CompScience runs AI on the security cameras a company already owns, spotting the hazards that turn into injuries - and selling insurance that pays you to prevent them.
It is 6 a.m. on a loading dock outside Atlanta. A forklift backs up a fraction too fast, a worker steps into its path for half a second, and then the moment passes. Nobody is hurt. Nobody even notices. On most days, in most warehouses, that is the entire story - a near-miss that vanishes the instant it ends.
At a company running CompScience, the story does not vanish. The same camera that has hung in the corner for years, doing nothing more demanding than recording footage no one will ever watch, just flagged that half-second. A model scored it against more than fifty known hazard patterns. By the time the safety manager pours her coffee, there is a clip, a location, and a note that this exact corner has now generated its third close call this month.
That is the whole pitch, really. CompScience did not invent a new sensor or a smarter hard hat. It taught the cameras you already bought to notice the things humans are too busy, too tired, or too far away to catch.
"Our AI system provides leaders with 300 times more data than traditional safety inspections."
— Josh Butler, Founder & CEOHere is the uncomfortable arithmetic of workplace safety. A safety inspector walks the floor maybe once a quarter, clipboard in hand, and takes a snapshot of a single moment. The other 99.9% of the time - the night shift, the rushed Friday, the corner nobody walks past - goes unobserved. Decisions about who gets hurt and how much it costs get made in that darkness.
Insurance, meanwhile, has historically been a business of waiting. A worker gets injured, a claim gets filed, an adjuster gets involved, and everyone agrees, a little too late, that something should have been done. The industry calls this "risk management." It is mostly accounting for risk after it has already cashed its check.
CompScience's founders looked at that and saw a data problem hiding inside a human one. The hazards were not invisible. They were just unwatched. And the one device capable of watching everything, all the time, was already bolted to the ceiling of nearly every workplace in America.
"Video data is the next wave of IoT. When you solve that data problem, you get full context on hazards."
— Josh Butler, on why the camera is the sensorA note on irony: the workers' comp industry spent decades perfecting how to price an injury. CompScience is betting the bigger market is in the injuries that never happen.
Josh Butler did not arrive from the insurance world. He spent roughly fifteen years building AI products - stints at Facebook and Microsoft, then running the self-driving car platform product team at NIO, where the entire job was teaching machines to perceive a chaotic physical world and react before something went wrong.
The pivot was personal. After a family member was injured on a construction site, Butler started asking why the perception technology pointed at highways could not be pointed at the places people actually get hurt: docks, factory floors, job sites. The cars had billions in funding to see a pedestrian. The warehouse had a clipboard.
He founded CompScience in 2019 and made a bet that most software companies would not: that selling the AI alone was not enough. To change behavior, you had to change the economics. So CompScience became part technology company, part insurance company - a Managing General Agent that underwrites workers' comp coverage and bundles its safety AI right into the policy.
"We reward customers who are ready to lean in on protecting their workers, and we invest an order of magnitude more in their success."
— Josh Butler, on the modelIt is a deceptively radical structure. A normal SaaS vendor sells you a dashboard and walks away. CompScience has skin in the game: if its AI fails to prevent injuries, it is the one writing the claim checks. The product and the incentive point the same direction.
The core of CompScience is a computer-vision and generative-AI system trained on more than two million hours of video and ten billion images. It plugs into existing security and CCTV feeds - no new hardware, no rip-and-replace - and watches for the fifty-plus behavioral and environmental hazards that precede most injuries: slip and fall risks, unsafe equipment operation, missing PPE, the dangerous corner that keeps generating near-misses.
Runs on existing cameras to detect 50+ hazards in real time. The engine behind the dashboards, hotspot maps, and alerts.
An "active" workers' comp product that bundles AI safety analytics with coverage, and rewards prevention with better economics.
Predictive claims analytics aimed at cutting Total Cost of Risk by 20-30% before claims ever land.
A 2025 mobile tool that helps frontline crews build safe work plans and safety briefings on the spot.
Translation for skeptics: it is not facial recognition and it is not a productivity narc. The system scores movements and conditions against hazard patterns - the goal is the corner, not the person standing in it.
"Don't leave safety to chance. Leave it to science."
— The CompScience tagline, doing exactly what taglines shouldClaims of "AI safety" are cheap. Results are not. CompScience reports that companies using its platform have seen workplace incidents drop by as much as 35%, with one customer, Propak, citing a 73% reduction in risk. Across its book, the company points to more than 2 million hazards flagged and over $30 million in customer savings.
The customer logos help the story. CompScience names Honda, Toyota, DHL and Conagra among its users, alongside case studies for Shelby Erectors, Peerless Products and Keystone Natural Holdings. On the insurance side, it has lined up serious partners - Swiss Re for reinsurance capacity, plus carriers including Nationwide, AmTrust and The Hartford.
The honest caveat: these are the company's own figures, and a 73% case study is the highlight reel, not the median. But the carriers writing real capacity behind the product are voting with their balance sheets, which is a harder thing to fake.
"With this investment, we will make AI-powered safety available to millions of workers and save thousands of lives."
— Josh Butler, on the Series BMission statements usually float somewhere above accountability - vague enough that nobody can ever check the math. CompScience picked a different kind of goal: prevent one million workplace injuries by 2035. It is specific, it is dated, and it is the sort of thing you can be wrong about in public.
The company frames its longer ambition as building "the AI control system for risk" - a phrase that sounds grand until you remember it started with one forklift, one half-second, one corner that kept showing up in the data. The grand version and the small version are the same thing at different scales.
It is worth noting who else is in this race. Voxel, Intenseye, Protex AI and others are all building computer vision for workplace safety. CompScience's wager is that software alone changes dashboards, while software plus insurance changes behavior. The market will decide whether that bundle is genius or just complicated.
"There are huge opportunities to prevent injuries. That's a blue ocean opportunity to uplevel safety management."
— Josh ButlerReturn to that corner outside Atlanta. The forklift, the half-second, the worker who stepped clear. In the old world, that moment is gone the instant it ends - one of a million invisible near-misses that quietly accumulate until, one ordinary morning, the timing is a fraction worse and someone does not step clear.
In CompScience's world, that corner has a file. By the third flagged near-miss, the layout changes - a mirror goes up, traffic gets rerouted, a shift briefing gets a new line. The injury that the old world was patiently waiting to pay for never arrives. There is no claim, no clip of an ambulance, no story at all. Which is exactly the point: the best safety outcome is the one nobody ever notices.
That is a strange thing to build a company around - success measured in events that don't happen. But ten billion images and two million hours of video later, CompScience is betting that the camera in the corner was never just recording. It was waiting for someone to teach it what to look for.
"The camera was always watching. CompScience just gave it something useful to do."
— The whole idea, in one lineProfile compiled from public sources including CompScience press releases, PR Newswire, SiliconANGLE, Valor Equity Partners, Crunchbase and CB Insights. Figures such as customer outcomes, employee count, and revenue are company-reported or third-party estimates and should be treated as approximate.