When Amazon launched its first Go store in 2018, the public lined up around the block to see the future of retail: a new experience where you could walk in, grab something off the shelf, and walk out. Sure, there were cameras on the ceiling and AI on computers tracking silently from above, but the promise was convenience through automation–maybe not The Jetsons, but a better 7/11 for certain.
Now Walmart has shared its version of the future of brick-and-mortar retail, the Intelligent Retail Lab, or IRL for short. Unlike Go, it doesn’t feature any futuristic user experience. There’s no automated checkout or similar whiz-bang head turner that people will Instagram about. Instead, IRL can track Walmart’s inventory in real time with unprecedented efficiency, making sure every item on every shelf is always in stock.
Rethinking the entire shopping experience, as Amazon Go has done, was not on the table. “It’s just not a priority for us right now, as we think about it,” says Mike Hanrahan, CEO of IRL (which is technically a startup within Walmart itself). Instead, the IRL store has 1,500 cameras hanging from the ceiling to ensure that when you walk up to the meat section, there’s in stock. “If you have really good inventory, it leads to a better managed store,” says Hanrahan. And a better managed store is a more profitable one.
What is the IRL store exactly?
Technically, IRL is a real, working Walmart store. It was built inside a 50,000-square-foot Walmart in Levittown, New York. This location is considered a busy Walmart, which is why it was chosen in the first place, though it’s roughly half the footprint of your average suburban Walmart Supercenter.
For the past 18 months, Hanrahan’s team of 75 has been retrofitting the store with the aforementioned 1,500 cameras (which include typical cameras and 3D depth-sensing cameras), along with various sensors on the shelves, which can measure weight, count remaining inventory when an item is taken, and even use capacitive technology, much like your iPhone touchscreen works, to measure the electrical resistance of items on the shelf to see their shapes and count them up.
All of this tech is connected to 150,000 feet of cabling (which Hanrahan eagerly points out is several times the height of Mount Everest) linking them to local servers that sit behind transparent glass, right in the customer’s view. “I think it’s always good to be transparent when it comes to the world of AI,” says Hanrahan. Nothing about Walmart’s cameras or computers is hidden. Walmart even welcomes school groups, and organizations like Girls Who Code, to tour the store.
If this all sounds insanely overbuilt, that’s only because it is. IRL is filled with every sensor Walmart could possibly want to use, so that it can train new AI models, and test various hypotheses (like can it detect a spill in an aisle and have it cleaned up, or count grocery carts to have them replaced when taken?). And somehow, all of this was installed without the Levittown location closing for a single day. That was an accomplishment unto itself. With over 4,800 Walmart stores across the U.S., neither Walmart nor its customers can afford for these stores to shut in the name of improvements.
Opening the store
The IRL store’s hardware flipped on in October. Walmart needed data with which to train its AIs–most of which operate with various types of vision recognition (which needs tens of thousands of examples of, say, a banana or an orange to understand what they are). So all it did was collect data for a few months. Around January, they began training AIs–software tools that could do little jobs, but do them well. [Image: Walmart]
One such tool is an inventory tracker for Walmart’s meat counter. It can identify every type of meat sitting on the shelf. When a customer grabs one, the AI knows just what is now missing, and an associate receives a text to restock it.
Inventory is notoriously hard for retail stores to track without literally scanning the barcode of every item in the store, Hanrahan explains. AI-infused cameras can automate the job.
That tiny function of tracking meat, Hanrahan explains, can have a huge impact on Walmart’s bottom line and the customer experience. Because if you go to Walmart to get chicken breasts, and they’re all gone, you will leave without them. Walmart loses a sale. You get angry for wasting your time. It’s bad for everyone. Multiply that phenomenon across the 30,000 different products in the Levittown store, and it begins to scale in a meaningful way. Then remember that having real-time inventory allows Walmart to update its website and app accurately, too, which helps it go elbow to elbow with Amazon and Target for quick shipping or in-store pickups. It all ties back to knowing what’s in stock.
Walmart takes AI national
Since the IRL store is a lab, its exact footprint won’t be duplicated across the U.S.. Instead, its best findings and developments will be applied and pared down across Walmart’s 4,800 exiting stores. Take the meat. Tracking the meat section isn’t difficult to scale to more stores, since Walmart does’t need 1,500 cameras to do it; just a few will do. They don’t need to be watching all the time, either; cameras can turn on every few minutes to grab a snapshot. That saves Walmart the overhead of a lot of data collection. It also means that battery-powered cameras could do the job instead of hardwired ones, which saves the trouble of wiring Mount Everest into every store.
When Walmart does roll out systems like the meat tracker to more locations within the year, “It’ll look like some cameras pointing at a shelf. Customers probably wouldn’t notice,” says Hanrahan.
While Amazon Go appears to track customers through cameras–allegedly having more than 20 people in a store at once used to crash its systems–the near future of Walmarts is just a bunch of cameras pointing at inanimate objects on the shelf. In the world of AI, which seems to do something new, crazy, and unethical every day, it almost sounds boring. But maybe boring AI is good for a change. Hanrahan insists that Walmart is not doing any sort of human tracking at all, at the moment. The IRL store pointedly doesn’t have cameras near restrooms, the pharmacy, or employee break areas to emphasize privacy, too. And if it were to eventually perform facial recognition or other analyses on customers, it would “absolutely” ask for consent first. “We could never do anything like that unless customers especially opted into that and it was going to create an experience that was really valuable to them,” says Hanrahan.
But he also makes it clear, “we’re not specifically taking anything off the table, either.”