Podcast and video: How this guy took center-store AI tech to the perimeter for ordering efficiency

Need to be active (or drive) while you learn? Listen to the podcast instead.

Matt Schwartz’s career has been shaped by his desire to make an impact.

“I believe that food more so than any other industry shapes the health of our planet and the health of people,” said Schwartz, cofounder and CEO of Afresh, which partnered with Albertsons Cos. in January to offer its artificial intelligence platform for forecasting, inventory, ordering, merchandising and operations.

Schwartz started a snack food brand and studied food and agriculture while earning his master’s degree in business at Stanford University.

Then he shifted his strategy to the perimeter of the grocery store, where he believed the focus of food was heading.

“At the same time, though, I observed that all of the critical technology built for retailers across functions like inventory management, merchandising, store operations, supply chain and more was built first from the center store, and stuff that had a box and had a barcode and lasts forever,” he said. “And this gap, as a result, left retailers with really tough challenges. How do you manage fresh food without technology?”

Fresh produce relies on the sophistication and hard work of store employees to use their judgment and workflows to try to make the best possible decisions.

Schwartz and his team discovered that, while there are incredible results that way, there was still more opportunity to reduce food waste, increase in-stock, improve freshness and help those store teams be even more effective at what they do.

“And so that was the thesis behind Afresh: the future of food is fresh,” he said. “The technology was all built for the CPG center store. If we could build it instead for fresh-food-first technologies, we could massively reduce food waste and multiply the profitability of retailers and really delight customers that are looking for the freshest, best offerings that retailers have to offer.”

Schwartz and his colleagues at Stanford interviewed more than 100 companies in the fresh food supply chain, looking for inefficiencies to find areas for intervention. In 2017, they observed that, across the industry, store ordering in produce was one of the most broken areas, but also one of the most foundational for a retailer because the store order ends up driving the distribution center-level order, and it drives the companywide procurement decisions.

And it drives the planning of the grower and distributors. And it also connects to merchandising, such as space allocation, pricing, promotions, markdowns and more.

“And so, we centered in on that because of the relative lack of technology that was available there, and how foundational it is to the rest of the supply chain,” he said.

The vision was to nail down and solve that problem and then scale it for around the rest of the store perimeter, expand across the supply chain and, finally, across different retail functions with the technology.

Some of the initial retailers he worked with said, “Why are these tools not built for us?”

That’s what fresh-first technology is all about.

So, what is it, exactly?          

In this artificial intelligence, or AI system, the first product is a store-level ordering tool for produce, and it covers the entire produce department.

Any item that's ordered in the produce department — fruits, vegetables, juices, croutons and whatever else the produce manager might be responsible for ordering — will be ordered by Afresh’s system.

The goal of the system is to optimize the sales and minimize shrink.

“What that looks like in practice is, we want a full shelf and an empty cooler or back room when the next truck comes in,” he said.

This method is not reliant on the traditional perpetual inventory system, he said.

“We're actually able to use artificial intelligence combined with the workflow of a produce manager in the store to override the need for a perpetual inventory and to drive accurate ordering,” Schwartz said.

Perpetual inventory

Perpetual inventory is the traditional way retailers try to understand digitally how many of a particular fruit or vegetable item are in the store. Take avocados, for example.

Think about how a store would determine the right number of avocados to order today, he said.

One question to ask: How many avocados do I have right now? Another one: What's my projection of how many avocados I'm going to sell over a period of time?

And then, there's an order calculus based on how much the produce manager thinks they'll sell and how much they will need, which could then also be determined by when the next truck is coming in.

Also: What is the holding power of the shelf? For instance, how many avocados or how heavy does a shelf need to be — what’s the number of cases? How big is the case? How perishable is the product?

“Solving the demand forecast is really hard to do in fresh because you've got items that are sold by weight,” he said. Also, promotions, growing season results and varietals affect the demand, too.
“There are a lot of questions like that that are extremely hard to predict, in addition to promotional effects and more. So that's a hard challenge in and of itself, but it was often overlooked, was this notion of inventory,” Schwartz said.

“Someone doing a period-ending inventory cycle count would say, ‘I’ve got 10 avocados today, I shipped in 10 on this last truck, so that should be 20. I sold five. So, I should have 15 left’” he said.

And that is how my perpetual inventory system works.

“The challenge in avocados is that they're going bad and being thrown out. And they're tough to track weights, because what are you going to do, drag a scale around and weigh them out as they go, when they're sold away? And that's actually something that retailers have tried doing before to solve this problem,” he said.

At the cash register self-checkout, the avocado might be rung as conventional when it's really organic because customers want to save some money, or the cash register makes a mistake. A supplier could ship a 20-pound case of them, but there's only 19 pounds in them.

The list goes on and on and on.

“And all of that, in turn, makes that perpetual inventory radically inaccurate,” he said. “Just understanding how many avocados are there is a really unsolved problem in the industry. And it sounds so simple, but I think any retailer who's worked in produce understands the profound challenge of that. And so, what we've done is build a system that doesn't rely on those perpetual inventory systems that really struggle in fresh food.”

AI teams with produce people

Afresh built a system in which the AI teams up with the store teams.

“We don't believe in fully automating away the process; we want to make it more efficient and automate what we can. But at the end of the day, we believe that nobody knows better than the produce manager who has his or her boots on the ground and has visibility into product,” Schwartz said. “And maybe some of the events and merchandising that could be happening that are unknown to the digital system. At the same time, we want to empower them with AI to make better decisions where they can.”

Afresh uses AI to make a better guess at inventory.

“We basically say ‘We think you have between eight and 12 avocados in the store.’ And when that range gets too wide, let's say it's now zero to 20, we're not sure. Then, we might ask the purchase manager to do exactly what they do today,” he said.

The purchase manager will use pen and paper to do a spot-check of how many avocados are there for a subset of the items. And then, the demand forecast for order-optimization kicks in with the AI that then generates a profit-maximizing, waste minimizing order for every item.

“So, the big difference is that we built this workflow to really accommodate the existing workflow of produce managers in stores and other fresh department managers, so that they can be maximally effective teaming up with the system,” Schwartz said.

To learn more, listen to the podcast or watch the video 14 minutes in.