AI: Action Center
It started as a request to redesign a small task panel. It ended with a new AI-powered platform adopted across Walmart's enterprise products and renamed by leadership.
AI · Enterprise · 0→1
Systems Design · Cross-functional
Role
Senior UX Designer (discovery, strategy, framework, execution)
Timeline
2025-2026
team
Merchants, Suppliers, Design peers, PMs, Engineers, Stakeholders, Me
platform
Web (Merch One + Supplier One)
The Real Problem
The request was small. A PM asked me to redesign the "My Workspace" panel in Merch One — Walmart's platform for merchants managing billions of dollars in supplier relationships. It was a little widget with two tabs: Opportunities and Tasks. It showed some data points and sent you elsewhere to actually do anything.
The ask was: make it better.
What I found when I started looking more carefully was that "better" wasn't the right question.
The widget existed. Merchants weren't using it. Not because it was poorly designed — though it was — but because nobody had ever asked them whether they needed it in the first place. Alongside this widget, three other features were being built in parallel by different teams: a notification system, an AI agent called Wally/Marty, and a messaging feature for merchant-supplier communication. Four different teams. Four different solutions. All solving overlapping problems in completely separate places.
Nobody had looked at them together.
Finding the Fix
Before touching a single design, I went to merchants directly. I showed them the existing widget and my early redesign concepts. The response was immediate and honest:
'We never use this. It doesn't tell us what to do or why.'
'I have my own task list. I manage it offline.'
'Notifications are fine but it's mostly noise.'
'We connect with suppliers and stakeholders ourselves. We don't need a system to track tasks for us.'
That feedback reframed everything. A better-looking widget with the same underlying logic wasn't going to change behavior. The real problem wasn't UI — it was that nobody had ever defined what role a task system should play in how merchants actually work.
Around the same time, I started talking to Megan, a design peer who was working on a messaging feature connecting merchants and suppliers. Her problem looked different from mine on the surface. But the more we talked, the more we realized we were both circling the same thing from different angles. Tasks, opportunities, notifications, and messages — four features, four teams, four roadmaps, and zero shared definition of what any of them meant or how they related to each other.
We decided the only way forward was to get everyone in the same room.
What Actually Happened
The Communication Ecosystem Sync workshop did something nobody expected. It didn't just align definitions — it revealed a priority.
Once everyone had agreed on what a task, notification, opportunity, and message actually meant in the context of how merchants and suppliers work, a pattern became obvious. Tasks and opportunities were the closest to a real, active, unsolved problem. Notifications were noisy but functional. Messaging was a separate thread. But tasks — the thing merchants were managing in offline spreadsheets because nothing in the platform helped them — that was where the biggest gap lived.
That's when Yutsen's use case arrived at exactly the right moment.
Yutsen, a peer working on store operations usecase in my team, brought a problem she'd been trying to solve: Walmart stores generate around 3,400 Not On File (NOF) tickets every month. A product gets scanned at a store register and the system can't find a matching item in the catalog. Merchants were working through these one at a time, averaging 173 hours per resolution cycle, jumping between systems with minimal AI support and no structured way to triage what mattered most.
It was the strongest real-world use case I'd seen. And it mapped almost perfectly onto what the workshop had clarified: a task with clear triggers, clear ownership, clear AI potential, and a real need to communicate with a supplier mid-workflow.
Me and my PM made the call: start with tasks, prove the framework with NOF, then stitch everything else in.
That stitching was the insight that made the Action Center more than a task manager. A merchant working on a NOF ticket shouldn't have to leave the canvas to message the supplier about it. The message should be right there — triggered by the action, in context, without navigation. A task could arrive as a push notification, be prioritized as an opportunity, and resolve through a conversation — all inside the same experience. Everything that had felt like four separate features started behaving like one connected system.
The design question shifted from "what does a task look like?" to "how does a merchant move from awareness to decision to resolution — without ever losing their place?"
I started wireframing the full-page Action Center around that question. The NOF ticket was the first use case to run through it. And because it was active, real, and had Yutsen's team invested in its success, every design decision had immediate feedback attached to it. That's the best possible condition for validating a new framework.
Several things evolved through iteration:
The name changed. "Tasks" felt like homework. Merchants responded better to "Actions" — something they were choosing to do, not something being assigned to them. The whole product became the Action Center.
The AI summary changed. My first version showed recommendations without context. Merchants didn't trust them. I added an AI summary with a confidence score derived from the ML model — explaining what signal triggered the recommendation and why it was prioritized. Merchants started engaging. They could decide to accept, modify, or override — human always in the loop.
The layout changed. An early version tried to show everything at once. I added filters by domain (a merchant managing Dry Grocery doesn't need to see Electronics tasks), due date, impact, and MFE. The full page went from overwhelming to scannable.
The homepage entry point changed. A compact Action Center widget on the Merch One homepage gives merchants a snapshot — highest priority, highest impact — and takes them to the full page when they're ready to work. The widget became useful once the full page gave it somewhere meaningful to go.
What Changed
The NOF use case validated the framework. Yutsen's team built the NOF ticket experience directly on the Action Center platform — and it worked. The target was to reduce ticket resolution from 173 hours to 48 hours, with 60% of tickets auto-resolved by the AI agent and 50% of Wally recommendations accepted by merchants.
After NOF, Alena — a catalog designer on my team — integrated a second use case: AI content quality updates. Merchants could now review and approve AI-generated content improvements for multiple items at once, directly in the Action Center, without navigating away. One more workflow removed from the noise. One more use case proving the framework held.
Then it spread.
The Action Center framework was adopted across Walmart's enterprise products — Seller, Data, Ads — as teams recognized it solved a problem they all had: surfacing the right action, to the right person, at the right time, with enough AI context to act confidently. Leadership renamed it "Next Best Action."
The widget that merchants never used became the AI platform that runs across the enterprise.
What I Had to Work With
Four overlapping features with four separate teams and zero shared definition. Getting tasks, opportunities, notifications, and messages defined consistently — across design, product, and engineering — required a structured workshop, a shared use case, and a lot of patience. The alternative was four teams building four products that users would never be able to make sense of.
No strong use case at the start. The widget redesign brief didn't come with a compelling reason to build something bigger. The NOF ticket problem arrived mid-project and became the foundation. Sometimes the right use case finds you — but you have to be far enough along in your thinking to recognize it when it does.
Building trust in AI with skeptical users. Merchants are experienced, high-stakes decision makers. Showing them a recommendation without explaining how the system arrived at it was never going to work. Every design decision around the AI layer — the confidence score, the AI summary, the ability to override — was about earning trust incrementally, not demanding it upfront.
Visibility from every direction. Once the Action Center concept started gaining traction, everyone had opinions. Design leads, PMs, product directors, and teams across the organization were watching what was being built. That's a good problem to have — but it meant every decision was scrutinized and every iteration mattered.
What I'd Do Differently
I'd run the Communication Ecosystem Sync workshop earlier — ideally before any design work started. We ran it after I'd already been designing for a while, and some of the early work had to be reconsidered once we got alignment on definitions. The workshop was the most important thing we did on the project. It should have been the first thing.
I'd also push for supplier-side validation sooner. The Action Center was designed primarily around merchant workflows, but suppliers interact with the outputs of many of those same actions — content quality requests, cost change approvals, item setup tasks. Getting supplier perspective earlier would have made the framework more complete from the start.
What I Learned
The most important design work is often reframing the problem. I was handed a widget redesign. The real problem was that four teams were building four disconnected systems for the same users. Recognizing that — and making the case for stepping back before moving forward — is the decision that made everything else possible.
AI needs to earn trust incrementally. The confidence score and AI summary weren't nice-to-haves. Without them, merchants ignored the recommendations. With them, merchants started engaging, learning the system's logic, and building the kind of familiarity that leads to long-term adoption. Trust in AI doesn't come from capability — it comes from transparency.
A framework that holds across use cases is more valuable than any single feature. The NOF ticket was the first use case. Content quality was the second. Supplier proposals, catalog workflows, and more followed. Each one validated the platform because the underlying logic was sound. Solving the first problem correctly — defining what an action is, how it should be surfaced, and how a human and AI should share the decision — made every subsequent use case faster and cleaner.
Naming matters more than you think. "Tasks" felt like a to-do list. "Actions" felt like a decision. "Next Best Action" felt like intelligence. The same underlying capability read completely differently depending on what you called it — and the name that leadership ultimately chose reflected exactly what the product had become.