Skip to content
Appsierra
Insurance · AI & LLM Engineering

AI Development for Insurance

By the Appsierra Quality Engineering Desk
Reviewed by senior engineers · Updated August 2026

AI development for insurance is the practice of building underwriting, pricing and claims AI that regulators and policyholders can be shown to be fair. It covers explainable decisioning, testing for proxy discrimination against protected classes, governance documentation aligned to the NAIC model bulletin and New York DFS Circular Letter 2024-7, and vendor-model oversight.

Part of Appsierra's Insurance & Insurtech engineering practice — see the full vertical overview.

Get a free QA audit →
AT A GLANCE
Industry
Insurance
Service
AI & LLM Engineering
Standards in scope
6
Questions answered
4
Updated
August 2026
A pod that already knows the constraint that changes the work in this sector.

Key Insurance testing & engineering challenges

Demonstrating that underwriting or pricing models do not proxy for protected classes
Explaining a declination or rating outcome to a policyholder in specific, accurate terms
Governing third-party and vendor models whose internals you cannot inspect
Validating external consumer data sources for accuracy and disparate impact before use
Keeping claims-triage automation from quietly disadvantaging a group through routing rules

Standards & regulations we test against

NAIC Model Bulletin on the Use of AI Systems by Insurers (2023)NY DFS Circular Letter 2024-7Colorado SB21-169Unfair trade practices / unfair discrimination lawEU AI ActNIST AI RMF

Key takeaways

Insurance AI regulation is state-led and already concrete — the NAIC bulletin and NY DFS Circular 2024-7 set real expectations.
You are accountable for vendor and third-party models too, including the data behind them.
External consumer data is where proxy-discrimination risk concentrates; it must be tested, not assumed clean.
Documentation is a required output: regulators expect to see how a system was tested and governed.

What do insurance regulators actually expect from AI systems?

The NAIC adopted its Model Bulletin on the Use of Artificial Intelligence Systems by Insurers in December 2023, and it has been the template for state adoption since. It applies across underwriting, rating, claims and marketing, and its central requirement is an AI governance framework ensuring systems are accurate, reliable, explainable and free from unfair discrimination.

New York went further in July 2024 with Circular Letter 2024-7, covering AI and external consumer data in underwriting and pricing. It expects insurers to demonstrate that systems do not proxy for protected classes or produce disproportionate adverse effects, to keep explanatory documentation, to allow regulator review of vendor tools, and to maintain internal oversight. These are engineering requirements as much as compliance ones.

How do you test an underwriting model for unfair discrimination?

The work centres on measurement rather than intention. Protected attributes are usually absent from the feature set, so testing focuses on whether outcomes differ across groups and whether permitted features act as proxies. That means estimating group membership carefully for testing purposes, measuring disparity in approval, rating and pricing outcomes, and probing which features drive any gap found.

External consumer data deserves particular scrutiny, because it is where disparity most often enters. A data source assembled for marketing may encode geographic or behavioural patterns that correlate strongly with protected characteristics. Testing it before adoption — and re-testing as it updates — is considerably cheaper than discovering the problem through a market conduct examination.

Who is accountable for a vendor's model?

The insurer is. Both the NAIC bulletin and the New York circular make clear that using a third-party model does not transfer responsibility for its behaviour, and New York expects regulators to be able to review vendor tools. This has a direct engineering consequence: you need contractual rights to documentation and testing access, and an internal capability to evaluate a model you did not build.

In practice that means treating vendor models as components you must independently validate — running your own disparity testing on their outputs against your book, monitoring their performance over time, and maintaining your own record of that oversight rather than relying on the vendor's assurances.

How does Appsierra help insurers build AI responsibly?

We pair AI engineers with evaluation engineers so that disparity testing, explainability and monitoring are built alongside the model, and the documentation regulators ask for exists as a by-product of development rather than a retrospective exercise.

We also perform independent evaluation of models an insurer has bought or built elsewhere, which is often the faster route to the oversight evidence the NAIC framework expects. We are an engineering and evaluation partner, not your compliance function or legal adviser.

Frequently asked questions

What is the NAIC model bulletin on AI?
It is the Model Bulletin on the Use of Artificial Intelligence Systems by Insurers, adopted by NAIC membership in December 2023 and used as a template by state regulators. It applies to AI used across underwriting, rating, claims and marketing, and requires insurers to maintain an AI governance framework ensuring systems are accurate, reliable, explainable and do not result in unfair discrimination.
What does NY DFS Circular Letter 2024-7 require?
Adopted in July 2024, it addresses AI systems and external consumer data sources used in insurance underwriting and pricing. Insurers are expected to demonstrate that these systems do not act as proxies for protected classes or produce disproportionate adverse effects, to maintain explanatory documentation, to permit Department review of vendor tools, to require vendor audits, and to keep internal oversight in place.
Are we responsible for a third-party AI model we licensed?
Yes. Regulatory guidance is consistent that accountability stays with the insurer, and New York expects vendor tools to be reviewable by the regulator. Practically this means negotiating contractual rights to documentation and testing access, running your own disparity and performance testing on the vendor's outputs against your own book, and keeping your own oversight record rather than relying on vendor assurances.
Can AI be used for claims decisions?
It is widely used for triage, fraud signals, document processing and adjuster support. Decisions that deny or reduce a claim attract the most scrutiny, and the governing expectations are the same as underwriting: explainability sufficient to give a specific reason, testing for disproportionate adverse effect across groups, human review where consequences are significant, and documented governance covering how the system was validated and is monitored.
No-risk start

Ship higher-quality insurance software, faster

Appsierra's expert-supervised AI & LLM engineering pods are productive in days and de-risked by our own evaluation platform — with senior accountability and a low-risk pilot. Tell us what you're building.

Get a free QA audit →
EXPLORE
Free ROI calculator What QA & dev cost Compare delivery models Hire a vetted pod Industries we serve
Vetted pods, productive in 7 days
Senior-reviewed pods · live in ~7 days · cancel anytime
Run the ROI numbers