How a 16-year-old company is easing small businesses into AI | TechCrunch

Amid all the “is this a bubble?” talk about artificial intelligence, the supply chain and logistics industries have become breeding grounds for seemingly genuine uses of the technology. Flexport, Uber Freight, and dozens of startups are developing different applications and winning blue-chip customers.

But while AI helps Fortune 500s pad their bottom line (and justify the next layoff to Wall Street), the right use of the tech is proving useful to smaller businesses.

Netstock, an inventory management software company founded in 2009, is working on just that. It recently rolled out a generative AI-powered tool called the “Opportunity Engine” that slots into its existing customer dashboard. The tool pulls info from a customer’s Enterprise Resource Planning software and uses that information to make regular, real-time recommendations.

Netstock claims the tool is saving those businesses thousands. On Thursday, the company announced it has served up one million recommendations to date, and that 75% of its customers have received an Opportunity Engine suggestion valued at $50,000 or more.

While tantalizing, one of those customers — Bargreen Ellingson, a family-run 65-year-old restaurant supply company — was initially apprehensive about using an artificial intelligence product.

“Old family companies don’t trust blind change a lot,” chief innovation officer Jacob Moody told TechCrunch. “I could not have gone into our warehouse and said, ‘Hey, this black box is going to start managing.’”

Instead, Moody pitched Netstock’s AI internally as a tool that warehouse managers could “either choose to use, or not use” — a process he describes as “eagerly, but cautiously dipping our toes” into AI.

Moody says it’s helping avoid mistakes, in part because it’s sifting through myriad reports his staff uses to make inventory decisions. He acknowledged the AI summaries of this info are not 100% accurate, but said it “helps create signals from the noise” quickly, especially during off-hours.

The “more profound” change Moody’s noticed is the software made some of Bargreen Ellingson’s less-senior warehouse staff “more effective.”

He highlighted an employee in one of Bargreen’s 25 warehouses who’s worked there for two years. The employee has a high school diploma but no college degree. Training this employee to understand all of the inventory management tools and the forecasting information Bargreen uses to plan inventory levels will take time, he said.

“But he knows our customers, he knows what he’s putting on the truck every day, so for him, he can look at the system and have this prosaic AI-driven insight and very quickly understand whether it makes sense or doesn’t make sense,” he said. “So he feels empowered.”

Netstock cofounder Kukkuk told TechCrunch that he understands the hesitancy around new technologies — especially because so many products are essentially mediocre chatbots attached to existing software.

He attributes the early success of Netstock’s Opportunity Engine to a few things. The company has more than a decade’s worth of data from working with retailers, distributors, and light manufacturers. That data is tightly protected to adhere to ISO frameworks, but it’s what powers the models that make the recommendations. (He said Netstock is using a combination of AI tech from the open source community and private companies.)

Each recommendation can be rated with a thumbs up or thumbs down, but the models also get reinforced by whether the customer takes the suggested action or not.

While that kind of reinforcement learning can lead to weird, sometimes harmful results when applied to things like social media, Kukkuk said he’s chasing different incentives.

“I don’t really care about eyeballs, you know?” he said. “Facebook and Instagram care about eyeballs, so they want you to look at their stuff. We care about: ‘what is the outcome for the customer?”

Kukkuk’s wary of expanding those interactions due to the limitations of current generative AI tech. While it might make sense for a customer to converse with Netstock’s AI about why a recommendation is or isn’t useful, Kukkuk said that could ultimately lead to a breakdown in accuracy.

“It’s a tightrope to walk, because the more freedom you give the users, the more freedom you give a large language model to start hallucinating stuff,” he said.

This explains the Opportunity Engine’s placement in Netstock’s typical customer dashboard. The suggestions are prominent, but easily dismissed. Google Docs cramming 20 AI features down a user’s throat, this is not.

Moody said he appreciated that the AI isn’t in-your-face.

“We’re not letting the AI engine make any inventory decisions that a human hasn’t looked at and screened and said, ‘Yes, I agree with that,’” he said. “If and when we ever get to a point where they agree with 90% of the stuff that it’s suggesting, maybe we’ll take the next step and say ‘we’ll give you control now.’ But we’re not there yet.”

It’s a promising start at a time when many enterprise deployments of generative AI seem to go nowhere.

But if the tech gets better, Moody said he’s nevertheless worried about the implications.

“Personally, I’m afraid of what this means. I think there’s going to be a lot of change, and none of us is really sure what that’s going to look like at Bargreen,” he said. It could lead to there being fewer data science experts on staff, he suggested. But even if that means moving those employees out of the warehouse and into the corporate office, he said preserving knowledge is important.

Bargreen needs people who “deeply understand the theory and the philosophy and can can rationalize how and why Netstock is making certain recommendations,” and to “make sure that we are not blindly going down” the wrong path, he said.