September 25, 2026 By: Swapnil Bidve
Why context, not models, will determine the next decade in property and casualty
Many insurance carriers are in the early stages of AI adoption and not yet seeing the ROI. Capgemini’s World Property and Casualty Insurance Report 2026, which surveyed more than 2,200 insurance employees, executives, and policyholders across 20 markets, found that 60% of carriers are still exploring AI or building proofs of concept, and that 42% aren’t even tracking AI results.
We sat down with Swapnil Bidve, senior vice president and head of the insurance and financial services practice at JK Tech, and Christina Lucas, Global Market Leader, Insurance, at Google Cloud to discuss why early returns for commercial carriers aren’t obvious. It’s not about the models, they say. Insurers already have the raw material they need, but not in a form AI systems can use. We asked them to explain what that means in practice.
Carriers have spent the last decade or more replacing core systems. How have those investments reshaped the insurance business?
Swapnil Bidve: Honestly, not in the way people expected. For the last ten or fifteen years, every company has worked on modernizing its systems, including policy administration, claims, and end-to-end operations. But as carriers are discovering, just implementing that operational layer doesn’t solve every business challenge.
Insurers want their combined ratio as low as possible. One big hurricane in Florida or a wildfire in California can eat a lot of your top line and your bottom line. The insurance workforce is aging, but the next generation is moving into AI and financial modeling. Meanwhile, customers want easy online experiences they can access on their phones, yet traditional companies still ask them to fill out a hundred questions on a form.
These challenges aren’t about the operational layer.
Carriers are now investing in AI to solve business problems like these. How is that going?
Christina Lucas: There’s huge progress, with many new paid and open-source models coming to market and real advances in reasoning. But progress is still too limited because the information those models can access is too limited.
Out-of-the-box models get answers by scanning the internet, which they’re able to do because publicly available websites are connected. Now imagine replacing public webpages with private data unique to your company. This gets complicated because AI doesn’t understand how you built your data over time. It might be in a database, in a document, on somebody’s laptop, or in somebody’s head. That’s how most companies are organized. Disconnected data becomes the bottleneck when you try to do anything with AI at an enterprise level.
Insurance has an even bigger version of this problem. It’s a two-hundred-year-old industry that’s been tracking claims data, continuously monitoring weather data, and buying data from third-party providers. The data just keeps coming. For AI to do its magic, you need to make all this data readable for any agent or system you build.
So, you’re saying that to get real results from AI, insurers need to get their data ready first?
Swapnil Bidve: Correct. You can have the most powerful model in the world, but if you don’t give the model the right context, you won’t get meaningful insights or results.
Everything your company knows about a single policyholder is spread across database tables, documents, chat histories, emails, or notes somebody typed after a site visit. These data sources are disconnected and expressed differently. You need to bring them into a standard format so AI has a full picture and context.
First, figure out where that single policyholder appears across all your data sources. Then you create one node for the policyholder and connect it to all relevant data. This is a knowledge graph, a kind of graphical representation of your data, which serves as a map for AI to follow. Building that map means creating a semantic layer and a unified ontology, connecting every data source, business rule, column name, and table name as far as you can.
Problems like reducing your catastrophe (CAT) exposure and improving your combined ratio have always been answerable in principle, because the answers are sitting in data you already own. You just need to build a map so AI can find all the relevant data you have.
You’ve described the problem and the shape of the solution. Where do Google Cloud and JK Tech come in?
Swapnil Bidve: We have an agentic AI orchestration platform called JIVA. It connects your enterprise data, builds the knowledge graph I just described, and lets you use whichever AI model you prefer to complete tasks with that data. Ontologos is our enterprise ontology product, which is where the domain modeling lives. Both run on Google Cloud infrastructure.
Since insurance is a regulated industry, it’s an added benefit that Google Cloud provides security and governance controls you can point to when somebody asks how a decision was made. AI-derived answers or decisions must be explainable. A human in the loop should be able to govern, trace, and audit your data to avoid bias in any policy, claims, or pricing decision made using that data.
Everything an insurance carrier does is linked to financial decisions, so you need to be able to show your work.
Once a carrier has a data foundation, what becomes possible?
Christina Lucas: Now AI can be more than a bot automating a process. It can analyze, make recommendations, and prepare things for you.
Let’s revisit the challenges Swapnil listed earlier. Reducing your CAT exposure means analyzing fifty years of data, which no human can do today. Feeding that data to a high-speed model returns valuable insights.
The same with underwriting. If somebody wants a policy on a forty-story building in New York, you have to analyze that building completely: the construction, what’s inside, how many exits, whether the contents are precious or ordinary. The better you understand what you’re taking on, the better you price it. AI can expedite analysis of every possible risk factor.
Now let’s take claims. Somebody’s car gets damaged in a storm, and they want it fixed. Behind the scenes, the insurance company is working through ten steps, which today takes two to five days. AI can quickly analyze the police report, video and images, policy coverage, repair estimates, and fraud risk. Within two days, the insurer can then tell the customer, “Everything is taken care of. Go to this dealer, and they’ll fix your car.”
Swapnil Bidve: There’s also compliance. Insurers report monthly and quarterly to the states, the Fed, and many other agencies. It never stops. AI can prepare the reports that a person then validates, signs, and submits. This reduces the pressure from your labor shortage.
What about distribution? Once you’ve connected your data and run models on it, you get hundreds of suggestions to take to your brokers and agents: which risks match your appetite, when to start a renewal conversation, and where cross-selling makes sense. These recommendations come from actual buying behavior and history.
By 2035, we’ll use weather data to predict catastrophic events before losses occur. Claims will settle in minutes using real-time data. Policies will be personalized to each individual, and pricing will stem from individual risk factors instead of ZIP codes.
If you connect your data with the right insurance context, then the possibilities Christina and I just described—and many more—will open up for you.
