Skip to content

Data and AI for insurance companies

To decide on a claim, the claims handler reads the notification, the loss adjusters' reports and the policy wording side by side, and the wording has to be the right edition. Cover changes with every edition, while policies issued under earlier ones remain in force.

Where data and AI change an insurer's work

A claim on a policy issued under an earlier edition

Today the claims handler receives the notification, looks for the wording of the edition under which the policy was issued and compares it by hand with the loss adjuster's report before proposing an amount, and sometimes reads the covers of a later edition. In the example, the agent opens the claim, retrieves the wording of the right edition from the insurer's knowledge base, checks coverage and prepares the settlement proposal, showing the clauses it relies on. The claims handler checks it, corrects it if needed and decides.

  1. The agent opens the claim and retrieves the wording of the right edition
  2. Checks coverage and proposes the settlement
  3. The claims handler checks, corrects and decides

Illustrative example

The product is described only once

In the Company Brain we record covers, exclusions and editions with the relationships between them, so that claims, underwriting and the network all read the same meaning. The expectations of IVASS, the Italian insurance supervisor, on data and model governance weigh here, as do the AI Act's obligations on high-risk systems for models that assess risk and set prices for individuals' life and health policies, while the health data in claim files requires separate access.

The claims handler is accountable for the decision

The agent opens the claim, checks coverage and proposes the settlement; the claims handler then checks it, corrects it if needed and decides. Product, claims and the network are processes the Join Business Group knows from the side of those who run them.

If the policy wording has no definite version, or if a product's editions cannot be told apart with certainty, we put the documents in order first and only then build the agents. This lengthens the first project.

The intermediary only needs to call once

We write the rules for clauses that are open to different readings with the people who know the product, so the network gets answers consistent with what the insurer actually sold.

Claim preparation time, exceptions and rework

We look at the time needed to prepare a claim and how complete it is when it reaches the claims handler, at how many exceptions the process produces and at how many claims have to be reworked, always against a starting point measured before we begin.

Frequently asked questions

If the agent proposes the settlement, who is accountable for the decision?

The claims handler, who checks the proposal, corrects it if needed and decides. Cases with anomalies reach the fraud unit with the file already prepared.

The wording of our products does not always have a definite version. Can we start anyway?

Yes, but first we put the documents in order, until every edition of a product can be told apart with certainty, and only then build the agents. The first project therefore takes longer.

Who sees the health data in the claim files?

We set up separate access for the health data in claim files, so only the people and agents authorised to process it can read it.

Let's think through one of your processes

We can start with the line of business where your claims sit idle longest, looking with the claims handlers at the step where they get stuck.