How to Build an Insurance Claims App with AI
Contents
Last updated: August 2026.
Insurance is regulated state by state in the US, not federally, and Colorado is the state that matters most right now for anything AI touches in claims, underwriting, or pricing. Auto and health insurers there just crossed a real compliance deadline, annual reports became mandatory on July 1, 2026, and the broader legal picture shifted twice in the months before that.
Colorado's SB21-169 is the law to build around
SB21-169 prohibits insurers from using external consumer data and information sources (ECDIS), and any algorithm or predictive model built on that data, in ways that produce unfair discrimination based on protected characteristics: race, color, national origin, religion, sex, sexual orientation, disability, and gender identity, among others. Life insurers went first, filing initial attestations in December 2024 and a second round in December 2025. Auto and health insurers followed, with their own annual compliance reporting requirement starting July 1, 2026. "Algorithm" here isn't limited to a flashy AI model, it covers any computational process, including a claims-scoring tool or a fraud-flagging system, that materially influences a decision about coverage, pricing, claims, underwriting, or policy servicing.
The broader Colorado AI law got repealed before it ever took effect
This is worth knowing precisely because it catches guides written even a few months earlier off guard: Colorado's general AI Act (SB24-205), signed in 2024 with an original effective date of February 2026, was delayed once to June 30, 2026, and then repealed and replaced entirely by SB26-189 in May 2026, before SB24-205 ever actually took effect. SB26-189 takes effect January 1, 2027, and requires deployers to notify consumers when covered automated decision-making technology influences a consequential decision, explain adverse outcomes within 30 days, and offer a meaningful human review option. If a piece of content or a vendor contract still references SB24-205 as the applicable framework, it's describing a law that no longer exists.
Build one governance program, not three
Insurers already compliant with SB21-169 are generally deemed compliant with SB26-189 in the practice of insurance specifically, and the same governance work (bias testing methodology, vendor oversight, incident management, documentation) covers most of what the NAIC's own Model Bulletin on AI Systems expects too. The practical approach is building one serious program against SB21-169's bar and treating the other regimes' requirements as incremental additions, a consumer notice document here, an AI Systems Program document there, rather than three separate compliance builds running in parallel. None of this program is something an app builder produces, Huxly generates the claims app's interface and connects a backend, the governance documentation is separate work with a compliance advisor regardless.
The technical side, once governance is sorted
Photo-based damage assessment (computer vision estimating repair cost from a few submitted photos) and telematics-driven auto pricing are the two AI features doing the most real work in claims apps right now, alongside fraud-flagging models that route suspicious claims to human review rather than auto-denying them outright. Every one of these counts as exactly the kind of algorithm SB21-169 and its counterparts are built to scrutinize, which is the reason the governance work has to come before, not after, these features ship. The claims-submission screens themselves, photo upload, status tracking, messaging with an adjuster, are the part a tool like Huxly generates quickly as a native app; the scoring and fraud-flagging logic behind those screens is what the governance program has to cover.
Fintech and healthcare development run into the same regulatory-first requirement, just with insurance regulators and state insurance departments standing in for banking and health regulators, see our guides on fintech and healthcare apps for how that plays out there. For the broader cost picture, see our breakdown of what it costs to build a mobile app in 2026.
FAQ
Is insurance AI regulated at the federal level in the US?
No, insurance regulation happens state by state. The NAIC coordinates a model bulletin that many states reference, but Colorado's own law (SB21-169) currently goes further than that model in requiring quantitative bias testing and formal attestation.
What is ECDIS, and why does it matter for a claims app?
External Consumer Data and Information Sources, any data not provided directly by the consumer, credit-based data, social data, and similar. If your app's algorithm uses ECDIS to influence coverage, pricing, claims, or underwriting decisions, Colorado's rules likely apply.
Is Colorado's AI Act (SB24-205) still the law I need to comply with?
No. It was repealed and replaced by SB26-189 in May 2026 before it ever took effect. Any reference to SB24-205 as the current framework is describing a law that no longer exists.
When do auto and health insurers in Colorado need to start reporting compliance?
Annual compliance reports became mandatory on July 1, 2026, following life insurers, who began filing attestations in December 2024.
Does complying with SB21-169 also satisfy other AI insurance regulations?
Largely, yes. Insurers compliant with SB21-169 are generally deemed compliant with SB26-189 in the practice of insurance, and the same governance work covers most of the NAIC Model Bulletin's expectations too.
Can Huxly help with the bias testing or governance documentation SB21-169 requires?
No. That work sits with a compliance advisor or the insurer's own governance team. Huxly's role is generating the claims app itself, the photo upload, status tracking, and messaging screens, not the algorithmic governance program behind the scoring or fraud-detection features.
What AI features are most common in claims apps right now?
Photo-based damage assessment that estimates repair costs from submitted images, telematics-based auto pricing, and fraud-detection models that flag suspicious claims for human review rather than auto-denying them.
Does a simple fraud-flagging feature count as a regulated algorithm?
Likely, yes. The definition covers any computational process that materially influences a coverage, pricing, claims, or underwriting decision, which includes a fraud-flagging tool even if it doesn't make the final call itself.
