Azimut AI
Illustrative scenario. The company on this page doesn't exist: it's a typical profile for the sector, built to explain how Azimut would be used. It doesn't describe any customer or claim measured results.
The company in this scenario
- 160 people: 100 in sales and account management, 30 in claims and 30 in support functions.
- Commercial lines: property, liability, fleets and transport.
- Broker management software, policy wordings as PDFs in a shared repository, email and Teams.
- Customer disclosure obligations under insurance distribution rules.
The problem
A broker's knowledge is in the policy wordings: hundreds of PDFs from different insurers, with exclusions that change from one version to the next. Comparing two liability policies means reading both end to end. Preparing a large renewal means manually gathering the claims history, the correspondence and last year's terms. And the senior brokers' judgment (which insurer responds best in each line) isn't written down anywhere.
How it would work with Azimut
Azimut would index the policy wording repository and email with the existing permissions, and the broker would publish four agents.
| Agent | What it does | Who uses it | What gets measured |
|---|---|---|---|
| Policy wording lookup | Answers what a policy covers and excludes, and compares coverage across insurers, citing the clause. | Brokers | Time to answer the client. |
| Renewal prep | Pulls the account history, correspondence and current terms into one report before renewal. | Account managers | Hours per renewal. |
| Claims follow-up | Summarizes each claim's status from email and documentation, and drafts the update to the client. | Claims | Response time to the client. |
| Pre-contract disclosure | Prepares a draft of the disclosure that distribution rules require from the proposed policy, for the broker to review. | Sales team | Hours per proposal. |
Comparing policy wordings calls for the top-capability model: it's close reading of long documents. Claims follow-up and client updates work well with a mid-tier one. Each user picks the model per task from OpenAI, Anthropic, Google and Mistral.
Where you'd start
You'd start with policy wording lookup. It's the most time-consuming task, the whole commercial side uses it, and its output is easy to verify because the agent cites the source clause.
What's out of scope
Explicitly out of scope: risk assessment and pricing for life and health insurance of natural persons. The EU AI Act treats them as high risk (Annex III), and Azimut excludes them in its terms of use. There's also no connector for broker management software or insurer portals: it works with documents, email and exports. The recommendation to the client is made and signed by the broker.
How it would be measured
Before turning on any agent, you set the baseline, with the company's own numbers rather than industry averages.
Week 0. Hours per renewal on the last quarter's large accounts, response time to coverage questions, and current AI use at the broker: who, with which tool and with which client documentation.
Month 3. The same numbers, next to Azimut's usage panel by user, area, model and tool.
Frequently asked questions
Can it price a policy or decide whether to accept a risk?
No. Pricing and risk assessment for life and health insurance of natural persons are high-risk cases under the AI Act and are excluded. The agent reads and compares documentation; the broker decides.
Does it connect to insurer portals?
No. It works with the policy wordings, correspondence and exports the broker already has.
What happens when an insurer changes its wording?
You replace the document in the repository and the agent answers with the new version; every answer cites the version it used.