If you are serious about AI in auto insurance, you need telematics data

This week, TPG Global’s Rise Fund and Allianz X led a $350 million strategic investment into Cambridge Mobile Telematics. The market is sending a clear message. Insurers that want to compete on AI need to start with the data layer.

The AI flywheel has a fuel problem

Across the insurance industry, AI has become a strategic priority. Carriers are investing in machine learning models for risk pricing, computer vision for claims automation, and large language models for customer engagement. Boards are asking about AI strategies. Technology budgets are growing. The ambition is real.

But there is a problem that is often overlooked: AI models are only as good as the data that trains them. This was highlighted by Ericsson Chan, Group CIO at Zurich at Insurtech Insights in London last week: data quality is a key part of the tech foundation of AI success factors.

And in auto insurance, the most predictive, most granular, and most actionable data available is driving behaviour data. Without it, an insurer’s AI grand plans are built on a foundation that cannot differentiate.

This is not a technical observation. It is a commercial one. The insurers pulling ahead on loss ratio, pricing precision, and customer retention are not simply the ones deploying the best models. They are the ones with the best data feeding those models. And the gap between them and their competitors is widening every year.

Why driving data is different

Traditional insurance data tells you who the driver is. Telematics data tells you how they actually drive. That distinction is fundamental.

Demographic and historical data points (age, vehicle type, claims history, postcode) correlate with risk but do not cause it. They are proxies of risk. Driving behaviour data has direct causality on risks. Hard braking, phone distraction, late-night driving, speed relative to context: these are the factors that provoke accidents. They are measurable, continuous, and extraordinarily predictive.

When you feed this data into an AI model, the model can do things that were simply not possible before. It can identify high risk behaviours before a claim occurs. It can segment a book of business with a precision that static rating factors cannot match. It can detect patterns across millions of trips that no actuary could find manually. It can personalise engagement in ways that demonstrably change driver behaviour over time.

According to our latest UBI Global Study, the most comprehensive analysis of the global connected insurance market, UBI has now reached over 60 million policyholders across 62 countries, doubling in just 5 years. The carriers driving that growth are not doing so by accident. They have recognised that telematics data is not a UBI product feature. It is the raw material for a superior AI-driven insurance model.

The Flywheel Effect

The reason telematics data is so strategically valuable is not just what it enables today. It is what it enables over time.

Each telematics programme generates data. That data trains better models. Better models produce more accurate pricing, lower loss ratios, and more personalised engagement. More personalised engagement attracts and retains better risks. Better risks generate more data. The flywheel accelerates.

Insurers who started building this flywheel 5 years ago are now in a fundamentally different competitive position to those who did not. They have proprietary datasets that cannot be bought or borrowed. Their models improve continuously. Their pricing gets more accurate with every policy written and every trip recorded.

This is what makes telematics data a strategic asset rather than a product feature. It is not just useful today. It compounds in value over time. And critically, it cannot be retrofitted. You cannot go back and collect 5 years of driving data you chose not to gather. Every year an insurer delays is a year of data – and a year of model training – they cannot recover.

TPG‘s decision to back Cambridge Mobile Telematics through its Rise Fund, its impact investing platform, is a reflection of exactly this dynamic. The capital is not flowing into AI as a standalone bet. It is flowing into the combination of AI and proprietary driving data at scale, accumulated over more than a decade and across tens of millions of drivers. That distinction is important.

AI without telematics is generic

The practical consequence of this is straightforward. In underwriting, an insurer that invests in AI without a telematics strategy is working with the same data inputs as every other carrier in the market. Census data. Bureau scores. Claims histories. This information is widely available, widely used, and increasingly commoditised. The AI models built on it converge to similar outputs. There is no durable competitive advantage to be found there.

Contrast that with an insurer running a mature telematics programme. Its AI models are trained on proprietary, continuously updated, behaviour-level data that no competitor can replicate. Its pricing reflects actual driving risk, not demographic proxies. Its engagement tools are informed by individual driver patterns. Its claims systems are fed by real-time crash data. Each of these capabilities reinforces the others.

The market leaders in connected insurance have understood this for years. The question now is whether the followers will act before the gap becomes insurmountable.

The barrier is no longer technical

One of the most important developments of the last decade in this market is the dramatic reduction in the cost and complexity of telematics deployment. Smartphone-based programmes offer ways to address the market without expensive hardware. Mobile SDKs can be integrated into existing insurer apps at relatively modest cost and at speed. OEM-connected vehicle data is increasingly accessible through standardised APIs.

The barrier to launching a telematics programme today is not technical. It is not financial. It is strategic will. Carriers that have been waiting for telematics to become easier have largely run out of reasons to wait. The infrastructure exists. The consumer acceptance is there. Our UBI Global Study shows that when insurers offer well-designed telematics programmes, customers prefer them to flat-rate alternatives. The demand is not the constraint.

What remains is a decision about whether AI-driven underwriting is a priority or a talking point. For carriers that are serious about the former, building the telematics data layer is not optional. It is the prerequisite.

Our outlook

PTOLEMUS has tracked the connected insurance market since its inception. Our UBI Global Study predicted in 2015 that telematics would uberise the auto insurance market. With over 450 active programmes and more than 60 million policyholders across 62 countries, that prediction has been borne out. The $350 million investment into Cambridge Mobile Telematics announced this week, backed by TPG, Allianz X, and State Farm, is consistent with the trajectory we have observed across markets for years: institutional confidence in telematics as foundational AI infrastructure is growing, not retreating.

We believe that the next phase of competition in auto insurance will be won by the carriers that combine mature telematics programmes with sophisticated AI capabilities. These are not two separate strategies. They are one strategy, and the data layer has to come first.

Insurers that treat UBI as a niche product rather than a data infrastructure decision will find that their AI ambitions outpace their data reality. The flywheel only works if you start spinning it.

If you wish to know more about the synergies between generative AI and UBI, our latest UBI Global Study includes 30 pages of analysis and case studies on the subject.

To obtain more information, please contact Frederic Bruneteau.

Article written by Frederic Bruneteau and Alex Tallon, under PTOLEMUS copyright