The Body Shop Cluster: Pattern Referral Under a Threshold Rule
You are the fraud risk lead at a mid-sized private motor insurer with 1.2 million policyholders. Q3 analytics flagged a 14% rise in third-party injury claims in one metropolitan region. Digging in, you find 38 open claims sharing only three repairers, two claimant law firms and two medical clinics; average notification lag is 27 days against a 9-day baseline. Most claims settle in a band of £4,000–£9,000, comfortably below the £10,000 mandatory SIU referral threshold, so none has been referred. Twelve policies were bought 30–45 days before the loss date, all through a single broker. Your SIU has four investigators and 60 open files. Fair-treatment rules and your own claims conduct policy both require that claimants not be delayed or denied without specific, evidenced grounds. One claimant has already complained about a six-week delay.
1. Which of these signals justify SIU referral, and how do you rank 38 claims against a four-investigator capacity?
Read the worked answer
Separate the indicators before you sort the claims. Three signals are weak alone: region, notification lag, and the claim-value band. Each has ordinary explanations. Three are stronger: the shared repairer/law-firm/clinic combinations, the tight policy-inception-to-loss interval, and the single-broker concentration on those twelve policies.
A 27-day lag can reflect injury progression. Combined with a 30–45 day inception gap and a shared clinic, it moves up the list.
Rank by:
- Combination strength - claims linking two or more strong entities (same repairer and clinic and law firm) go first.
- Irreversibility - anything near settlement or limitation, or where payment is imminent.
- Pattern value - the £4k–£9k band means individual losses are small; the value is in proving or disproving the network, so prioritise cases that test the hypothesis.
- Group coherence - the twelve broker policies form one testable unit; investigate them together.
The £10,000 figure is a referral floor, not a ceiling. Ask for the rule to be changed to permit pattern-based referral. Then stratify: roughly 8 active investigations, 15 data-only reviews, the rest monitored with a trigger list.
2. Set out the analytical checks you would run before any claimant is contacted, and explain what each is designed to prove or disprove.
Read the worked answer
Run checks in this order, each with a stated purpose:
- Entity resolution - normalise names, addresses, phone numbers, bank details, VINs and invoice numbers across the 38 claims and the wider book. Purpose: confirm the cluster is not a data artefact and find its true size.
- Inception-to-loss distribution - compare the twelve broker policies against that broker's whole book and the regional baseline. Purpose: test anti-selection against coincidence, and check for timestamp inconsistencies between bind, payment and endorsement dates.
- Repair invoice forensics - labour hours against benchmarks, part numbers against VIN compatibility, photo metadata, duplicate images. Purpose: distinguish inflated repair from fabricated repair.
- Treatment pattern - duration and modality at the two clinics against comparable injury claims from other clinics. Purpose: test whether treatment is injury-driven or claim-driven.
- Relationship mapping - are any claimants, clinic staff or law-firm staff connected to the repairers or the broker? Purpose: evidence of conspiracy rather than parallel conduct.
- Prior claims history - cross-market and prior-policy checks. Purpose: establish pattern versus first occurrence.
Write the null hypothesis before each check. That discipline is what stops you pattern-matching noise into a fraud finding, and it is the record you will need if a claim is later challenged.
3. What conduct and fair-treatment risks does your approach create, and how would you mitigate them?
Read the worked answer
Three risks need active management.
Delay. Referral to SIU is not a lawful reason to stop handling the claim. Set a service clock: acknowledge, investigate, and pay any undisputed element - for example vehicle damage where only the injury is disputed. Write to the claimant explaining the claim is taking longer, without accusation.
Profiling. The flagged region will correlate with ethnicity, income and immigration status. Score on behaviour and relationships, not postcode or surname. Have the model reviewed for disparate impact and keep the review on file.
Vulnerability. Injury claimants are often in pain, off work and financially stressed; an investigation can be experienced as intimidation. Screen for vulnerability at first contact and route to a specialist handler.
Governance. Any denial or reduction needs an independent second reviewer and a reason that would survive an ombudsman. Keep a decision log recording what was checked and found - for the claims you clear as much as for the ones you challenge. Pull the existing complaint file before you touch that claim, since it may already be on a regulatory clock.