Corporate AI8 min read

How Can Insurance Underwriters Use AI?

On this page

Insurance underwriters can use AI to summarise long proposal forms and risk survey reports, build checklists of missing information, draft referral notes and explain terms to brokers and clients in plain words. AI prepares the paperwork faster. The underwriter still decides whether to accept a risk, on what terms and at what premium, and personal or medical details stay out of public AI tools.

TL;DR:

  • The best starting points are summaries, checklists and referral notes, where AI saves reading and writing time.
  • Give AI only masked documents: no names, ID numbers, KRA PINs, number plates or medical details in public tools.
  • AI never accepts, declines, loads or prices a risk. That judgement belongs to the underwriter and your authority limits.
  • Every summary is checked against the original document before anyone relies on it.
  • Health data is sensitive under Kenya's Data Protection Act, 2019, and section 46 limits who may process it.

Who this guide is for

This guide is for underwriters, underwriting assistants and underwriting managers at insurers and reinsurers in Kenya, across general, life and medical business. It covers six tasks with prompts you can copy, what to check, and the lines AI must not cross.

For the claims side, read how insurance companies in Kenya use AI for claims. For the sales side, see how insurance agents can use AI to serve clients. For the wider picture, start with AI for SACCOs and insurers in Kenya.

The rules that shape underwriting work

The regulator. The Insurance Regulatory Authority (IRA) supervises insurers and intermediaries under the Insurance Act, Cap. 487. On the IRA pages we checked on 15 November 2026, we found no rule written only about AI. Your existing duties still apply, and your compliance officer has the final word.

Cover starts with the premium. Section 156(1) of the Insurance Act says no insurer shall assume a risk in Kenya "unless and until the premium payable thereon is received by the insurer". So an AI-drafted email must never tell anyone they are covered. Cover confirmations come from your system and your authorised staff.

A draft to watch. In April 2026, Kenya's financial regulators, including the IRA, asked the public to comment on a draft Financial Consumer Protection Framework, dated March 2026. As of 15 November 2026, we found it only as a draft. It would require that any computer program or algorithm, "including any model or tool that relies on artificial intelligence", used on a digital platform is checked for risks of harm to customers, tested and documented. If your company plans automated underwriting tools, that draft is worth reading now.

Data protection. Proposal forms are full of personal data. Under the Data Protection Act, 2019, health status, family details and property details are sensitive personal data. Section 46 says health data may only be processed by or under the responsibility of a health care provider, or by a person under a legal duty of professional secrecy. A public chatbot is neither. Section 35 gives people the right, with some exceptions, not to face a decision based solely on automated processing that significantly affects them.

Six tasks where AI helps underwriters

Each prompt uses the four Cs we teach at FuKazee: Clarity, Context, Constraints and Customisation. Replace the square brackets, and mask every document first. If your company has an approved AI tool for client data, follow its rules instead.

1. Summarise a risk survey report

Survey reports can run to many pages. A short summary helps you see the hazards quickly.

Clarity: summarise the masked risk survey report below into a one-page note. Context: I am an underwriter at a Kenyan general insurer, reviewing a [fire / burglary / engineering] risk for "Insured A", a [type of business] in [town]. Constraints: use only the report. Use these headings: premises and construction, main hazards, protections in place, recommendations made by the surveyor, and open questions. Quote the surveyor's recommendations exactly. Do not suggest terms, premium or acceptance. Customisation: bullet points I can read in two minutes. REPORT: [paste masked report]

Check: open the original report and confirm every hazard and recommendation. Look for anything the summary left out, especially warnings buried in the middle.

2. Check a proposal form for gaps

Below is a masked proposal form for [class of insurance]. List every question that is unanswered, unclear or contradicted by another answer. Show the result as a table: question, what is missing or unclear, and a polite question to send back to the broker or client. Do not guess any answer. FORM: [paste masked form text]

Check: compare the list with your own underwriting guide. AI can miss a gap that only matters for your class of business.

3. Build a checklist for a class of business

Create a checklist of the documents and information an underwriter typically needs before quoting [class, for example contractors' all risks] for a business client in Kenya. Group it under: about the client, about the risk, past losses, and documents. Mark each item "required" or "useful". Add a note at the top: "Check against our current underwriting manual."

Check: your underwriting manual and reinsurance treaty terms are the real rulebook. Edit the list to match them, then save it as a team template.

4. Draft a referral note

When a risk is above your authority, a clear referral saves your manager or reinsurer time.

Write a referral note to my underwriting manager about a risk that is above my authority limit. Facts: [paste masked facts: class, sum insured, location type, loss history, survey findings]. Headings: summary of the risk, why I am referring it, key hazards, information still missing, and my questions. Under 300 words. Do not recommend acceptance or decline; leave a blank section headed "Underwriter's view" for me to fill in.

Check: the facts match the file. Write your own view in the blank section, because that judgement is the point of the referral.

5. Explain terms in plain words after you decide

I have decided to offer cover for [class] with these terms: [paste the terms, excess, conditions and exclusions you set]. Write a short explanation for the broker to share with the client. Use plain English with no jargon. Explain what each condition means in practice and what the client must do. Under 250 words. Do not change or soften any term, and do not mention premium or say the client is covered.

Check: read every line against your quotation. A plain explanation must say exactly what the policy says, never more.

6. Review loss experience in Excel

I have a masked table of [number] years of claims for this account with columns: year, claim type, amount paid, amount outstanding. Give me Excel formulas to total paid and outstanding by year and by claim type, and build a simple chart. Then list three questions an underwriter might ask about the pattern. Do not calculate a premium.

Check: test the formulas on a few rows by hand. For step-by-step help with Copilot, see how to use Copilot in Excel. Microsoft says Copilot in Excel may not be included in every Microsoft 365 subscription, so check with your IT team.

What AI must not do in underwriting

  • Make the risk decision. Accepting, declining, loading, excluding and pricing stay with the underwriter, within authority limits and treaty terms.
  • Confirm cover. Section 156(1) ties cover to the premium being received, and confirmations come from your system.
  • See unmasked proposals in a public tool. No names, ID numbers, KRA PINs, number plates, addresses, bank details or medical history.
  • Judge medical evidence. Medical underwriting decisions follow your medical underwriting rules and medical advisers.
  • Treat a summary as the document. The original proposal, survey and policy wording are what count.

For more on data rules, read our guide to the Data Protection Act and AI at work.

Questions people ask

Will AI replace insurance underwriters?

Not on the evidence we have seen. AI can read and summarise quickly, but it cannot carry an authority limit, defend a decision to a regulator or understand a client relationship. It takes over some of the reading and writing, which leaves more time for judgement.

Can I paste a proposal form into ChatGPT?

Only after you remove everything that identifies the person or business, and only if your company allows it. Medical proposal forms should stay out of public tools altogether.

Does the IRA have rules on AI in underwriting?

On the IRA pages we checked on 15 November 2026, we found no AI-specific rule. The draft Financial Consumer Protection Framework includes the IRA and covers AI on digital platforms, but it was still a draft when we checked.

Can AI help me read a reinsurance treaty?

It can summarise a masked treaty into plain headings, such as limits, exclusions and reporting duties. Always check the summary against the full treaty, and ask your reinsurance team when something is unclear.

Which AI tool should an underwriting team start with?

Start with the tool your company already approves. For teams on Microsoft 365, that may be Copilot. Otherwise, the free version of a mainstream tool is enough to practise with masked documents.

Sources

All sources last checked 15 November 2026.

Where FuKazee fits in

FuKazee is a Nairobi training company that has trained more than 3,000 people since 2022. In our insurance training, underwriters practise reducing a long risk survey to a one-page note of hazards and open questions, while the underwriter sets the terms. We train claims, underwriting, customer care and distribution teams separately, then bring leaders together to agree shared rules. Sessions run at your head office or branch by arrangement, or live online. We sell no software and recommend no single product. We are AI trainers, not underwriters, so your underwriting manual sets the limits.

To plan training for your underwriting team, email lead@fukazee.com

Want guided support instead?

Plan AI training around your team's real work. Answer a few quick questions about your sector, team size and goals, and we will recommend the right programme and format. Sessions are built on your workflows and include a usage policy.

Rodney MainaFounder & Director, FuKazee

Rodney has run FuKazee's training since 2022 and still teaches inside every programme line: corporate sessions, the postgraduate cohort and the KCSE examiner webinars. First Class Honours in Journalism and Mass Communication, University of Nairobi.