AI In P&C: Real vs. Hype
AI In P&C: Real vs. Hype
(Image source: CCC Information Services.)
The term AI has gone from being an infrequent suggestion of future possibilities at the fringe to being one of hottest buzzwords in the insurance industry. As with “configurable” or “cloud” before it, AI is likely to find its way into industry announcements whether it’s the most accurate description or not. This is nearly as true of life insurance as it is of property/casualty, but a recent announcement from CCC Information Services (Chicago) suggested the growing normality of AI in the P&C value chain. The vendor referenced 300 active, customer-specific AI models deployed. Does this mean AI has arrived in P&C insurance? We asked industry analysts for their take.
It would be reasonable to assume, based on the number of soft industry pronouncements these past couple of years about the potential of AI for this or that insurance application that real, scalable and multi-purpose AI solutions are still more aspirational than real, acknowledges Stephen Applebaum, Managing Partner Insurance Solutions Group (Chicago). However, he insists, nothing could be further from the truth. “AI is real, it’s in use and its adoption across the insurance enterprise is impressive,” he says.
We’ve been hearing about the promise of AI for insurance pain points for so long we could be forgiven for being surprised when faced with the actual extent of adoption, Applebaum continues. He acknowledges being surprised by CCC’s announcement but says, “This is AI at scale, customized to meet the objectives of individual insurers.”
What has enabled this seemingly sudden adoption of AI is the rapid rate of digitization of the data needed to train and continuously improve the underlying algorithms. The realization of that prerequisite has enabled a range and variety of use cases may surprise those who are primarily focused narrowly on one operational area, Applebaum says. Those include:
More specifically to auto insurance—the P&C industry segment CCC serves—uses cases include the following, according to Applebaum:
- auto and property claims ranging from touchless automated damage estimating to workflow distribution;
- auto and property Underwriting including risk assessment, risk avoidance and premium pricing;
- auto casualty claims including injury causation, fraud deterrence and medical cost management;
- digital payments to claimants, policyholders and vendors;
- end-to-end Total Loss claims management and cycle time compression.
Adoption of AI for these use cases is no longer limited to proofs of concept or pilots but also includes fully scaled up applications by very large and smaller insurers, Applebaum says. “AI has come of age in insurance—and it still has so much more to offer,” he concludes. “Little wonder that CCC’s investors announced last month that they value the company at $6.5 billion as they prepare to take it public in a SPAC transaction.”
AI in its various forms is absolutely becoming real in P&C, with many use cases and strong results, affirms Mark Breading, a partner at Boston-based analyst firm SMA. Chatbots and robotic process automation (RPA) are already spreading across the industry and improving operational efficiencies, he says. Even more sophisticated AI technologies—such as computer visioning, machine, learning, and natural language processing—are already providing high value in underwriting and claims, he reports.
While the digitization of data is critical to the rise of AI, Breading sees a broader menu of enabling technologies starting with raw increases in computer power. “The whole AI world has experienced a great leap forward over the last 5-10 year, with massive computing power, big data capabilities, the evolution of cloud computing, and advances in algorithms,” he elaborates. “During this time there has also been a new wave of companies and also significant numbers of talented individuals with data science and AI skills.”
There are two major categories of those talented individuals—specialists and generalist, according to Martin Ellingsworth, an analyst with Celent (Boston). “Specialists—at vendors mostly—are going big and at scale with the ‘art of the possible’ and productizing good use cases, then selling to insurers trying to realize impactful business value,” he says. “Generalists—who are mostly internal insurer staff—are doing test-and-learn sandbox pilots around proving good use cases and then failing to reach implementation. Lacking focus on a use case, technologists in academia and the insurance industry end up with more research than results.”
Insurers’ Struggle to Get Results from AI
Internal analytic generalist teams scattered across multiple business units and IT use data scientists more casually, focusing on executive wish-list problems, Ellingsworth laments. “Fielding dozens of pilots across the enterprise on data they scrape from multiple internal silos and systems. They ‘fail to Scale’ at insurers due mostly to inertia, lack of support, and flat footedness. Oftentimes, the MLOps/AIOps and AI Lifecycle management governance roles do not exist in their IT organizations to sustain the AI models they can create into a production setting. If you go to the LinkedIn job boards, you can literally see this gap now being addressed as these types of job titles are the most hotly recruited outsiders for large insurers.”
The good news, according to Ellingsworth, is that vendors, for whom viable use cases are a life-or-death matter, are productizing solutions. “They are bringing Data, AI, and Cloud to the party,” Ellingsworth quips. “Furthermore, they can learn to make their solutions better, faster, cheaper, and more satisfying while operating in a ‘test, learn, do’ loop, so their business improves.”
Nor is insurers’ investment in AI for naught, Ellingsworth suggests. “It seem like dabbling in data science is not producing impact, but in most cases it serves to move the bigger obstacles out of the way for an ‘adopt and adapt’ acceleration of customizable vendor solutions,” he says.
A vendor/insurer dichotomy is essential for understanding both how far AI has come and how far it still has to go, suggests Jeff Goldberg, EVP, Research & Consulting, Novarica (Boston). Insurers may still be in relatively early stages in achieving AI expertise, but the vendors represent the vanguard.
“I think more AI is being utilized by the insurance industry than the insurers themselves realize, because the vendors are leveraging the technology,” Goldberg comments. “The vendors are taking advantage of AIU in both direct and indirect ways to influence the solutions they provide. Insurers thus benefit even if they’re not investing in AI themselves.
Two Ways to Invest in AI
There are two ways for insurers to invest in AI, Goldberg says: the first is to adopt general purpose AI platforms—such as Azure Cognitive, Google AI or IBM Watson. “To use these, you need your own AI experts on staff doing their own modeling and operationalizing the models,” he explains. “Only the bigger carriers have the resources to invest in the necessary talent and technology.”
The other way to invest—knowingly or otherwise—is to adopt purpose-built solutions where vendors have applied AI to industry problems. As examples, Goldberg cites applications that use aerial or satellite imagery to estimate roof condition, or submitting business and receiving a real0tie risk score from a predictive model powered by AI. In the later case, Goldberg observes, “From the insurer’s perspective, it’s just a score—a number to help underwriters make decisions. However, AI is directly helping them, even though they are not directly investing in AI.”
Investing in Business Value
It is by no means a problem that insurers aren’t chasing technology, in this case AI, Goldberg implies. “The reasons some insurers don’t realize the benefit is that they don’t care about AI—they care about business value,” he stresses. “What they’re investing in is better decision making, better operational processes or better customer engagement.”
The Hartford Adopts Tractable for AI-Driven Claims Processing