Ask an insurance executive whether their company is “doing AI” in claims or underwriting, and the answer is almost always yes. Ask whether it’s actually moving loss ratios, cycle times, or combined ratios, and the answer gets quieter.
That’s not a coincidence. Adoption and impact are two different metrics, and insurance is much further along on the first than the second. In underwriting specifically, AI adoption is projected to jump from 14% today to 70% by 2028 — which means most of that growth is still ahead of most carriers, not behind them.
The insurers pulling ahead aren’t running fundamentally different AI. They’re running the same models on top of a fundamentally different engineering foundation.
What is AI claims and underwriting automation?
AI claims and underwriting automation applies machine learning and generative AI to the document-heavy, judgment-heavy work inside claims and underwriting: reading policy files and loss runs, extracting data from unstructured documents, flagging risk, routing claims to the right adjuster, and surfacing recommendations an underwriter can approve or override. Done well, it doesn’t remove people from the workflow — it gets better information in front of them, faster.
Why are most insurers still stuck in pilot mode?
The reason has less to do with the technology than most executives expect. BCG’s research on the sector breaks the obstacles to scaling AI in insurance into two buckets, and the split is lopsided: 70% are organizational and people problems — unclear ownership, siloed operations, cultural resistance — and only 20% are genuine technology barriers. AM Best’s survey of more than 150 insurers and MGAs backs this up from another angle: 41% are already using AI in core functions, but fewer than 1 in 5 have reached advanced implementation.
Where the technology barriers do show up, they’re specific. Accenture’s underwriting research found the top two blockers are legacy system integration, cited by half of insurers surveyed, and data quality or accessibility, cited by 45% — not the AI models themselves.
What does it cost insurers to stay in pilot mode?
- Cycle times that don’t move. A pilot that never reaches production doesn’t change how long a claim or a policy actually takes to close.
- Compounding legacy debt. Every year automation doesn’t reach the policy admin system is another year of manual workarounds layered on top of it.
- Underwriting talent doing extraction work. Skilled underwriters spend their time reading and re-keying documents instead of assessing risk.
- A widening gap with carriers that did scale. Aviva has deployed more than 80 AI models across its claims operation and generated roughly $82M in savings in a single year. Lemonade settles straightforward claims in as little as 2 seconds and automates roughly half the claims it receives. That gap compounds every quarter a competitor stays in pilot mode.
What does engineered claims and underwriting automation actually look like?
- Document processing built for audit, not just accuracy. Getting a model to read a policy file or loss run is the easy part. Getting the extraction logged, validated, and traceable back to the source document for an examiner is what actually ships.
- Integration with the policy admin systems already in place. Most carriers aren’t rebuilding core systems to accommodate AI. The automation has to work with legacy platforms that were never built with modern APIs in mind.
- Human validation built into the workflow, not bolted onto it. The carriers seeing real cycle-time and loss-ratio improvements aren’t removing underwriters and adjusters from the loop — they’re using automation to get better information in front of them, faster.
- Data-handling controls that hold up under scrutiny. Claims data includes medical records, financial details, and other sensitive information. Where it’s processed, where it’s stored, and who can access it need clear, defensible answers before an automation project goes live, not after.
What does this look like in practice?
A recent Gorilla Logic engagement shows what this looks like end to end. An insurance carrier’s requirements documentation for carrier onboarding was taking one to three weeks of manual work per submission — a slow, error-prone process that delayed new business. Gorilla Logic built a document-extraction and human-validation workflow that read incoming submissions, extracted the required data, and routed it for underwriter review. The result: what took one to three weeks now takes minutes, a 95% reduction in processing time, without changing the underlying approval process the carrier’s compliance team already trusted.
A claims and underwriting AI-readiness checklist
Five questions reveal how close a claims or underwriting AI initiative is to production-ready:
- Can we trace every AI-assisted claims or underwriting decision back to a source document an examiner could review?
- Does our automation integrate with the policy admin system we actually run, or would it require replacing that system first?
- Are underwriters and adjusters still the ones approving decisions, or is the system deciding unsupervised?
- Do we know exactly where claims and policy data is processed and stored, and who can access it?
- Could our compliance team explain, today, how a specific AI-assisted claim decision was reached?
If more than two answers are “no,” the constraint isn’t AI strategy. It’s the engineering underneath it.
Questions insurance leaders ask about claims and underwriting AI
Does AI replace adjusters and underwriters?
No. The insurers seeing measurable results are using AI to surface information and recommendations faster, with adjusters and underwriters still making the final call on anything that matters. AI removes the manual document work, not the judgment.
How long does it take to move claims automation from pilot to production?
It depends more on integration complexity than model selection. Carriers that treat it as an integration and governance project, not an AI experiment, typically move from pilot to a production workflow in a matter of months rather than years.
What’s the biggest technical blocker to scaling claims AI?
Legacy system integration and data quality, consistently, across every major industry survey on the topic. The model is rarely the limiting factor.
Can AI help with compliance and audit requirements, not just efficiency?
Yes. Document processing that logs and traces every extraction back to its source gives compliance teams a stronger audit trail than most manual processes ever produced.
Your claims and underwriting AI initiative doesn’t need a better model.
