The Insurance Data Readiness Checklist (Before You Buy Any AI)

Before you sign an AI contract, spend twenty minutes on this instead. It'll tell you more about whether the project will succeed than any vendor demo.

Roughly 80% of the effort in a successful insurance AI project goes into data, not models. So the useful question isn't "which AI should we buy?" It's "is our data ready for any AI at all?" That question is measurable. Here's how to measure it.

Score each item 0, 1, or 2. Be honest — the point is to find gaps, not to pass.

Section 1: Foundation

1. Single source of truth. Is there one place where policy, claims, and billing data live together and can be queried in one join?
0 = data lives only in source systems · 1 = a warehouse exists but is partial/stale · 2 = one governed platform, all core domains

2. Stable customer key. Can you identify the same person across every product line, reliably, today?
0 = no · 1 = partial matching, no durable ID · 2 = resolved identity with a persistent key

3. Legacy access. Can you get data out of the policy admin core without a manual extract?
0 = manual exports · 1 = scheduled batch · 2 = CDC / near-real-time

Section 2: Quality

4. Measured data quality. Do you have a per-field completeness and consistency score, tracked over time?
0 = never measured · 1 = ad hoc audits · 2 = continuously scored with thresholds

5. Structured where it matters. Are the fields your models need actually structured — or trapped in free text and PDFs?
0 = mostly free text · 1 = partially extracted · 2 = document pipeline producing structured fields

6. Deduplicated entities. Do duplicate customers, properties, and vehicles get resolved?
0 = duplicates everywhere · 1 = periodic cleanup · 2 = resolution in the pipeline

Section 3: Timeliness

7. Real-time where it counts. Are FNOL, fraud, and quote signals available in seconds — not overnight?
0 = all batch · 1 = some intraday · 2 = streaming for decay-sensitive signals

8. External data integrated. Are weather, telematics, property, and credit feeds joined to your records?
0 = none · 1 = a couple, manually · 2 = integrated and joinable

Section 4: Governance

9. Lineage. Can you trace any given number back through every transformation to its source record?
0 = no · 1 = documented informally · 2 = captured automatically end-to-end

10. Decision logging. If a regulator asks why a specific application was declined, can you produce the inputs, model version, and reasoning?
0 = no · 1 = partial logs · 2 = immutable per-decision record

11. Fairness sliceable. Can you slice outcomes by cohort to test for unfair discrimination?
0 = no · 1 = with heavy manual work · 2 = routinely, as a report

12. PII controlled. Is sensitive data tagged, access-controlled, and handled per HIPAA/GDPR?
0 = ad hoc · 1 = partial · 2 = enforced in the platform

Scoring

0–8 — Not ready. Any AI you buy now will stall. Not "underperform" — stall. Your money is better spent on the foundation for the next two quarters. The good news: this is fixable, and fixing it is a known quantity.

9–16 — Foundational. You can run narrow, well-scoped use cases (one line of business, one claim type) if you're disciplined about scope. Broad AI will disappoint. Target your lowest-scoring section.

17–20 — AI-ready. Your constraint is genuinely the use case and the model now, not the data. Go.

21–24 — Ahead of the market. Most carriers aren't here. You should be doing the agentic and real-time work your competitors are still writing slide decks about.

The pattern in the answers

When carriers run this honestly, two things usually happen. First, the score is lower than expected — often in the 6–10 range at organisations with ambitious AI roadmaps. Second, the low scores cluster in governance, because it's the section nobody funds until a regulator asks.

That second point matters more each quarter. With the NAIC Model Bulletin adopted across 23 US jurisdictions and the EU AI Act treating most insurance AI as high-risk, items 9–12 are moving from "good practice" to "prerequisite."

The point of this checklist isn't to be discouraging. It's the opposite: it turns "our AI project is mysteriously stuck" into a list of specific, ordinary engineering tasks. That's a much better problem to have.

We help insurers score honestly and close the gaps — the data foundations, pipelines, and governance that AI actually runs on. More at IntelliBooks.