Gyanesh Kumar Jha Argues Retail Automation’s Hardest Problem Isn’t the Algorithm

Ask what actually slows down AI adoption in retail, and the honest answer is rarely the model itself. Most large retailers can buy or build a demand forecast. Getting merchandising, engineering, and store operations across four countries to run on the same forecast is a different problem, and it is closer to the one that Gyanesh Kumar Jha has spent his career solving.

Jha is a Staff Tech Operations leader at Walmart, where he leads technical program management for automation and optimization platforms within the company’s International Merchandising Technology group.

“I lead technical program management for automation and optimization platforms within Walmart’s International Merchandising Technology, Space, and Assortment organization, supporting international markets including Mexico, Canada, Central America, and Chile,” Jha said. “I manage a cross-functional team of more than 75 across product, engineering, and program roles.”

None of that is an argument against investing in better models. Retail forecasting has gotten measurably better because the underlying algorithms have. But Jha’s career suggests the industry has been solving the easier half of the problem. A model that predicts demand accurately in a lab, or even in one market, still has to survive contact with the merchandising team that doesn’t trust it. The engineering team built it slightly differently than the spec described. The store operations team finds out about the change two weeks after everyone else did.

That is the half Jha has spent most of his working life on. Before Walmart, he led the SAP implementation across mainland China, Hong Kong, and Taiwan for the audio company BOSE, then moved to Belgium as a Scrum Master in a five-team implementation of the same client’s systems across Europe. He later led a technical program for Mars from Shanghai, where the account’s delivery team grew as Tata Consultancy Services expanded its contract with the client. Two decades of that kind of work, spread across India, China, Belgium, and the United States, led him to the U.S. in 2019, first consulting Walmart’s technology organization and then with Reckitt, before Walmart hired him directly in 2022.

In each of those engagements, the technology mattered less than getting a distributed set of teams to agree on how to use it. A forecasting model can be technically sound and still fail in production if the team responsible for acting on its output doesn’t trust it, doesn’t understand how it was built, or is working from a different version of the underlying data than the model was trained on. None of that shows up in an accuracy score. It shows up months later as a project that quietly falls out of use.

Jha is careful about what he’ll put a specific number on, and this piece leans on that caution rather than around it: no internal performance figures appear here, by his own preference. What he will describe is the shape of the coordination itself, a team of more than 75 people spanning three functions, aligned across four markets with different store formats and different retail habits. That alignment is the actual product, whether or not it ever gets its own line item.

If there is a lesson in that for retailers further behind on automation, it has less to do with which model they choose and more to do with whether merchandising, engineering, and operations are already working from the same plan, in the same language, before anyone hands them a smarter algorithm to argue about. Most companies find that out the hard way, usually after the algorithm has already shipped.