How an insurer turns claims data into revenue

A commercial insurer that has settled fifteen years of claims understands risk in its segments better than most underwriting models do. The understanding is real, it was expensive to acquire, and it is currently being applied to exactly one purpose.

Consider what that insurer actually holds: policy administration, claims, risk and inspection, billing, and third-party enrichment. Each system is complete on its own and each answers its own questions well. A person who knows the claims system thoroughly can tell you almost anything about a claim.

They cannot tell you the interesting thing, because the interesting thing is not in the claims system.

What becomes visible when five systems resolve into one

Resolve those five sources into one structure, with entities matched and relationships typed, and several findings surface that no single system holds.

Claims severity turns out to correlate with characteristics of the insured site that underwriting captures inconsistently and inspection captures well. Loss patterns cluster around specific repair and remediation networks in ways neither the claims team nor the underwriting team can see, because each sees only its own half of the picture. Certain classes of business generate early operational signals, in first notice of loss language and in inspection scheduling, that precede a severity outcome by months.

And fifteen years of settled claims constitute a body of evidence about how particular risks actually behave, as distinct from how the rating model assumes they behave.

A catalog would have reported that the estate holds a claims table with forty columns. That is metadata. The clustering is knowledge, and it only becomes visible once the entities are resolved and the relationships are typed.

Six candidates, generated in days rather than a quarter

From those findings the assessment produces candidates rather than a recommendation, and in this case it produces six.

A risk benchmarking service for brokers, and a loss prevention advisory offering for large insureds priced separately from the policy. Segment-level portfolio intelligence for reinsurers and capacity providers, alongside a feasibility position on the usage-based and parametric structures the historical evidence would actually support. An underwriting appetite service for the distribution network. And a repricing case for the insurer’s own book.

Each one carries the strength of the evidence behind it, stated per claim rather than in general. That distinction matters commercially and not just technically. A claim resting on four corroborating sources is a different proposition from one resting on a single unconfirmed field. An offering built on the first is defensible. An offering built on the second fails on contact with its first serious customer.

What the buyers actually said

The insurer took them out.

Brokers found the benchmarking service interesting and would not pay for it, because a competitor already bundles something adequate. Large insureds said the loss prevention offering was worth a great deal, provided it came with an engineer rather than a report. Reinsurers did not engage at all.

That is four candidates resolved in a few weeks of conversations, three of them retired.

The retirements are the point. Under the old arithmetic, where establishing whether one offering was even possible took an analyst team most of a quarter, an organization evaluates two or three ideas, picks one, and commits. The chosen idea then has to work, which makes everyone cautious about picking it and reluctant to abandon it afterwards. Small number of large bets, taken slowly, defended after the fact.

When a credible candidate costs days, the strategy inverts. Twelve candidates evaluated costs less than one built on a hunch.

The one that never needed a customer conversation

Note what happened to the sixth candidate.

The repricing case never went to a buyer, because there was no buyer. It is an internal action, and it flows into the operating model as a record with an owner, an expected outcome, and a horizon. Organizations look outward first when they think about data monetization. The internal answer frequently wins, and it is often the largest number on the list.

The most commercially significant thing an insurer can do with fifteen years of claims evidence may not be to sell an insight product at all. It may be to price its own book differently.

Where the boundary sits

The product here is derived understanding: patterns across many events, commercially valuable precisely because they generalize. What an insurer knows about how a category of risk behaves is a different asset from the file of the customer who generated a claim, and only the first is in scope.

A candidate that depends on identifying individuals to an external party is the wrong candidate rather than a problem to be engineered around. Permitted use gets evaluated as part of the assessment, with the terms under which data was collected treated as a constraint on what can be built from it. Candidates that cannot clear permitted use get retired early, before a business case has been written and a partner approached, rather than late as a legal objection.

For the two that survived, delivery is where most initiatives fail, and the failure is always the same one. A product delivered as a periodic export begins aging the moment it is produced, the recipient acts on stale figures, and the provider gets blamed for a decision made against data it delivered accurately three weeks earlier. A product built on a resolved foundation is a governed, versioned, continuously current view instead, with lineage on every figure, a confidence rating that moves when the underlying evidence weakens, and every version reconstructable so a dispute nine months later is settled by showing the state the recipient actually saw.

Two offerings worth building because buyers said so, and four retired in a fortnight before anyone spent real money on them. That is what a good quarter looks like.

Twelve candidates, and what happened to them

The PolyPhaze white paper Twelve Ideas, Not One Big Bet sets out the four-question sequence, the four shapes an opportunity takes, and how a governed data product is delivered. Download the full insurance data monetization ebook for the worked example in full.