SIU - Special Investigation Unit
What is an SIU: the Special Investigation Unit explained
An SIU, or Special Investigation Unit, is the team inside an insurer that investigates claims and policy applications suspected of fraud. Handlers refer a file when fraud indicators appear, the unit gathers evidence, and the claim comes back with findings attached rather than suspicion.
Composition varies by market. Most Special Investigation Units mix former police investigators, forensic accountants, medical and engineering specialists and data analysts, with counter-fraud analytics running alongside them. In several jurisdictions an SIU function, or an equivalent anti-fraud plan, is a regulatory requirement rather than a choice.
What does an SIU do?
- Investigates referred claims: statements, site visits, document verification, forensic review of images and invoices.
- Screens applications: policy-inception fraud, misrepresented risk details, and ghost broking, which increasingly arrives before any claim exists.
- Detects organised patterns: linked claims, repeat addresses, repeat garages, staged collision networks.
- Supports repudiation: building a file that will hold up under complaint, ombudsman review or litigation.
- Feeds prevention back: what the unit learns should reshape intake questions and triage rules, not just close individual cases.
What triggers an SIU referral?
Referral criteria are normally a written list of fraud indicators. The common ones: a loss reported very soon after inception or just before renewal, inconsistencies between the account given and the damage shown, documents that do not match the supplier's usual format, an unusual claims history, pressure for a fast cash settlement, and images whose provenance cannot be established.
The list is not a scoring sheet. One indicator on an otherwise ordinary claim is a prompt to ask a question, and the innocent explanation turns up far more often than not.
SIU example: a claim eleven days after inception
A new motor policy is taken out, and eleven days later a claim arrives for accident damage with a repair invoice from an unfamiliar garage. The photographs show real damage, but the plate is only legible in one frame.
The handler refers it with two specific points noted: the proximity to inception, and the fact that the images were emailed as attachments with no traceable origin. The investigator checks the registration against salvage records, contacts the garage, and either clears the file in a week or builds a case. What made that referral useful was the specificity. "Feels off" would have cost the unit a week.
Why fraud indicators are a reason to look, not a verdict
This is the discipline that separates a competent unit from an expensive one. A claim is not fraudulent because a system flagged it. It is fraudulent when there is evidence, and the decision belongs to a person who has weighed that evidence in context.
Two failure modes follow from getting it wrong. Treat flags as findings and you generate wrongful declines, complaints and regulatory attention. Treat flags as noise and organised fraud walks through the middle of the book. The answer is risk-based review: let signals decide how much attention a file gets, never what the outcome is.
The scale explains the investment. The Coalition Against Insurance Fraud's 2022 study put the cost of insurance fraud in the United States at $308.6 billion a year across all lines, the first update to that figure in twenty-seven years. In the UK, Aviva reported detecting more than 18,400 suspicious claims worth £233 million across its brands in 2025, equivalent to over £638,000 a day, while Admiral reported £86.8 million of detected fraudulent motor, home and travel claims in 2025, up 71% on 2024, and linked part of the rise to AI tools that alter images and produce documents that never existed. In Belgium, Assuralia recorded €181 million of proven fraud in 2025 across 7,720 cases, up from €147.5 million in 2024, while estimating real fraud closer to €800 million a year. All of these are detection figures, and they move with investigative capability as well as with criminal activity.
Mistakes claims teams make with the SIU
- Referring too late: by month three the vehicle is repaired, the scene is gone and the witnesses have moved.
- Referring on volume rather than quality: a unit buried in weak referrals gets slower on the strong ones.
- Going silent on the customer: an investigation still has service standards, and unexplained delay is the fastest route to a complaint that succeeds.
- Treating a withdrawn claim as proven fraud: withdrawal is not a finding, and recording it as one creates its own liability.
- Never closing the loop: if what the unit finds does not change what intake asks for, the same pattern arrives again next quarter.
A growing share of SIU work is now visual, and two threats sit behind it that need separate responses: synthetic material produced by AI, covered in AI-generated insurance fraud, and edited real files, covered in shallowfake insurance claims. The second is the volume problem, because making one needs no AI at all.
The strongest thing a claims operation can do for its Special Investigation Unit is upstream: make the evidence entering the file traceable, structured and verifiable from day one, so investigators spend their time on judgement instead of on reconstructing where a photograph came from. The standards behind that are in insurance claim documentation.
