On this page
Kenyan claims teams use AI mostly for the paperwork around a claim: summarising what the customer reported, checking a file for missing documents, drafting status updates, and listing details that do not match so an investigator can look closer. The decision to accept or decline a claim, and how much to pay, stays with a qualified person. Kenya's Data Protection Act also limits decisions made only by machines.
This guide is for claims handlers, supervisors and customer care staff. For regulators and a vendor checklist across insurers and SACCOs, read our guide to AI for SACCOs and insurers in Kenya.
Why claims teams are looking at AI
The volume is large and the clock is real. The Insurance Regulatory Authority's (IRA) claims report for April to June 2026 shows that:
- General insurers recorded about 3 million policyholder (non-liability) claims reported or revived in the quarter, worth KES 22.30 billion.
- 142,202 third-party (liability) claims were still outstanding at the end of June 2026.
- The IRA registered 354 complaints against insurers in the quarter. In both general and long-term insurance, most were about delays in settling claims.
Section 203 of the Insurance Act sets the clock. Where the claimant has submitted all the relevant documents, the insurer must admit or deny liability, work out the amount, confirm who the claimant is and pay, within 90 days of the date the claim was reported. If it cannot, it must ask the Commissioner, who can allow up to 30 more days. After that, a penalty of 5% of the unpaid amount falls due. Where an assessment was done, the claimant must get a copy of the report. Index-based and micro-insurance claims are excluded from this section.
AI cannot make a garage send documents faster, but it can save hours of reading and rewriting.
Six claims tasks where AI helps today
First, remove names, ID numbers, KRA PINs, phone numbers, number plates, policy numbers and medical details, or use a tool your company has approved for real claims data. Use labels such as Claimant A and Vehicle 1.
Each prompt follows the four Cs we teach: Clarity (what you want), Context (who you are and the situation), Constraints (the rules) and Customisation (who will read it).
1. Intake summaries
I am a claims handler at a Kenyan general insurer. Below are my notes from a customer's call reporting a motor accident, with identifying details replaced by labels. Turn them into an intake summary with these headings: what happened, when and where, vehicles and people involved, injuries reported, police report mentioned (yes or no), documents the customer says they have. Use only my notes. Write [NOT STATED] where something is missing. Keep it under 200 words for a claims supervisor.
Check: compare it with your notes. Summaries drop details, and nothing should appear under "injuries" that the customer did not say.
2. Missing-document checks
Below is our document checklist for [claim type], then the list of documents received so far for Claim A. Return a table with three columns: document, received (yes, no or unclear), notes. List missing items first. Do not judge whether the claim is valid. Do not add documents that are not on our checklist.
Check: use your current checklist, not one the tool suggests. Section 203 links the 90 days to having all the relevant documents, so ask for everything missing early, in one message.
3. Customer status updates
Write an SMS of under 300 characters and a short email to Claimant A about their claim. Facts: [stage reached, what we are waiting for, who it is coming from, date of our next update]. Constraints: plain English, calm, no promise of payment or of any date I have not given, no blame on the garage or hospital. Customisation: the customer is worried and may read this on a basic phone. Then give a Kiswahili version of the SMS.
Check: a fluent Kiswahili speaker reads the translation. A supervisor approves each template once, and staff reuse it. Our guide to training customer service teams on AI covers tone in more depth.
4. Explaining an assessment report
Below is an assessor's report on Claim A, with labels in place of names. Write a one-page cover note for the claimant explaining, in plain English, what the assessor found, the repair estimate and any items the assessor left out, quoting the report for each point. Do not change any figure. Do not say whether the insurer will pay. End with who the claimant can call with questions: [contact].
Check: every figure against the report. The decision letter comes separately from the claims officer.
5. Fraud flags that a person reviews
Below is an anonymised claim file: claim form, driver's statement, police abstract summary and assessor's report. List every point where two documents disagree (dates, times, places, damage described, people present), quoting each line exactly with the name of its document. Do not say whether fraud is likely. Do not suggest a motive. Do not score the claimant.
Check: treat every line as a question for an investigator, never a finding. If a document looks altered, confirm it with whoever issued it, not with a chat tool. Section 204B of the Insurance Act makes it an offence to knowingly present false or misleading information in a claim, and names "computer-generated documents" specifically.
6. A claims ageing tracker in Excel
I use Excel. My columns are: Claim Ref, Date Reported, Date All Documents Received, Status, Date Paid. Give me formulas for days since reported, days since all documents arrived, and a flag that shows "REVIEW" at day 60 and "URGENT" at day 75 for unpaid claims. Explain each formula in one sentence.
Check: test it on three claims you have worked out by hand. How your company counts the 90 days is a question for compliance, not the tool. See how to use Copilot in Excel.
What AI must not do in claims
- Decide a claim. Section 35 of the Data Protection Act gives people the right not to be subject to a decision based solely on automated processing that significantly affects them. Where an exception applies, the person must be told in writing and can ask for a fresh decision that is not solely automated.
- Call anyone a fraudster. An AI flag can start an investigation. It is never the outcome.
- Handle medical files in public tools. Under section 46, health data may only be processed by or under the responsibility of a health care provider, or by someone under a legal duty of professional secrecy. Medical claims teams need legal advice and an approved system.
- Quote policy wording from memory. Paste the actual clause.
- Write to a customer unchecked. A person reads every message before it goes.
The Kenyan rules to know
- Registration. The ODPC's registration regulations list "provision of financial services" among the purposes that must register with the Data Commissioner, even for businesses small enough to be exempt otherwise.
- Sending data abroad. Section 25 says personal data should not leave Kenya without proof of adequate safeguards or the person's consent. Most public AI tools run outside Kenya.
- Complaints. Section 204A lets any insurance customer lodge a written complaint with the Commissioner. The IRA's website gives a complaints form, complaints@ira.go.ke and a toll-free line, 0800 724 499.
- AI rules from the IRA. We found no AI-specific rule on the IRA pages we checked on 14 November 2026. Its list of circulars did not load for us, so ask compliance.
Put these points in your AI usage policy, and see the Data Protection Act and AI at work.
Questions people ask
Can AI approve insurance claims in Kenya?
Not on its own. Section 35 of the Data Protection Act limits decisions based solely on automated processing, and section 203 of the Insurance Act makes the insurer responsible for admitting or denying each claim. AI organises the file. A qualified officer decides.
Can I paste a claim file into ChatGPT?
Not with real names, ID numbers, plates or medical details in a public tool. Remove them first, or use a tool your company has approved. Medical files need the most care.
Does AI make claims faster?
We found no published Kenyan figures on this. Time one task, such as status updates, for two weeks before and after, including checking time.
Can AI detect fraudulent claims?
It can list inconsistencies between documents for a person to follow up. It cannot prove fraud, and it should never be the reason a claim is declined. Confirm suspicious documents with whoever issued them.
Which tool should a claims team start with?
The one your company has approved. If you have none, practise with free tools on non-personal material such as templates and checklists, and ask IT before any real claims data goes into a tool.
Sources
All sources opened and last checked on 14 November 2026.
- Insurance Regulatory Authority, Insurance Industry Quarterly Claims Statistics, April to June 2026 (PDF)
- Insurance Act, Cap. 487, as published on the IRA website (legislation as at 11 December 2023), sections 203, 204A and 204B (PDF)
- Insurance Regulatory Authority, Lodge a Complaint and home page contact details
- ODPC, Data Protection Act, 2019, sections 25, 35 and 46 (PDF)
- ODPC, Data Protection (Registration of Data Controllers and Data Processors) Regulations, 2021, regulation 13 and Third Schedule (PDF)
- FuKazee facts: AI training for insurance companies in Kenya
Where FuKazee fits in
FuKazee is a Nairobi training company, and we have trained more than 3,000 people since 2022. In our insurance sessions, claims handlers, underwriters, agents and customer care teams practise on the letters and files they handle every week, with customer details masked. A person still makes every claim decision, and AI never labels anybody a fraudster. We use Copilot for firms on Microsoft 365 and the free versions of other tools. See AI training for insurance companies in Kenya.
To plan training for your claims team, email us at 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.





