AI ActHigh-risk AI rules for life and health insurance pricing apply from 2 December 2027.See what to keep
Sign inBook a demo
Fraud signals

The model points. A person decides.

Link analysis lays out claims, parties, repairers, phone numbers and addresses so an investigator can follow a pattern. Every signal shows where it came from. Nobody is refused a claim by a model alone, and a referral to the SIU is a decision with its own record.

Every signal with its sourceNobody refused by a model aloneAuthorised before data is collected
honestas · fraud · link analysisSample data
Link analysis showing one roofer linked to twelve storm claims through a shared phone number
Twelve claims, one roofer. Castell Roofing invoiced twelve Storm Fionn claims in six days, with its phone number on two of them and every bill within £700 of the others.Sample data
What people see

Reasons, not scores. Sources, not hunches.

A handler sees why a claim was flagged, from which records, and the case for the customer beside the case against. The claim keeps moving while they decide.

Warning signs on one claim

Three kinds of reason from three different places, each with its source, and no score. The case for the customer sits beside the case against.

  • A link, a photo date, words that match other claims
  • The case for and the case against
  • Holding a payment is a separate decision
honestas · fraud · one claimSample data
Fraud indicators on one storm claim, each with its source, and the case for and against
Fraud indicators on a claim. Sara decides whether to raise HX-39713. The claim does not slow down while she does.Sample data
From signal to referral

Four steps, and a person at each one that matters.

01

Signals found

A nightly run and link analysis look for shared phones, shared words, photo dates and repeat repairers.

02

Shown with sources

Each signal names the record it came from, beside the case for the customer. There is no score to hide behind.

03

A person decides

Raise it to investigations, set it aside with a reason, or ask the customer a plain question.

04

Each claim on its own

A group can be investigated together. Each claim is still decided, and paid, on its own merits.

The SIU referral

A referral is a decision. It gets its own record.

Sending a claim to special investigations changes how a customer is treated, so it is decided like any other decision. A named person raises it, a named person authorises it before any data is collected, and the scope of what may be collected is written down and sealed.

  • Authorised before collection, never after
  • The scope of what may be collected, in writing
  • The customer is not tipped off, and not treated differently on a call
  • Closing a referral needs a reason in words
Referral record · REF-2026-0412

Castell Roofing, referred to the SIU

Decision
Refer twelve claims to special investigations
Raised by
Idris Kane, claims handler, in his own words
Authorised
Sara Lindqvist, claims lead, before any data was collected
May collect
Invoices and phone numbers on these twelve claims, Storm Fionn only
Payments
Each claim decided on its own. None held because of the group
On calls
The customer is not tipped off and not treated differently
Scope
It records why a person looked. It does not say anyone committed fraud
sha256 41c7e0 … 9b12Sealed 17 Sep 2026 · 15:22:08 UTC
Sample record
The rules it answers to

Fair to the honest customer. Firm with the rest.

Each referral and each refusal carries the reasons a person gave, so it can be explained to the customer, an ombudsman or a supervisor.

  • GDPRRegulation (EU) 2016/679, Article 22No decision with legal or similar effect based solely on automated processing, without safeguards.
  • SCHUFACJEU, Case C-634/21, 7 December 2023A score can itself be an automated decision when others rely on it heavily.
  • Dun & Bradstreet AustriaCJEU, Case C-203/22, 27 February 2025People can ask how an automated decision about them was reached, in words they can follow.
  • EIOPA opinion on AIAI governance and risk management, 6 August 2025Fairness, human oversight and records for the AI systems insurers use.

Not legal advice.

Questions

About fraud signals.

Does HONESTAS give claims a fraud score?

No. It shows reasons, each with its source, beside the case for the customer. A person decides what they mean.

Can a model refuse a claim?

No. A model can point at a pattern. Refusing a claim is a decision a named person makes, with reasons, in a decision room.

Is the customer told about a referral?

Not while it is open, so an investigation is not prejudiced. They are never treated differently on a call, and every decision about their claim carries reasons that can be explained to them.

What can investigators collect?

Only what the authorised scope allows, written down before collection starts. Asking for more is a new decision.

Can investigators see behind walls?

No. A path through a special investigation behind a wall is counted, so others know it exists, and never drawn.

Book a demo

Bring the referrals you already make.

Bring a handful of last year’s SIU referrals. We will show each one with its signals, its sources and a record of who decided.