI learns behavior: "you already have milk," "you normally buy cinnamon with oatmeal"
Catches missed items, prevents buying extras

In-Store Navigation

Optimized route using phone GPS and store layouts
Guides users to items in the most efficient path
Substitution suggestions when items are out of stock

Error Handling

Graceful fallbacks when connectivity or AI fails mid-shop
Options: reconnect, shop offline, check out later, or get human assistance
Users are never left stranded — AI and technology are not infallible

Video Walkthroughs

Walkthroughs narrated with my design decisions.

An in-depth explanation of why I made certain design choices.

Onboarding

Recipes

Food Inventory

Grocery Lists

In-Store Navigation

Error Handling

04 — Validate

Testing showed us where to push harder.

A mixed-methods testing approach combined qualitative depth with quantitative scoring — giving a complete picture of what was working and what still needed iteration.

Qualitative

Think-Aloud Testing

Recruited 11 new participants for think-aloud testing (ages 28-63)
VADER sentiment analysis across all sessions
1 open-ended trust question focused on AI reliability

Quantitative

Scoring Frameworks

SEQ — Single Ease Question per task
NPS — Net Promoter Score
SUS — System Usability Scale
DRT — Desirability, Reliability & Trust of the AI: I designed a custom 4-dimension Likert scale to measure Desirability, Reliability, and Trust in the AI features

Sentiment Analysis

Sentiment distribution chart showing shift to positive after iteration

Participants had an overall more positive attitude toward MealMap after the first round of iteration — sentiment shifted meaningfully toward positive and neutral responses.

Quantitative Results

Quantitative scoring results post-iteration

Post-iteration, both main tasks (planning a meal and in-store navigation) reached acceptable SEQ scores. SUS, NPS, and DRT all improved.

Score Improvements After Iteration

SEQ
Meal Planning: 5.5 → 6.2 (already at threshold) & In-Store Nav: 3.5 → 5.4 (reached acceptability after iteration)

Single Ease Question

Both tasks at acceptable threshold

SUS
(65.8 → 79.5)

System Usability Scale

Improved overall ease of use

NPS
(-50 → +40)

Net Promoter Score

More promoters; sample too small for statistical claim.

DRT
(3-4 → 4-5)

AI Trust Score

Improved AI desirability & trust

Lessons Learned & Reflection

What worked. What I'm still solving.

🏆 Wins

AI-generated meals directly addressed the decision-fatigue problem that drove Participant 7's meal-kit dependency. This was the single clearest design-to-research-finding link in the project.
Reminders to add frequently used items to the grocery list were a standout favorite feature — users described them as helpful rather than intrusive.

🔬 What I'd run next

In-store map confirmation still has too many steps. Moving confirmation off the map to the list reduced friction but didn't solve it — my next test would compare three confirmation patterns (list-only, gesture-on-map, audio-confirm) with a within-subjects design.
Several participants completed tasks without exploring the broader app — meaning we missed feedback on discoverability features. Next round I'd add explicit exploration prompts after task completion, before the post-test interview.

Next case study

Autonomous Rideshare