ISO ClaimSearch and how fraud analytics flag claims

ISO ClaimSearch is a database, run by the analytics company Verisk, that holds claims reported by insurers across the property and casualty industry. When an adjuster opens your claim, the system compares it with the claims already on file, lists matches such as prior losses or a phone number seen before, and flags combinations that look unusual. A link analysis tool maps connections between people, providers and vehicles, and scoring models rank claims by resemblance to past fraud. None of this is a finding that you did anything wrong. A flag asks a person to look, and honest claimants are flagged more often than they expect because ordinary life produces coincidences.

What ClaimSearch is and who feeds it

Verisk describes ClaimSearch as a claims intelligence network built from contributions by insurers. Its history, as Verisk's own timeline tells it (read November 2026): the Index System was founded in 1920 by an insurance trade association; a property loss register launched in 1980; and in 1998 ISO ClaimSearch was formed from the merger of the INDEX and PILR databases and the auto theft claim system of the National Insurance Crime Bureau (Verisk ClaimSearch). The database reached 1 billion claims in 2015, according to the same timeline, and surpassed 1.8 billion claims in 2024.

The company's 2022 product brochure gives a sense of scale and users. As of September 2022, it said that an adjuster entering a claim compares it with more than 1.5 billion claims, that more than 90 percent of property and casualty insurers use the product, and that other users include more than 1,000 self-insured organizations, all 27 workers' compensation insurance funds, third-party administrators and state fraud bureaus (Verisk brochure, 2022). Those are Verisk's marketing figures, dated, and they have grown since.

What goes in is claim information: who was involved, what was insured, what was lost, where, when, and the claim's status and payments. The product also holds loss photographs, and Verisk reports that its image database reached 500 million photos in 2024, and describes image forensics that analyze photo metadata to detect potential fraud. Contributing a claim also routes some required reporting. The brochure lists services such as Medicare and child-support lien checks, the federal vehicle title database (NMVTIS) and state fire marshal reporting.

It is an insurer-facing tool. Drivers do not log in. That matters later in this article, when we look at what you can and cannot see.

What happens when your claim is entered

A match report is the first thing an adjuster sees. Verisk's brochure says a match can turn up prior related losses, multiple policies, previous SIU involvement, several addresses for the same individual, and a Social Security number used for more than one person. Each claim is scanned automatically and color coded: red for claims with alerts, yellow for matches with no alerts, and green for neither. Verisk's own guidance is that green claims might be fast-tracked, yellow claims with matches "may need a quick review," and red claims "may need further evaluation and possibly SIU involvement."

The 2022 figures show how ordinary a match is. Verisk reported an average of 5.6 matches per claim and an average hit rate of 84 percent. If a typical claim has several matches, then having one tells an adjuster almost nothing. The question is which matches, how recent, and whether they fit together in a suspicious way.

The step that follows depends on the insurer. A small, straightforward claim with a yellow result usually continues. A red result, or a claim that a scoring model ranks high, is more likely to be routed to a special investigations unit (SIU), a team that investigates suspected insurance fraud. The guide on how insurers investigate fraud describes what that team does next.

How link analysis finds connections

Fraud is often organized. Staged crashes, inflated treatment and recycled damage involve repeat participants: the same drivers, passengers, clinics, lawyers, tow operators and body shops. A single claim can look normal. A web of claims is harder to hide. Link analysis converts claims into a network in which people, phone numbers, addresses, vehicles, providers and policies are points, and a shared detail between two claims is a line. Verisk lists tools that "help detect coordinated claim fraud activity," and CNA describes the same method as "visual mapping of relationships among parties to a claim" (CNA special investigations unit).

An invented example shows how it works. Say four auto claims in 18 months list different drivers, but each lists the same cell number as a contact, three name the same clinic, and two put the claimed injuries in a car that was reported with the same VIN on an earlier claim. The pairs mean little alone. The cluster, with the phone number at the center, is what a link chart is built to surface, and it is why investigators can describe a "ring" long before anyone has confessed.

The same math produces false hits. A shared address can be an apartment building with 200 units, a family home with three generations, or a mail drop. A shared phone number can be a parent, a spouse or a workplace. Two people can go to the same clinic because it is the closest urgent care. A towing company can show up in dozens of unrelated claims because it holds the city contract. A network tool cannot tell a conspiracy from a convenience, which is why its output should lead to questions and not to denials.

Scoring models and red flags

Scoring adds a ranking. Verisk's brochure describes predictive models that "provide enhanced claim scores and reason codes to detect potential fraud and support investigations," and the timeline dates claim scoring to 2003 and the introduction of machine learning to 2018. The general method, which other vendors share, is to train a model on past claims that investigators confirmed as fraudulent and past claims that were not, then score new claims by how closely they resemble the first group. Reason codes show which features drove the score, so a reviewer can see why it fired.

Many of the features come from patterns that investigators have documented for decades. The NICB's guide for law enforcement lists indicators for staged or caused accidents, and says plainly that they "only 'indicate' the potential of fraud and are not actual evidence" (NICB scene indicators, undated). The table below pairs several with an innocent explanation, which is the point reviewers have to weigh.

An investigator reading a score sees the first column. A fair investigator also considers the third.

When an honest claimant gets flagged

A few ordinary situations raise flags:

• A recent claim on another policy or a prior claim for the same body part. A match report shows prior related losses even when each was legitimate.
• A new policy or a coverage increase shortly before a loss.
• An address, phone number or Social Security number that appears on another claim, whether through family, a landlord or a typo.
• A provider or repair shop that other flagged claims also used.
• A name that matches a different person with the same name.
• Data entry errors: a transposed digit in a VIN or phone number can create a false link.

The effect is usually delay and extra questions rather than denial. Insurers must still follow claim-handling rules. California's regulation, for example, requires an insurer to accept or deny within 40 calendar days of proof of claim, to send a written notice every 30 days if it needs more time and say what more it needs, and to "conduct and diligently pursue a thorough, fair and objective investigation" without persisting in "seeking information not reasonably required." The same regulation says the insurer need not disclose information that could alert a claimant that the claim is being investigated as possible fraud (10 CCR 2695.7). So you may notice an unexplained delay, and be told only that the claim is "under review." Other states set their own deadlines; see how fast an insurer must respond to your claim: state time limits.

If you think you were flagged

1. Answer questions fully and consistently. Your recorded statement, your claim form and your documents should say the same thing. Inconsistency does more damage than a flag.
2. Explain connections before they look suspicious. If you share a landlord, a phone line or a relative with another claim, say so.
3. Fix errors. Check that your VIN, address and phone number on the claim are right and correct them in writing.
4. Ask for the status in writing. Ask what additional information is needed and when the insurer will decide, and keep dated notes.
5. Provide records promptly: repair estimates, medical records, photos, receipts, a police report.
6. Escalate if deadlines pass. If the insurer is silent beyond your state's deadlines, your state insurance department takes complaints. My claim is under SIU review: what to expect and your rights and when an automated fraud flag delays your claim cover this path in more detail.

How ClaimSearch differs from CLUE

People often use "CLUE" for any insurance claims record, but the products have different jobs. The table compares what I could confirm from each company's own pages and the Consumer Financial Protection Bureau.

The CLUE row comes from LexisNexis's C.L.U.E. Auto page, which describes "up to seven years of personal automobile claims information" with 99.6 percent of the auto industry contributing, used by underwriters. The CFPB's list of consumer reporting companies includes C.L.U.E. with a free report, and its A-PLUS page says A-PLUS "collects and reports insurance claims and loss history associated with homes, auto or personal property" and will provide one free report every 12 months on request. Verisk's FCRA page lists the A-PLUS Consumer Inquiry Center for requests.

I did not find a consumer request route for ClaimSearch itself. Whether a given database or report is a "consumer report" under federal law is a legal question, and a licensed attorney is the person to ask about yours. In practice, the CLUE and A-PLUS reports are the ones you can pull, and they are covered in Pulling your own records: CLUE, MVR and claim file and Your CLUE report and motor vehicle record explained. How a claim ends up on your report and what it does to a renewal is in How a claim reaches your CLUE report and your renewal price.

The distinction explains a frustrating pattern. A flag can slow a claim today and never show up on any report you can request. Conversely, something on your CLUE report can raise a renewal quote even if no one suspected anything.

Privacy and fair-claims rules around the databases

Three layers of rules apply, and none is a blanket shield for either side.

Sharing for fraud prevention is permitted. The federal Gramm-Leach-Bliley Act restricts how financial institutions, which include insurers, disclose nonpublic personal information. But its general exceptions allow disclosure "to protect against or prevent actual or potential fraud, unauthorized transactions, claims, or other liability," and to "insurance rate advisory organizations" (15 U.S.C. 6802(e)). That is part of the legal basis for contributing claims to shared systems. States add their own insurance-privacy rules and fraud-reporting immunity laws, which differ; see mandatory fraud reporting and immunity laws for insurers.

Fair-claims laws still apply, whatever tool produced the flag. The NAIC's Model Bulletin on the use of artificial intelligence systems by insurers, adopted December 4, 2023, says that AI is used across the insurance life cycle, including "claim management, and fraud detection," and that insurers' actions "must not violate" unfair trade practices or unfair claims settlement practices laws "regardless of the methods the Insurer used" (NAIC Model Bulletin). It also expects insurers to govern third-party vendor models and data. A model bulletin is guidance that binds only in states that adopt it, so the NAIC's implementation map and your state insurance department tell you whether your state has.

An insurer cannot deny a claim merely because a score was high. A denial needs a basis in the policy or in evidence. That is the logic behind rules like California's requirement of a fair and objective investigation. What a flag legitimately triggers is more scrutiny.

Common questions

Can I see what ClaimSearch says about me?

I could not find a consumer disclosure process for ClaimSearch. You can request your CLUE report from LexisNexis and your A-PLUS report from Verisk, and you can ask your insurer about its claim file. Rules about access to the claim file vary by state.

Does a flag mean the insurer thinks I lied?

No. A flag means a rule or model found something worth a human look. The match report may list nothing but prior claims that were entirely proper.

Will a flagged claim end up on my CLUE report?

A claim reported by your insurer can appear on a CLUE report as a claim regardless of any fraud review, and its outcome (paid, closed without payment) is shown as the insurer reports it. The review itself is not a separate mark on the report that I could confirm.

Do these systems catch most fraud?

I have no figure I can verify for how much fraud they catch or miss. The NAIC points out that insurers increasingly use predictive modeling, link analysis and artificial intelligence, and that 42 states and the District of Columbia operate insurance fraud bureaus (NAIC insurance fraud page, updated June 2026). Treat any single detection-rate claim with caution.

What if I really was involved in a prior claim and did not mention it?

Tell the truth, and fix the record now. A match for a prior claim you did not disclose can look like concealment, even if you forgot. Correcting an honest omission early is far better than being caught on it later.

Key takeaways

• ISO ClaimSearch is an insurer-facing industry database; Verisk said in its timeline that it passed 1.8 billion claims in 2024, and its 2022 brochure put the share of property and casualty insurers using it above 90 percent.
• Matches are routine (Verisk reported 5.6 per claim in 2022), so what matters is the pattern, not the presence of a match.
• Link analysis and scoring models route claims to people. A flag is a prompt for review, not evidence, and NICB itself says its indicators "are not actual evidence."
• Honest claimants are flagged by shared addresses, prior claims, new policies and data errors. Consistent answers, early explanations and written follow-up are the best response.
• ClaimSearch is not CLUE. CLUE Auto and A-PLUS are the underwriting records you can request for free; fair-claims laws apply to any claim, whatever tool raised the flag.