Meijer · 2025
AI-Powered Associate Experience
Store teams needed trusted answers and operational tasks completed without leaving the floor.
Interface recreated for portfolio presentation to protect confidential product information.
My role
Senior UX Designer
Impact
- 35% improvement in task efficiency
- 45% faster information retrieval
Scope
- Research
- Experience strategy
- Conversational interaction
- End-to-end product design
- Design systems
- Prototyping
- Cross-functional alignment
Timeline
2025
The challenge
Associates moved between multiple internal tools while a customer waited. The information existed; the path to it did not fit the moment.
The opportunity
One conversational surface, grounded in internal knowledge, that answers in the associate's language and hands off cleanly when it cannot.
What research changed
Speed beats completeness
An answer that arrives after the customer's patience runs out is not an answer. Short, confident responses were preferred over thorough ones.
Trust is earned per answer
Showing where an answer came from mattered more to adoption than making the answer marginally better.
Failure is a design surface
The moment the assistant does not know is the moment that defines the product, so we designed that state first.
How might we
How might we make the fastest path to an answer also feel like the most trustworthy one?
How it changed
V1
Search-centric experience
A better index over existing content.
User signal
People still interpreted results manually
Associates read several results before acting, which put the synthesis work back on them.
V2
Conversational response
A single written answer instead of a result list.
Problem
Unstructured output felt unpredictable
Prose answers gave no consistent place to look, so confidence dropped even when content was correct.
Final
Conversational entry + structured action card
Ask in your own words; receive a typed answer with the source and the next action attached.
Decisions
Decision 01
Conversational entry with structured results.
- Search-first
- Structured workflow
- Conversational entry
Options considered
- What changed my mind
- Research showed associates framed tasks as questions but expected an actionable answer, not reading material.
- Why
- It preserved flexibility of phrasing without making generated output feel unpredictable.
- Tradeoff
- More component logic and state management, in exchange for significantly clearer user expectations.
Decision 02
Express confidence through wording, sourcing and available actions.
- Hide model confidence
- Show a numeric score
- Express confidence in language and layout
Options considered
- What changed my mind
- Numeric confidence invited debate about the number rather than the answer.
- Why
- Associates needed to know how far to trust an answer before acting on it in front of a customer.
- Tradeoff
- Harder to write and to QA than a score, and it required content rules the team had to maintain.
Decision 03
Deliver the assistant as tokens and components in the design system.
- Ship the assistant as a standalone screen
- Ship it as system components
Options considered
- What changed my mind
- Neighbouring teams began asking for the same answer pattern inside their own tools.
- Why
- The answer pattern was more valuable as shared infrastructure than as one destination.
- Tradeoff
- Slower first release, and more upfront alignment across teams.
Constraints we designed around
Operational environment
Layouts assume one hand is holding something else, on a handheld device, under store lighting and noise.
AI confidence
The system cannot be right every time, so every response pattern also had to work as a partial or unknown answer.
Where it landed
Answer-first responses
The answer leads, the reasoning follows, the source stays one tap away.
Actionable results
A response can become an action rather than ending as text.
Graceful escalation
Low confidence routes to a person with the context already attached.
Outcome
35%
Improvement in task efficiency
45%
Faster information retrieval
What I took from it
The hardest part of conversational design is not the conversation — it is deciding what the product does when the conversation breaks.
Next project
Walmart
My Assistant + Intelligent Operations