By Sharon Fisher | May 17, 2022
Every company knows stuff about its customers. Too much stuff. There’s so much material that it can’t be summarized into anything actionable.
That’s where SturdyAI comes in.
“We take all of the communications that a business has with its customer – support tickets, email, chats — organize them in one spot, and use artificial intelligence or natural language processing to extract relevant relationship triggers,” explained Steve Hazelton, founder of SturdyAI. The software “de-noises” the message stream and then uses robotic process automation to route information to the right people, such as sending happy customers to marketing to act as references, and looking for unhappy customers to help them out. “Someone who reports three bugs in 90 days is four times more likely than someone else to cancel, so we identify and triage that,” he explained.
“There’s a huge amount of lift a business can get when they use data they never thought of using,” Hazelton said. “All this information and they have no data at all.”
Identifying grumpy customers
Hazelton came up with the product because it was something he needed himself. “I’ve run businesses my whole life,” he said. “I started my first company when I was 24 or 25. When you’re running a business, you tell the people working for you, ‘If our most important customer gets grumpy, let me know.’”
At Hazelton’s previous company, he brought people on board and trained them to identify these relationship signals. “It was all done manually,” he said. “It’s fragile. It requires someone to have empathy and cognition to recognize these signals.”
But that doesn’t scale as a company grows. “When you’re a small company, that happens telepathically,” Hazelton said. “When you have 50 customers, you know which ones are happy and which ones want to be left alone. As the company gets larger, you lose that.”
So when Hazelton’s company got bought by another company with 25,000 customers, it became virtually impossible to manage all the customer relationships, he said. “As you get larger, you agree that you’ll let customers fall by the wayside,” he admitted. “You say, ‘What’s going on with our best customer?’” which requires getting five people in a room to discuss it, he said. “It doesn’t make sense to do it that way.”
Training a system
So in 2020, Hazelton went back to a previous business he’d sold and asked them for 200,000 to 300,000 tickets to train a language model, he said. “The cool thing about what we do is it isn’t ‘creepy AI,’” he said. “When you contact a business – ‘this product didn’t work today’ – you want them to do something with that message, whether it’s route it to support or billing or sales.”
While the product got a few pilot customers last year, it was officially launched in January and now has about 10 customers generating about 200,000 messages a month, Hazelton said.
Hazelton is not your typical tech entrepreneur. He went to school at the University of New Hampshire, studying political science, and worked for the New Hampshire Legislature his senior year, which he described as the “third-largest legislative body in the world,” after the U.S. Congress and the United Kingdom’s Parliament.
When Hazelton graduated, he managed political campaigns, but soon gave that up. “I was sick of being poor,” he said. “You don’t do well managing state senate campaigns,” because you had to get a new job every two years, he said.
Instead, in the late 1990s, Hazelton got hired by a fellow running a recruiting firm in Ketchum. “The model was, once you got good enough, he’d give you money to start your own firm.” Eventually, he moved to San Francisco and bought the fellow out.
But Hazelton didn’t really like that business, so he started doing outsourced recruiting instead, which did well enough that the company had to build software to manage its business. Eventually, he sold off the recruiting and human resources part, and kept the software part, starting what he said was the first cloud-based applicant tracking system.
“That was the right place at the right time,” Hazelton said. “Before, it you were in human resources, you had to use Outlook or paper.”
After building that company up to 2,400 customers, Hazelton sold it to Paycor, which at the time was the fourth-largest payroll company in the country. He worked for them for two years and then retired.
“I’m terrible at retirement,” Hazelton explained. “I was retired for about six months. It was a very unsuccessful retirement.”
That brought Hazelton back to Idaho, along with family reasons. “One of my goals for moving back was to bring more tech business to the Boise area,” he said.
Hazelton was thinking about getting angel funding for SturdyAI in the fall of 2020, but when he talked to one investment company, they said, “Why don’t you do a seed round and get it over with?” So he raised $2 million in a seed round from a company based in the Bay Area.
Unlike some of Idaho’s other entrepreneurs, Hazelton hasn’t been doing the competition circuit. “I have not done any of the startup competition in Boise,” he said. “The level of money we’re trying to raise, $25,000 checks don’t cut it. That wouldn’t even cover payroll for a week.”
Instead, the company is likely to close another round of seed funding from a company outside Idaho in the next couple of weeks, Hazelton said.
Sturdy AI, which currently has just Hazelton in Boise and several other employees working remotely in various places, will be doing some hiring in Boise – three people or so in the next couple of months, and a total of 25 or so in the next couple of years – primarily in areas such as product design, content, digital marketing and strategic marketing, Hazelton said. He’s also considering a Boise office in the downtown or North End area, he added.
However, Hazelton hasn’t settled on an exit strategy yet. “It’s too early to talk about that now,” he said. “I don’t plan businesses for the exit, I plan them for the growth. Our goal is to hit as many customers as possible right now. It drives tremendous value when you can start mining this stuff. Now we just have to get out and start selling it, and then we’ll think about an exit strategy.”