Source: AI as a “Safety Net” for Claims Adjusters: Inside CLARA Analytics with CEO Heather Wilson – Transcript
TLDR:
- CLARA frames its AI as a “safety net” that flags risk on a claim while the adjuster stays the decision-maker, and that framing is what wins over cautious carriers.
- The models read adjuster notes and documents, not just form fields, to reconstruct 80 to 85 percent of a claim’s story and refresh it daily.
- Acting on those alerts produced real money: 2 to 3 percent loss cost savings within six to nine months, and reserves cut by millions on over-reserved files.
CLARA Analytics is a claims AI platform whose models have learned from roughly 10 million cases over the past decade. CEO Heather Wilson, a former chief data officer at a tier-one carrier, sat down to explain how the company gets risk-averse insurers to adopt AI without triggering the usual fears about replacement. Her core move is a reframe that any claims leader can borrow.
Why “safety net” beats “robot adjuster”
The fear around claims AI splits into two extremes: it either does almost nothing useful, or it takes everyone’s job. CLARA sidesteps both by positioning the model as a second set of eyes. Wilson compares it to a navigation app like Waze. You may already know the route, but the app still warns you about the crash ahead. The adjuster remains the driver and makes the final call on whether to reroute, change a provider, or raise a reserve.
Here is how that plays out. A worker files what looks like a minor finger cut, a severity-one case. As more medical documents arrive, CLARA’s alerts push the case to severity four. It turns out to be an amputation. A busy adjuster who had not opened the file in 90 days might miss that escalation. The model catches it, flags the reserve gap, and surfaces possible litigation exposure. That is the safety net in action.
The data problem nobody talks about
Most carriers assume their structured data is the foundation. Wilson says the opposite. The fields adjusters type into often have single-digit fill rates because of basic human behavior. So CLARA built its models to lean on two richer sources instead: the adjuster notes, which she says read like a novel about the case, and the documents themselves, meaning medical records, police reports, and legal filings. Those two sources reconstruct 80 to 85 percent of the story and keep updating it daily.
Getting there requires what Wilson calls data engineering as a service. CLARA refuses to start until the right data is flowing, even if that delays go-live by six weeks. The payoff is that carriers do not have to fix their own data first.
How they escape pilot purgatory
Insurance is full of pilots that go nowhere. CLARA’s answer was to find an early non-pilot client and deliver measurable outcomes within three months. Just as important, the team removed integration pain by acting as a data orchestration layer. As Wilson puts it, the data can sit in 10 different places and CLARA will hook up its own pipes and pull it, so they rarely need to involve IT. Adoption then sticks because the alerts live inside the existing claims workflow, and acting on them rolls up to an aggregated financial outcome that gets the CEO demanding adoption.
The Bottom Line
Sell the safety net, not the robot. Carriers buy claims AI when it augments a senior adjuster’s judgment, plugs into existing workflows, and proves loss-cost savings inside two quarters. Start with the messy notes and documents you already have, secure executive sponsorship early, and let the financial results pull adoption through the organization.