Eighteen months after putting an AI receptionist on its phones, a Saskatchewan brokerage says most of its early mistakes came from asking the technology to do too much at once.
Brock Longworth, chief operations officer at Prince Albert-based Cornerstone Insurance Services, told Insurance Business his main advice for other brokerages is simple.
“I would say narrow the scope of what you want it to do,” he said.
Read more: Your group life insurance won't cover your mortgage, broker says
Cornerstone started using voice AI from Sonant in June 2025 to answer and route incoming calls. The brokerage handles far more calls than most, with peak traffic of 300 to 400 a day, Longworth said.
“There was no number of receptionists that could ever handle peak call times,” he said.
The original plan was to have the AI answer every call under its own name, Quinn.
“We backed off on that within the first four days because it’s clearly not working,” Longworth said.
The brokerage also gave the system an enormous script covering what it could and couldn’t do, then spent the first year pulling it back, he said.
“I think you should go in with a very specific use case and just launch on that and then build out,” Longworth said. “Don’t go in thinking it’ll do everything.”
Early on, the system struggled with the nuances of the brokerage’s workflows, he said. It could send a customer renewing their auto policy to the right team, but if that team didn’t answer and the caller switched to a question about home insurance, it lost its way.
“It would get stuck in a loop where it wanted to do one thing and it had a hard time breaking out of it to do something else,” Longworth said.
It also had trouble drawing out what callers wanted when they were vague, and knowing to route a call to a backup when a staff member was away, which happens regularly in a company of 50 or 60 people, he said. Those problems have eased considerably over the past year and a half.
Staff were the first concern. When reception lines were full, every phone in the office rang, so the AI promised relief, but the initial reaction was wary, Longworth said.
“I think our staff’s initial reaction was: ‘Are they trying to replace me?’,” he said.
The brokerage held a company-wide meeting before launch to make clear that it wouldn’t, and let employees test the system ahead of time, Longworth said. Most ended up excited about it, he said.
Read more: A rogue AI agent could hack your company, and some cyber policies may not cover it
“We still have the same number of staff, lots of job postings out, so we’d love to hire some more people,” he said.
Customers were harder to read. Cornerstone has received a couple of negative Google reviews about the system, though Longworth said positive reactions match or outnumber the negative ones. He and a colleague listened to every call for more than a year and found no consistent profile among those who disliked it. Problems tended to arise when a caller was on a poor Bluetooth connection or had a television on in the background, he said.
“It’s more circumstantial, I think, than it is individual,” Longworth said.
The brokerage has lost a few clients over the AI, he said, but it had tracked losses before the system too, and those were worse. The retention problems caused by unanswered phones “were significantly more impactful to us than the odd client who doesn’t like AI,” Longworth said.
The clearest gain was capacity. The system can take what is effectively an unlimited number of calls, Longworth said, and staff got back focused time for the work they are meant to do.
“We don’t have to have customers calling in and getting a dialling sound or going right to voicemail anymore, which is really nice,” he said.
Being upfront about the technology also helped. Callers now choose between the “digital agent” and a human receptionist, and the AI introduces itself as a digital agent rather than by name.
“We’re double announcing that it’s AI,” Longworth said.
To limit E&O exposure, the AI can explain general terms such as a deductible but cannot address a customer’s specific situation, and must say a broker needs to confirm. It can see into the brokerage’s management system but is barred from sharing coverage details.
“We’ve given it that box to work in with very general stuff,” Longworth said.
Read more: Canadians ignore their group benefits until a crisis hits, broker says
A second tool listens to calls between staff and customers, flags missed points against criteria the brokerage sets, sends managers a next-day report and files a summary to the client record. Coaching staff on how to handle customers who object to the AI also mattered, he said.
Cornerstone now plans to carefully loosen some restrictions, such as letting the system give callers their policy number, Longworth said. It is also preparing an outbound agent to call people who request online home quotes, collect missing information and book them with a broker.
Finding the right use for outbound calls took some trial and error, he said. At one point, he said, “I had a solution for a problem I didn’t actually have.”
“I think that the next step of the evolution of this system is opening its capabilities up slowly and carefully,” Longworth said.
