An assistant that guesses at coverage is a market conduct problem, not a productivity gain. Agnotic Technologies builds insurance assistants that read the claim file, the submission or the policy form, cite what they used and leave every coverage and claim decision with a licensed person.
Trusted by global innovators
Artificial intelligence assistants for insurers are AI tools that help adjusters, underwriters and policyholders with reading and drafting work, inside the controls insurance regulators expect. Agnotic Technologies builds three kinds: a claims file assistant, an underwriting submission assistant and a policyholder assistant grounded in the policy forms, each logged and reviewed by people.
Each one has a narrow job, a defined set of documents and a named person who reviews its output.
Reads the claim file so the adjuster starts from a summary, not a stack of PDFs.
Turns a broker email and its attachments into a structured submission.
Answers questions about the customer's own policy from the forms and endorsements on that policy.
The failures are specific to insurance, and most of them are compliance failures rather than technical ones.
The assistant answers from a sample form instead of the edition and endorsements on the customer's policy.
A summary drops the one document that changes the coverage position, and the adjuster never opens the file.
A friendly answer about what the policy covers reads like a coverage decision the insurer never made.
The assistant starts suggesting products or limits, which is advice a licensed producer must give.
Outbound reminder calls use a synthetic voice without the consent the phone rules require.
Nobody can show a regulator which model version wrote which draft, or who approved it.
We agree this table with your claims, underwriting and compliance leads before we build. Rules differ by state, so your compliance team confirms the last column for each state you write in.
| Task | May the assistant do it? | Who approves | What limits it |
|---|---|---|---|
| Summarize a claim file | Yes, as a draft linked to the source documents | The adjuster reads it before acting | Claims standards expect a reasonable investigation, so the adjuster still reviews the file |
| Draft a reservation of rights letter | Yes, as a draft only | The adjuster, and coverage counsel where your process requires it | Misstating policy provisions is an unfair claims practice, so a person checks every clause cited |
| Deny a claim or decide coverage | No | An adjuster with authority for that claim | Coverage and claim decisions stay with people under your claims authority rules |
| Extract a broker submission | Yes, with low-confidence fields marked | An underwriting assistant or underwriter confirms flagged fields | Your AI governance program, testing and documentation |
| Answer a policyholder's coverage question | Yes, by quoting the customer's own form, with a handoff option | Licensed staff take any follow-up that needs advice | Advice on the terms or benefits of a contract needs a licensed producer |
| Recommend a product or quote a premium | No | A licensed producer, with premium from the rating system | Selling, soliciting and negotiating insurance need a licence; premiums come from filed rates, not the model |
| Place outbound calls with a synthetic voice | Only with the consent the TCPA requires | Compliance sets the consent rules | AI-generated voices count as artificial voices under the TCPA |
Controls come before features. These three carry most of the weight.
Retrieval filters by policy number, form edition and endorsement, and the answer shows the passage it used. No passage means no answer.
Assistants write drafts and summaries. Sending a letter, paying, denying or quoting stays a human action in the core system.
Each assistant is listed with its purpose, data, tests, owner and model version, so the record exists before a regulator asks for it.
These are the parts we build for insurers. The general chatbot build process is described on our AI chatbots and assistants page.
Declarations, forms and endorsements indexed per policy with edition dates, so retrieval finds the right wording.
OCR and classification for the documents in a claim file, with each summary line linked to its page.
Fields from applications, schedules and loss runs with a confidence score, ready for intake rules.
Every answer shows its source passage, and the assistant declines when it cannot find one.
Requests for advice, quotes, complaints or claim decisions routed to a queue with the conversation attached.
Real files and questions with expected answers, rerun on every model or prompt change.
Prompt, retrieved sources, model version, output and reviewer stored for each interaction.
Consent checks before any outbound call or text, and no synthetic voice without it.
The path one question takes through the assistant. Every step is logged.
The customer signs in or is matched to an account. Without a verified policy, the assistant gives only general information.
The declarations, form editions and endorsements attached to that policy on the date in question are selected.
Search runs only over those documents, and the best matching passages are ranked.
The assistant quotes the passage. If the question asks for advice, a quote or a claim decision, it routes to licensed staff instead.
The exchange is stored with sources and model version, and your quality team reviews a sample on a schedule you set.
Chosen to run inside your cloud agreements and to leave an audit trail.
Measures we put on the assistant's dashboard from day one.
Answers that quote a source passage from the right policy.
Questions that needed advice, a quote or a decision.
How much adjusters change each draft before sending.
Test cases that fail after a model or prompt change.
Small first releases, each tied to one line of business and one review step.
Summaries for one line of business, read by adjusters before they act.
Reservation of rights and acknowledgement drafts from your approved templates.
Broker emails turned into structured submissions for one program.
Answers grounded in one product's forms, with handoff to licensed staff.
Structured loss details captured from email or chat for an adjuster to review.
Answers from your claims manual and authority guidelines, with citations.
The assistant reads from these systems and writes drafts back only where a person will review them.
We build secure, scalable products designed for privacy, interoperability, and regulatory readiness from day one across every sector we serve.
Implement lawful consent flows, data minimization, and secure processing for global data privacy.
Verified controls for security, availability, and confidentiality of enterprise data systems.
Adhering to the international gold standard for managing information security risks.
We combine deep technical expertise with industry-specific knowledge to deliver solutions that aren't just functional, but transformational.
We implement rigorous security protocols and compliance standards (HIPAA, GDPR, SOC2) across all industrial solutions to protect sensitive data.
Our architectures are built to handle massive data loads and user bases, keeping performance steady whether you're serving ten or ten million.
Using our suite of internal tools and proven frameworks, we cut setup work so your first release reaches users sooner.
Beyond simple wrappers, we build deep-learning integrations and predictive analytics directly into the core of your industry-specific workflows.
We start with one assistant on one line. We agree a fixed first release on the scoping call.
The task, the line of business, the documents and who reviews the output.
The assistant added to your AI inventory, with its test plan and owner.
Real files and questions with expected answers, agreed with your experts.
One assistant in production with human review on every output.
More lines or tasks once evaluation and review results hold.
Short facts we design around, each linked to its primary source. States adopt NAIC models individually, so your compliance team confirms what applies where you write.
The NAIC Model Bulletin on the Use of Artificial Intelligence Systems by Insurers defines an AI System as a machine-based system that can generate outputs such as predictions, recommendations or content, including text and sounds, and expects an insurer's AIS Program to cover AI Systems used for regulated insurance practices whether the insurer or a third-party vendor developed them.
Source: NAIC Model Bulletin, adopted 4 December 2023, sections 2 and 3 (1.8)The NAIC model bulletin says an insurer's AIS Program should include processes for telling impacted consumers that AI Systems are in use, with access to information suited to the stage of the insurance life cycle where the AI System is used.
Source: NAIC Model Bulletin, adopted 4 December 2023, section 3 (1.9)The NAIC Unfair Claims Settlement Practices Act lists knowingly misrepresenting relevant facts or policy provisions to claimants and insureds, and refusing to pay claims without a reasonable investigation, as unfair claims practices when committed flagrantly or often enough to be a general business practice.
Source: NAIC Model Law 900, Unfair Claims Settlement Practices Act, sections 3 and 4Under the NAIC Producer Licensing Model Act, a person may not sell, solicit or negotiate insurance without a licence for that line of authority, and negotiating includes offering advice directly to a purchaser about the substantive benefits, terms or conditions of a contract.
Source: NAIC Model Law 218, Producer Licensing Model Act, sections 2 and 3In Declaratory Ruling FCC 24-17, adopted 2 February 2024, the FCC confirmed that the TCPA's restrictions on calls using an artificial or prerecorded voice encompass current AI technologies that generate human voices.
Source: Federal Communications Commission, Declaratory Ruling FCC 24-17, CG Docket No. 23-362We don't just build products; we forge lasting partnerships. See how we've helped industry leaders transform their vision into technical reality.
“I can clearly see how Agnotic has a unique way of handling end-to-end development. They are always active on quick chat and provide support quickly.”
Aaron Phelan
Founder, Benchmark
“Agnotic combines deep technical expertise with strong domain knowledge. They understand the business context, anticipate challenges, and make collaboration smooth and effective.”
John Pasmore
Founder, Latimer
“Agnotic is the best technical team we evaluated. Their engineering excellence made our work dramatically easier and allowed us to stay focused on what matters most for maternal care outcomes. They took full ownership of the technical execution, and we are always happy to continue working together.”
Kim Smith
Founder, My Lauren
This page covers assistants for insurers. These pages cover the general AI work and the insurance systems they sit on.
Our general chatbot and assistant build process.
Extraction and classification for high-volume documents.
Voice agents for inbound calls, with consent and handoff rules.
The rating, forms and filings layer the assistants read from.
Our wider insurance practice for carriers, MGAs and startups.
Explore other production-grade engineering platforms we deliver across insurtech.
The technical patterns behind this platform translate naturally into adjacent verticals.
