How AI powers Sam’s Club’s holiday product assortment

When a member of one of the 600 Sam’s Clubs across the U.S. wants a fresh pumpkin pie to serve this holiday season, Vinod Bidarkoppa, SVP and chief technology officer at Sam’s Club, wants to make sure it’s on the shelf and ready to go.

Unlike mass-market retailers — like Sam’s Club’s parent company Walmart — satisfying and retaining members is the most important strategy for a membership-only warehouse chain. That’s why Sam’s Club has steadily increased its artificial intelligence (AI) efforts, based on connected intelligence. Since Bidarkoppa joined the company in February 2020, he said his team has worked hard to figure out how to make sure the right assortment of products gets into the right stores at the right time — especially during the peak holiday season.

“With accurate AI forecasting, last year we sold enough pies to cover the length of 450 football fields while staying ahead of customer demand and reducing food waste,” Bidarkoppa told VentureBeat. “If we have actually delighted our members and every member who wanted a pumpkin pie got it, that’s a victory for us.”

Sam’s Club needs to understand demand and inventory

The journey of Sam’s Club’s pumpkin pies begins by determining how many pies members will buy across the 600 clubs — and, therefore, how many need to be ordered from suppliers and shipped to stores.

“We do a machine-learning [ML] based forecast looking at a number of variables, including local weather conditions, holiday dates and shelf life, and come up with an unbiased, unconstrained demand forecast,” he said. “Then, suppliers replenish pumpkin pies into distribution centers, where machine learning also exists to help plan the most efficient way to ship all of those pumpkin pies to the 600 clubs.”

After they arrive at stores, the pumpkin pies — which arrive semi-frozen and need to be prepared for sale — need to be stocked on the shelves. But how many are needed to keep up with member demand while not allowing leftovers to go to waste? That’s where AI and ML also play a part.

“If you overstock, there’s a lot that’s going to be wasted,” he said. “But if you’re under-produced, then you have a disappointed member who says, ‘What’s the holiday without a pumpkin pie?”

The company’s Fresh Sales App uses AI to predict — in some cases as much as hourly — how many pies or other products need to be stocked, based on ML forecasts generated for each item at the club.

“This system takes into account the minimum presentation level  — how much inventory I want in that club at a particular time,” he said. “So it almost always feels like it’s full and we don’t disappoint the members.”

AI tools capture store shelf data

Last week, Sam’s Club unveiled its latest computer vision-based ML tool, an autonomous floor scrubber that does, indeed, clean the floors but also has a mounted camera that scans and captures all the items on the store’s shelves twice a day. The scrubber then sends “hot item” alerts to the backroom, so associates can prioritize the receipt of items most at risk of selling out.

“We have patented models around object detection which have roughly about 90% accuracy and coverage of the entire club space,” Bidarkoppa said.

Once a truck filled with pumpkin pies arrives from the distribution center, Sam’s Club’s ClubIQ tool prioritizes in what order associates need to unload the pies and other items from the truck. The system also analyzes historical item sales behavior for upcoming holiday events and tells associates which items to put on the floor first based on the popularity of that item – down to the club level.

Connecting Sam’s Club’s AI ecosystem

“This is an ecosystem that exists in a machine-driven model,” Birdarkoppa explained. “In the end, the goal is really to delight our members, but powering all of this is the AI behind it.”

This connected AI and ML-powered ecosystem also continues to learn over time.

“It doesn’t stop with the data signals we pick up,” he said. “We feed it back into the model and these become learning systems. So, the more we learn from the signals, the better the models become and the better the outcome is for both our associates and our members.”

Club associates, he insists, love the inventory intelligence and item recognition technology.

“Instead of having to manually find out all of these details, the machine helps them very precisely,” he said. “They love the fact that there are things that help our members instead of having to do unnecessary administrative manual tasks — today we are north of 15% of tasks that are automatically handled by because of this technology, and imagine that at our scale.”

Sam’s Club plans to build on AI success

Birdarkoppa says Sam’s Club plans to build on its AI efforts — which clearly go beyond pumpkin pies.

“Whether it’s planning, buying, planning, moving or selling across all of those core retail processes, we will continue to double down on machine learning,” he said. “There’s a long list of things that are on my to-do list and really excited with what we can do with this technology, for both our members and associates.”

As a membership-driven organization, Birdarkoppa again emphasized the company’s need to have a member-focused lens across all of its processes and tasks.

“That bleeds into the technology stack and how we work within the tech organization,” he said. That applies to AI-driven tools, too — for example, one of the objective functions in any Sam’s Club algorithm is how it will affect membership retention and growth.

“In the end, we are in the business of making our members happy,” he said.