Nicky Koscinski, a 911 communicator with Halton Regional Police Service, works with software from Hyper that uses AI to triage incoming non-emergency calls, at police headquarters in Oakville, Ont.Nick Iwanyshyn/The Globe and Mail
Most people don’t realize that when calling a police department’s non-emergency phone line – to report the proverbial cat stuck up a tree, say – you can be connected with a 911 operator who handles actual life-or-death situations.
These less urgent calls can gum up the lines and take precious seconds away from operators when they need to jump on an emergency situation.
This is where Hyper comes in. The startup, which has offices in Toronto and San Francisco, has developed an artificial-intelligence-powered voice agent to handle non-emergency calls for police departments. The system is used or piloted in Ontario, Manitoba and in the U.S.
The goal is not to handle every single call, co-founder and chief executive Ben Sanders said.
“It’s just to remove as many as we can, to save as much time as possible,” he said.
Hyper co-founder Ben Sanders works with Ms. Koscinski at the Halton Regional Police Service headquarters.Nick Iwanyshyn/The Globe and Mail
In some cases, Hyper is fielding up to 65 per cent of non-emergency calls, according to the company, easing the burden on 911 operators (or communicators, in police parlance) and allowing them to focus on more serious issues.
Mr. Sanders started Hyper in 2023 with Damian McCabe. Both have founded companies before – e-commerce financier Clearco for Mr. Sanders, and product development company Connected for Mr. McCabe. Earlier this year, Hyper raised a US$6.3-million seed round that included funding from Ripple Ventures in Toronto and Eniac Ventures in New York.
Hyper’s system can tackle minor vehicle collisions, thefts, noise complaints, lost items and suspicious activities, among others. In some cases, the bot can take information to fill out the necessary forms to report issues, inform callers about next steps and escalate problems beyond its scope to live operators. Mr. Sanders said Hyper is equipped to handle more than 100 non-emergency scenarios, though some departments start with a much narrower band.
In some cases, Hyper is fielding up to 65 per cent of non-emergency calls, according to the company.Nick Iwanyshyn/The Globe and Mail
While many startups are building AI bots to automate customer service for a variety of industries, there is a lot of value in specializing, said Ripple Ventures founder Matt Cohen.
“Everything is going very, very niche. You can’t just take an insurance voice agent and apply it to 311,” he said, referring to the number that residents in various cities can call about municipal services. He added that Hyper is well-positioned to work with government institutions, too, as Ottawa has said it wants to integrate AI into the public service.
Halton Regional Police Service, west of Toronto, switched on Hyper in October to operate 24/7 after testing the technology. The department named its bot Sara, for Smart Answering Routing Assistant, and some staff refer to the agent as “her” or “she.”
Inside the call centre, a digital board displays the number of emergency and non-emergency calls. On a Thursday morning in October, the tally read 57 emergency and 60 non-emergency, the latter of which had been routed there by Hyper.
Analysis: From tractors to ChatGPT: Lessons from agriculture on how AI will transform work
Because operators prioritize 911 calls, the non-emergency ones can pile up. “You could be waiting 10 seconds or 10 minutes depending on the call volume,” said Brian Dodd, communications supervisor with Halton police. “If you just had a question that was simple, and Sara takes care of it, that’s great.”
For those answering the phones, the mental shift from dealing with a crisis to listening to someone fume over a car blocking a driveway is a taxing one. “You tend to lack empathy and patience going to that, and it’s to no fault of anybody,” said Zakiah Taha, who has been a 911 operator with Halton for 15 years.
Sometimes, they will take a non-emergency call only to determine the issue doesn’t fall under the department’s jurisdiction. “The ability for the AI to redirect things away that are not to do with Halton at all is definitely helpful,” Ms. Taha said.
An obvious question might be: Why not hire more people?
These are hard roles to fill, Mr. Dodd said. The job comes with intense mental and emotional demands, training can take more than a year and not every applicant makes it. When he presented to the Police Association of Ontario about Hyper, members asked if the technology will replace jobs. “My answer was, you don’t even have enough communicators,” he said. “Every communications centre in Canada and the U.S. is short-staffed.”
Halton Regional Police Service fully incorporated Hyper’s software in October. The AI bot now operates there 24/7.Nick Iwanyshyn/The Globe and Mail
To show how adept the AI voice agent can be, Mr. McCabe with Hyper offered a demo by calling it up and saying he had been in “a bit of a fender bender.” The AI agent confirmed the issue and asked a series of questions – whether it occurred on a 400-series highway, whether he was in the vehicle, whether a driver failed to remain at the scene and so on.
The voice was female, polite but insistent, and the sound of a clacking keyboard filled the silences while the AI model was processing to help prevent talking over one another. The agent told Mr. McCabe to report the issue to the region’s collision centre and texted him the link. Aside from one brief lag, during which Mr. McCabe pursed his lips anxiously, the process went smoothly.
“We can really focus on the questions that are most applicable to this situation,” he said, “and there’s a decent amount of questions.”
When Ms. Taha takes a call, she immediately knows what to ask. She has helped hone the AI agent for the department, and trying to transfer her knowledge has shown her how specialized her skills are.
Hyper co-founders Ben Sanders, right, and Damian McCabe at the Halton Regional Police Service headquarters on Oct. 30. Mr. Sanders says the goal of Hyper’s software is not to field all non-emergency calls to police, but to ‘remove as many as we can, to save as much time as possible.’Nick Iwanyshyn/The Globe and Mail
Unlike the human mind, AI models have trouble generalizing and can be thrown off by edge cases. In one example in Halton, Hyper took a call from a man whose parked car had been struck by another one. The bot told him it was not necessary to report the incident since he wasn’t in the vehicle at the time. Ms. Taha knew it should have been reported because the caller owned the vehicle, something the bot did not ask.
“These very minute pieces that you wouldn’t think would have a profound impact on misdirecting the call actually do,” she said. (The questions around fender benders have been revised to be more precise.)
The Winnipeg Police Service switches on Hyper’s AI agent for a few hours each day, with the goal of eventually replacing its legacy phone tree system. There has been a learning curve for callers. At first, some barrelled into calls by bleating “Eight, eight” in hopes of reaching a live operator.
“There’s some people that just hate it,” said Gord Spado, an inspector with WPS. “We had people actually yelling at the AI.” (Some people hate the phone tree too, he added.)
The department has seen benefits as the kinks are worked out. The percentage of hang-ups, for example, has fallen from 35 per cent under the legacy system to 25 per cent when Hyper is running. “Overall, I think it’s doing very well,” Mr. Spado said. “The AI being more conversational has actually been fairly well-received from the public.”
While Hyper is focused on recruiting more customers (Mr. Sanders said the number of deployments is in “the teens”), it’s also looking to improve its technology. Departments retain years of recorded calls that can be used to train future AI models, for example.
“It’s really been hard to get this to work right,” Mr. Sanders said. “But we’ve done more than 10,000 calls this month, and it’s really starting to take off.”