Data Foundations: The Prerequisite for AI Expansion
In this article, we look at why the gap between AI ambition and AI adoption in insurance has little to do with the technology itself, and everything to do with the data underneath it. We set out a five-stage framework for placing your organisation's data foundations, and what it actually takes to move up a stage.
It's now well established that most insurance leaders believe AI is important for their business. But belief isn't the bottleneck. Capgemini's 2026 World Property & Casualty Insurance Report, based on interviews with 344 senior executives across 18 markets, found 60% of insurers are still stuck in the exploration or proof-of-concept stage, with 42% not even tracking whether their AI investment is working.
That gap has nothing to do with AI itself. It lives in the underlying data foundations. WTW's March 2026 Advanced Analytics & AI Survey found data quality and accessibility issues are the single biggest barrier to analytics adoption, cited by 42% of P&C insurers, ahead of every other obstacle in the survey.
The London Market tells the identical story from a different angle. The LMA's April 2026 AI survey, based on 39 responses representing over 60% of Lloyd's market stamp capacity, found adoption has more than doubled in twelve months, with 93% of firms now having, or building, formal AI governance frameworks. But deployment remains concentrated in operational efficiency rather than frontline underwriting or claims decisions. Ambition is not the constraint. Trust in the data is.
The London Market tells the identical story from a different angle. The LMA's April 2026 AI survey found adoption has more than doubled across Lloyd's in twelve months. But deployment remains concentrated in operational efficiency gains and cost-saving exercises. Only a small minority of managing agents have taken the further step of putting agentic or generative AI directly into live underwriting decisions. Ambition is not the constraint. Trust in the data is.
A framework to place yourself on
Rather than leave "data foundations" as an abstract phrase, it's more useful to think of it as a framework with five recognisable stages. Most organisations aren't uniformly at one stage, and different pillars of the business often sit at different points. But most leaders can place their organisation's overall AI-readiness quickly against these markers.
Foundational. Data quality and structure vary widely by team or system. No single source of truth for key risk or claims data. AI use, if any, is exploratory and disconnected from live decisions.
Emerging. Pockets of good practice exist, usually where a specific team has taken ownership. Data standards (Core Data Record, ACORD alignment) are understood but inconsistently applied. AI pilots exist but haven't scaled.
Developing. A data governance framework exists on paper and is partially operational — ownership is clearer but monitoring and quality checks are still manual or intermittent. AI is trusted for monitoring and drafting, not decisions.
Established. Data standards are consistently applied and actively monitored, not just documented. Governance covers model inventory and explainability, not just data quality. AI is contributing to some live, governed decisions in underwriting or claims.
Advanced. Data is a genuinely trusted, real-time asset across the business. AI operates within clear guardrails on live decisions, with human oversight calibrated to risk rather than applied uniformly. The organisation can absorb new data sources or use cases quickly because the foundation, not just the model, is built for it.
Most of the market sits somewhere between Foundational and Developing (consistent with Deloitte's 25% figure). Very few organisations have genuinely reached Established, and Advanced remains rare enough that it's currently a sizeable competitive advantage.
What moving up a stage requires
It's rarely a technology purchase. In practice, it comes down to a few specific, unglamorous actions.
An honest inventory before an ambitious roadmap. Most organisations can name their AI ambitions faster than they can describe the current state of their underlying data. The second is more useful to know first.
Ownership that sits with someone senior enough to move it. Data quality initiatives don't stall from lack of budget but from lack of a single accountable owner with the authority to prioritise it against other demands.
Standards alignment treated as infrastructure, not a project. Core Data Record and ACORD alignment isn't a one-off migration; it's an ongoing discipline, because new products, systems and data sources arrive continuously.
Governance built for AI specifically, not retrofitted from data governance. Model inventory, explainability and human-oversight thresholds are a different discipline from data quality monitoring, even though they depend on it.
Why this is worth doing before, not after, the next AI use case
It's tempting to treat data foundations as a prerequisite to finish "eventually," while the AI use cases get the visible investment. The Deloitte gap suggests that's exactly the trap 75% of the market is currently caught in. There's genuine recognition of AI's importance, without the underlying readiness to act on it credibly. The 30% of initiatives that never leave proof-of-concept are, in most cases, not failures of the model. They're failures of what the model was standing on.
We'll be exploring this exact framework, and how organisations move themselves up it, with a small group of senior data, underwriting and claims leaders next month. If placing your own organisation on this curve raised more questions than it answered, that's a great reason to join us and your industry peers. Get in touch to secure your place.