How is Walmart using AI to engage with customers?
We are innovating ways in which we can serve our customers through AI. We are changing the way in which we run our business through AI — how we move our products all the way from manufacturers, through our warehouses and to the store. Physical is meeting digital.
The power of AI is that it can not only understand and personalise the experience but also help understand the context of what you are trying to do. AI is democratising access to information and insights in a way where our associates get access to data and insights a lot more naturally and faster.
What AI-led features do you have for customers?
We have rolled out Sparky, our AI-enabled agent. With Sparky, instead of having to shop individual products at a time — because of AI’s ability to understand context — you can tell it what it is that you are trying to do. Instead of going from keyword-based shopping, it [the process] becomes much more natural.
We also have Code Puppy, which allows non-engineers such as merchants and store managers to create a dashboard to help solve problems. The process used to take several hours by hand but takes minutes now. That’s being adopted across hundreds of different stores.
AI allows us to have real-time intelligence in optimising how we serve customer orders in a way that is most efficient and meets the needs of the SLAs (service level agreements). This becomes very important in the real world because it is always messy. Real-time optimisation is a big unlock.
How have such agents improved customer engagement and developer efficiency?
Our basket sizes have gone up by 35 per cent compared with the time when customers were not using Sparky. Wibey, which our engineers use internally, has helped us a lot in what we call adopting a platform way of building. That means instead of building systems individually for every country, we are now building a common set of capabilities from which every country benefits. We go from multiple disconnected systems, individually built, to a common set of capabilities. This transformation is being hugely accelerated through Wibey. Rolling out one particular feature across Walmart US, Sam’s Club, Mexico and Canada would have taken us six to seven months. That has compressed to several weeks.
What is the most significant work that your India tech centre has done?
One area we started working on is our cataloguing. We sell millions of different products. When our suppliers give information to us, many times it is either incomplete or inaccurate. This used to take a lot of effort manually to go look at what the right information is and curate it. We started using AI to automatically look for defects and clean them. This work started in India and is now actually spreading across all of our different centers.
What was the critical reason for downsizing, which also impacted the Indian centre?
I want to separate two things. There is AI but there is also platformisation. Instead of building things separately for Mexico, Canada, or the US, a couple of years ago we started an effort to be able to build common capabilities. This transformation meant that the way in which we work shifted. That is the primary emphasis for some of the changes that we made last year. It had nothing to do with AI but more to do with the way in which we were actually building common capabilities rather than a lot of duplicated stuff.
Are there any other operations where AI is becoming critical?
We are now starting to see AI enter areas that traditionally have been where the capabilities were not there. One is around being able to orchestrate across different workflows. Take supply chains as an example. We used to optimise last-mile delivery as its own thing. We used to optimise replenishment inside the store as its own thing. But these two are tied together. How you replenish inside the store really depends on how you forecast how many products need to be pulled out and how you optimise the delivery of these products. It was very hard to build models to be able to look across these kinds of domains.
One part of it is really to make sure that the work that we have done in terms of building custom models to be able to optimise in different pieces continues to stay. It is also about looking at how signals can flow from one part of a system to another. And look at the capabilities of AI in being able to tie different signals together. But it’s much more about making sure that the main models that we have built can actually start talking to each other.
 "Suresh")