Artificial intelligence at Walmart does not begin with a robot replacing a cashier or some futuristic vision of retail.

It begins with much more ordinary problems.

A supervisor spending an hour and a half planning an overnight shift. A fashion team trying to get a trend onto store shelves before the trend has passed. A merchant working through reports and spreadsheets to understand why a product is not selling. A supply-chain employee spending hours figuring out why particular stores did not receive enough merchandise.

Walmart has been applying AI to each of those problems. The interesting question is not simply what technology the company deployed.

It is what was hurting, what Walmart changed, and what happened afterward.

Ninety minutes before the work begins

Consider an overnight stocking shift.

Merchandise has arrived. Shelves need replenishing. Different areas of the store have different priorities. Before the crew can be directed toward the work that matters most, somebody has to make sense of the workload and organize the shift.

Walmart says its associates were navigating what it describes as a complex, tool-based environment. For team leads, turning that information into an actionable plan could consume about 90 minutes.

Walmart introduced an AI-directed task-management system designed to interpret the workload, prioritize tasks and recommend where associates should focus their efforts.

The manager is still managing the shift. The difference is that the first version of the work plan no longer has to be assembled entirely through the old process.

According to Walmart, team leads and store managers involved in the initial overnight-stocking deployment estimate that planning time has fallen from 90 minutes to about 30 minutes.

The technology matters. But the operational result is easier to understand:

An hour of a supervisor’s shift came back.

Getting a fashion trend onto the rack before it disappears

Fashion creates a different problem.

By the time a retailer recognizes a trend, researches it, develops a collection, communicates specifications to suppliers, manufactures the merchandise and gets it onto shelves, customers may already be interested in something else.

Walmart’s designers traditionally researched sources including fashion shows, social media and other trend information. Ideas then had to become colors, textures, styles and mood boards before eventually becoming a technical package detailed enough for a supplier to manufacture the product.

Walmart built a system called Trend-to-Product to compress that process.

The system analyzes trend information and uses generative AI to produce initial concepts and mood boards. Human designers and merchants then refine those ideas, incorporate sales data and apply their own judgment. Once the collection has been developed, the system can generate the technical package sent to suppliers.

Walmart says the research-and-design portion of the process can move from weeks to minutes, while an AI-assisted concept can be developed in roughly an hour.

More consequentially, Walmart says the overall process can put merchandise onto shelves in six to eight weeks — as much as 18 weeks faster than its traditional timeline.

That changes the problem AI is solving.

This is not simply about designers producing mood boards faster. For a retailer, eighteen weeks can be the difference between recognizing demand and actually being able to sell something while that demand still exists.

“Why isn’t this product selling?”

Now consider a Walmart merchant responsible for a product that is underperforming.

Finding out why can require more than looking at the sales number.

The merchant may need to examine several measures, run multiple reports, work through spreadsheets and combine the results before reaching a useful explanation.

Walmart says that process required significant time and effort.

Its response was Wally, a generative-AI assistant built around Walmart’s proprietary data.

Rather than manually assembling all of that information first, a merchant can use Wally to analyze complex datasets and investigate why a product is performing above or below expectations. The system also handles some data-entry work and can answer operational questions.

Here Walmart has not publicly supplied the kind of clean before-and-after number it provided for shift planning.

That distinction matters.

What Walmart has documented is a change in the workflow: multiple reports, complicated spreadsheets and manual synthesis are being replaced in part by a system capable of interrogating the underlying data and helping identify the likely cause.

The merchant still has to decide what to do about it.

But the work required to reach the question “What should we do?” is changing.

When the problem is somewhere inside the supply chain

The same principle becomes more consequential when the question involves inventory.

Imagine several Walmart stores received fewer units of a product than expected.

Someone has to determine which products were shorted, where the shortages occurred, what the underlying inventory data shows and what should happen next.

Walmart says that kind of investigation can require hours of analysis.

Its newer agentic-AI tools allow an associate to ask questions such as which items were shorted across particular stores. The system can analyze the information and return both insights and recommended next steps.

Walmart describes the change as moving from hours of analysis to seconds of action.

Again, the useful part of the story is not that Walmart has an AI agent.

It is that an employee previously had to spend hours turning scattered operational information into a decision. The company is attempting to collapse much of that intermediate work.

The pattern underneath the technology

These are four different jobs inside one enormous company.

The overnight supervisor is not doing the same work as the fashion designer. The merchant is not doing the same work as the supply-chain employee.

Yet Walmart appears to be applying AI to them in a remarkably similar way.

First, find where work is getting stuck.

Then identify the information people need to move forward.

Then use AI to perform some of the searching, organizing, analyzing or prioritizing that sits between the problem and the human decision.

The result is not always a job disappearing.

Sometimes it is 90 minutes becoming 30.

Sometimes it is weeks of research becoming minutes.

Sometimes it is a merchant reaching the useful question without first wrestling with several reports and spreadsheets.

And sometimes it is hours of inventory analysis becoming seconds.

Those figures should also be understood for what they are: results reported by Walmart, not independent measurements conducted by Zen Plasma. In several cases, Walmart has provided considerably more detail about the new workflow than it has about how the reported performance gains were measured.

But the implementations provide something more useful than another prediction about what artificial intelligence might eventually do.

They show what happened when one of the world’s largest employers found specific pieces of work that were slow, repetitive or difficult to navigate — and started handing parts of them to machines.