There is a particular kind of optimism that drives most insurance AI projects: the belief that deploying a capable model against an existing data environment will produce reliable results quickly enough to validate the investment. This belief is understandable. It is also, in the majority of cases, incorrect. The results that come back are uneven, inconsistent, and ultimately insufficient to justify enterprise commitment.
Bruce Broussard, the leader behind Percipience, articulates why with a directness the AI market rarely applies to itself: AI does not fix data fragmentation. Without proper preparation, it amplifies the problem. When data is siloed, incomplete, or misaligned across systems, AI scales those inefficiencies rather than resolving them. The organizations that understand this are building the data foundation first. The ones that do not are accumulating proof-of-concept results that cannot translate to production.
What fragmented insurance data actually looks like
Almost every insurer’s data environment is the product of years of system accumulation: policy administration platforms that predate current integration standards, claims systems that developed their own data models, third-party feeds that were connected ad hoc, and homegrown applications that hold institutional logic no one fully documented. None of these systems were built to share a common data structure. The result is an environment where the same data element may be defined and stored differently across three systems.
Running AI against this environment does not produce a blended, useful output. It produces outputs that reflect whichever system’s version of reality the model encountered most heavily in training. For underwriting decisions, claims models, or pricing analytics where precision is directly tied to financial outcomes, this is not a theoretical concern. It is where implementation costs accumulate and where production models underperform relative to pilot results.
The strategic argument for data infrastructure
Broussard frames insurance data infrastructure not as a technical prerequisite but as a strategic foundation. Without a unified, governed data layer that creates a single source of truth, insurers cannot fully unlock the value of AI, analytics, or digital transformation initiatives. This reframing matters for how CIOs and CDOs build the internal case for what is otherwise a difficult budget conversation.
Data infrastructure investment is unglamorous and its returns are not immediately visible in the way a deployed AI model is. It does not generate demos. It does not show up in vendor announcements. But it is the work that determines whether every subsequent investment in analytics, AI, or digital capabilities delivers what the business case projected. Organizations that skip it are not moving faster; they are building on a foundation that will require expensive remediation once production systems reveal the gap.
Protecting investments from platform changes
One dimension of the data infrastructure problem that Broussard identifies and that CIOs should weigh carefully is investment durability. Insurance technology stacks change. Vendors are replaced, platforms are retired, and the AI products sold today may not be the ones the organization is running in four years. When analytics and machine learning investments are tightly coupled to a specific platform, they are exposed to obsolescence every time the underlying system changes.
A well-constructed enterprise data layer decouples analytical investments from the systems that feed them. The organization owns its data constructs independently of any vendor relationship. When platforms change, the data foundation persists and the analytical work built on it is preserved. This is a longer-term argument than most AI investment conversations engage, but it is the one that determines whether the organization’s data capability compounds over time or resets with each technology cycle.
A ReSource Pro perspective
The pattern across insurance organizations that have made durable progress on AI is consistent: they invested in data governance and infrastructure before, not after, they tried to scale AI applications. That investment is not just technical. It requires organizational discipline around data ownership, documentation of business rules that live only in people’s institutional memory, and the process structure to maintain data quality over time. Process and data management capability is the foundation on which AI value is built at scale.
Bruce Broussard: Why insurance data infrastructure must come before AI adoption
Author: Bruce Broussard
Publication: The Insurance Lead
Original Publication Date: June 17, 2026
