Claims fraud detection that reads every new claim and refers the suspect ones to the SIU, with the proof
Referrals in days, not weeks, after first notice, each with its evidence.
See one case, screen by screen ↓Software that looks at claims for signs of fraud and sends the suspect ones to the special investigations unit (SIU). Claims Fraud Detection reads every new claim at first notice — notes, documents, photos and prior claims, not only the structured fields — and gives each one a score out of 1,000 with its reasons in plain words and the proof behind each.
How to choose AI solutions for insurance →
Updated
The signals are in the file, but they are spread across it: a policy bound nine days before the loss on the declarations page, an application answer that does not match the contents list, a prior theft at the same address under another household member’s name, two receipts with numbers in sequence. Rule scores see the structured fields. The adjuster with 140 open claims has a feeling, and no time to prove it.
So the referral comes weeks after first notice, or never. When it does arrive, the SIU intake analyst spends the first hours rebuilding what the adjuster already saw — pulling the declarations, the application, the claim history — before deciding whether to take it. A referral that cannot hold up gets sent back as weak.
Estimated split for a typical claim referred by hand, and the same claim read at first notice.
Six agents read every new claim, score it with reasons and the proof behind each; the SIU intake analyst decides.
Darren Meachum reports a burglary at home on Tue 20 Oct at 09:14 and claims $48,620. Here is what happens next, screen by screen, in the working solution.
Imani opens the referral queue. 1,284 claims were read in the last 24 hours, ranked by score, with a dashed line at 600: claims above it wait for the SIU, claims below pay normally. Six are waiting for her; the desk has already decided 17 today — 14 accepted, 2 returned, 1 closed.
“Referral line · 600 — above waits for the SIU · below pays normally.”
HO-26-118734 was read 17 minutes after first notice. The run log shows each step: the call, 6 adjuster notes, 41 pages and 14 photos turned into 31 facts with sources; one prior claim at the address; six indicators matched from list v22; two innocent explanations found. A baseline of 120 for homeowners theft, six reasons adding 760, the police report and a hotel receipt taking 68 off: 812, which is 212 above the line.
Imani opens the top reason. A to-scale timeline runs from the online purchase on Wed 30 Sep at 21:48, to the policy start on Thu 1 Oct, the loss on Sat 10 Oct at about 02:00, the online police report on Mon 19 Oct and first notice on Tue 20 Oct. Beside it, the declarations page and the online application, with the passages highlighted: the personal property limit suggested at $42,000 and selected at $70,000, and “No” to prior homeowners insurance in the last 3 years.
Why it holds up: “Home owned since 22 Mar 2019 (county record) — yet no prior insurer declared.” And its limit: “Not proof on its own — 41% of referrals with this indicator were confirmed last quarter.”
A theft from the residence on 9 Mar 2024, claimed by Dana M. Meachum at 2716 Quail Hollow Way with another carrier: $22,150 paid, for jewelry, a laptop and a tablet. The match rests on two identifiers — the address and the phone — and the declarations page lists Dana M. Meachum as a resident. The application answered “No” to losses at this address in the last 5 years.
A match needs two identifiers, and a prior claim is never treated as proof.
Marquise Fine Jewelry receipt No. 004417, dated 14 Feb 2025, for a 14k tennis bracelet at $9,800; No. 004418, dated 2 Oct 2025, for a diamond pendant and chain at $8,600. Both paid in cash, same pad and layout, photographed together on 20 Oct at 08:58 — after the loss.
Ten sources were read — first notice call, adjuster notes, declarations page, online application, police report, jewelry receipts, contents list, damage photos, hotel receipt and the prior claim record. Each highlighted line says which reason it feeds. The 14 damage photos are read too, for what they show and what they leave out.
“What the photos do not show: the slider lock or frame, any item before the loss. All 14 were taken on 20 Oct between 07:40 and 08:06.”
Imani has three choices: accept the referral, return it to Felipe with a question, or close it with no referral. She accepts. Case SIU-26-0412 opens in the Sacramento SIU, assigned to Kevin Hsu, with the packet attached — one day from first notice. Felipe Arguello is told to keep the claim moving on its clock.
A referral never stops the clock. The adjuster keeps the claim moving.
The packet writer drafted it in seconds: a summary with every sentence linked to its evidence, the indicators and their points against an index of 9 evidence items, the innocent explanations, and the claim’s California clocks. Three suggested first steps: recorded statements from Darren and Dana Meachum, separately; proof of ownership for the two jewelry items; and the 2024 claim file through the industry claims history.
“Examination under oath is the investigator’s call under the policy conditions. Nothing here recommends a denial.”
The outcome tracker reads what the SIU found and proposes changes to the indicator list. Three wait for Corinne: lower “reported more than 7 days after the loss” from 110 to 60 points when a police report was filed within 48 hours; raise “receipts numbered in sequence” from 140 to 170; retire “damage photos taken at night, location not shown”. Tested on last quarter’s 1,412 closed referrals, they mean 67 fewer weak referrals and 4 more confirmed frauds above the line.
“2 confirmed frauds would drop below the line — review before approving.” Every change needs Corinne and a second approver.
Suspected fraudulent claims go to the California Department of Insurance within 60 days of the SIU’s determination. The Andrade staged-collision ring of 5 claims, determined on Wed 2 Sep, is due Sun 1 Nov with 11 days left; the draft is ready, written from the case file, for Corinne to edit or approve and file.
The same claim, seen by the four people who carry it — what their week looked like, and what it looks like now.
Six agents read every new claim, score it with reasons and the proof behind each; the SIU intake analyst decides.
Reads first notice, adjuster notes, documents, photos and invoices on every new claim, and writes the facts with their source line. Describes photos by what they show and when they were taken.
Finds prior claims for the policy, the address, the people, phones and bank accounts, internally and in the shared industry claims history.
Matches the facts to the approved indicator list — industry and the insurer’s own — and writes each reason in plain words with its points.
For each reason, looks for facts in the file that explain it innocently — a prior insurer, a lapse letter, a police report, travel proof, a witness — and lowers the score when it finds one.
Writes the SIU referral packet: summary, indicators, explanations, suggested first steps and an evidence index, with every sentence linked to its evidence.
Records what the SIU found on every closed referral against its indicators, proposes point changes where the confirmed rate differs, and tests them on last quarter.
Decides every referral; the fraud manager approves list changes. The agents propose; a named person decides.
The SIU desk can ask in plain words about a claim, a reason, an indicator or a deadline. Every answer cites its source.
Why is Darren Meachum’s claim above the line?
It scores 812 against a line of 600. Three reasons carry most of it: the policy started on 1 Oct, nine days before the loss; a theft at the same address was paid to Dana M. Meachum, who lives there, in March 2024; and the two cash jewelry receipts are numbered 004417 and 004418 but dated eight months apart. The police report and a hotel receipt for the Reno trip take 68 points off.
Is a policy bought 9 days before the loss enough to refer?
On its own, no: it is one indicator, worth 190 points, and 41% of last quarter’s referrals with it were confirmed. It holds up here because the home has been owned since March 2019 with no prior insurer declared, and the contents limit was raised from $42,000 to $70,000 during the quote. The claim keeps its own clock either way: accept or deny within 40 days of proof of claim.
Which referrals did the SIU send back as weak this month?
11% in October, against 38% before go-live on 3 Aug. The most common pattern: reported more than a week late, with a police report filed within two days of the loss — only 9% of those were confirmed. The outcome tracker proposes lowering that indicator from 110 to 60 points in that case; it waits for Corinne Lathrop and a second approver.
What must we report to the Department of Insurance this month?
Two drafts are ready. The Andrade ring (SIU-26-0351) was determined on 2 Sep, so it is due by Sun 1 Nov — 11 days left; Wesley Darrow (SIU-26-0364) is due by Fri 13 Nov. Both are due 60 days after the determination. Kinsey Paulsen was filed on 15 Oct.
13 screens from the working solution, on its sample data. Pick one to see it large.
Referral lines, baselines, clocks and who decides are settings. Changes are versioned and need a second person from the SIU.
| Setting | Default | Choose from |
|---|---|---|
| Referral line, score out of 1,000 | 600 | Set per line: homeowners, personal auto, commercial auto, workers’ comp |
| Audit sample below the line | 1% | Share of claims below the line sent to the SIU at random |
| Baseline points a claim starts with | 120 for homeowners theft | Per loss type: homeowners fire 150 · auto theft 140 · auto injury, 3 or more people 150 · glass 40 |
| Fair-claims clocks | California | Acknowledge 15 days · accept or deny 40 days after proof of claim · written status every 30 days |
| Fraud bureau report | 60 days | Days after the SIU’s determination · Ins. Code §1872.4 |
| Answer to a returned question | 1 business day | Business days for the adjuster to answer |
| Shared claims history match | 2 identifiers | Identifiers that must agree before a prior claim counts |
| Reading | Every new claim | At first notice and when a document arrives · re-read can raise or lower the score · tell the adjuster when a claim crosses the line |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample claims data · “target” = the design goal, measured in the live solution · “estimated” = our estimate · “typical” = published figures (Coalition Against Insurance Fraud, The Impact of Insurance Fraud on the U.S. Economy, 2022, and fraud stats — about 10% of P&C losses). Clocks shown are California’s: 10 CCR §2695.5(e) and §2695.7, Cal. Ins. Code §1872.4. People, companies and claims named on this page are fictional — characters and sample data in the working solution.
Software that looks at claims for signs of fraud and sends the suspect ones to the special investigations unit (SIU). Claims Fraud Detection reads every new claim at first notice — notes, documents, photos and prior claims, not only the structured fields — and gives each one a score out of 1,000 with its reasons in plain words and the proof behind each.
Each claim starts from a baseline for its loss type, adds points for each indicator on your approved list that the file supports, and takes points off for facts that explain a reason innocently, such as a police report or travel proof. The score is only the sum of the reasons shown. Claims at or above the referral line wait for the SIU.
No. No agent denies, delays, pays or contacts the insured, and the packet never recommends a denial. The adjuster keeps handling the claim on its fair-claims clocks while the SIU decides on the referral.
Yes. The SIU intake analyst accepts, returns or closes every referral. Changes to the indicator list need the fraud manager and a second approver, and changes to settings need a second person from the SIU.
No. Race, colour, religion, national origin, sex, marital status, age, disability, ZIP code or neighbourhood on its own, and credit-based insurance scores are locked: they are not on the list and cannot be added. Every indicator must describe the claim, the policy or a documented fact about it.
The outcome tracker records what the SIU found on each closed referral and proposes point changes where an indicator’s confirmed rate differs from its points. Each proposal is tested on last quarter’s closed referrals — showing, for example, confirmed frauds that would drop below the line — before the fraud manager and a second approver decide.
Yes. It tracks suspected fraudulent claims due to the state fraud bureau — in California, within 60 days of the SIU’s determination under Insurance Code §1872.4 — drafts each report from the case file for the fraud manager to edit and file, and exports the log.
The Agentic Solution Engine builds and deploys it from your requirements — your indicator list, referral lines, SIU packet format and a sample of closed referrals — and it goes live once every quality gate has passed. We will walk you through it on your own claims first.
Every claim document over a set amount tested for edits before it is paid.
1-min demoFraud Network AnalysisOrganized rings seen across claims, not missed one claim at a time.
1-min demoContractor Fraud DetectionContractor clusters raised before payment, not found after it.
1-min demoWe’ll run Claims Fraud Detection on a sample of your own closed referrals.