---
title: "Cross-carrier fraud data networks (IDN, ISO ClaimSearch, NICB): what they catch and what they miss"
description: "Cross-carrier fraud data networks - ISO ClaimSearch, NICB, and Shift IDN - answer one question: have we seen this claim before? A match is a lead, not a verdict. What they catch, what they structurally miss, and the investigation layer they leave open."
date: "2026-07-20"
lastModified: "2026-11-16"
author: "Nitish Badu"
tags: ["Guides"]
canonical: "https://gethesperai.com/blog/cross-carrier-fraud-data-networks-limits/"
---

# Cross-carrier fraud data networks (IDN, ISO ClaimSearch, NICB): what they catch and what they miss

> **TL;DR** A cross-carrier fraud data network matches an incoming claim against the industry's shared claim history to answer one question: have we seen this before? That match is a lead, not a verdict. ISO ClaimSearch, NICB, and Shift IDN are strong at anything with a prior record, but a match engine cannot see first-time or synthetic fraud, and it never investigates the flag it raises. The flag still lands on a human SIU investigator - and that is the layer, not the data, that caps how much fraud a carrier actually resolves.
>
> - ISO ClaimSearch holds 1.8B+ records across ~95% of US P&C
> - A network answers 'seen this before?' - not 'is this fraud?'
> - Synthetic identity fraud grew from $8B to $30B+ (2020-2025)
> - Manual SIU resolves only ~25% of the flags a network raises

- **1.8B+** - Records in ISO ClaimSearch (~95% of the US P&C market, 2,800+ contributors (Verisk))
- **$308.6B** - Annual US insurance fraud (~10% of P&C losses (Coalition Against Insurance Fraud))
- **$8B → $30B+** - Synthetic identity fraud, 2020-2025 (The fraud a match engine cannot see (Claims Journal / RGA))
- **~25% → 100%** - Flagged-claim coverage (Manual SIU vs Hesper on the flags a network raises (Hesper benchmark))

A cross-carrier fraud data network answers exactly one question: have we seen this claim, person, vehicle, or provider before? That match is a lead, not a verdict. ISO ClaimSearch, NICB, and Shift's IDN surface anything with a prior record. None of them investigates the claim they flag.

The two questions get conflated in procurement. A match engine tells you a claim resembles others in the system. It does not tell you whether this specific claim is fraudulent, with documented evidence a state DOI examiner or a court would accept. For the wider three-layer model, see [prevention vs. detection vs. investigation](/blog/insurance-fraud-prevention-vs-detection-vs-investigation/).

## What a cross-carrier fraud data network is

A cross-carrier fraud data network is a shared or consortium database that checks an incoming insurance claim against claim history contributed by other carriers, then surfaces matches. It is a matching engine. Three networks run the US P&C market.

ISO ClaimSearch, operated by Verisk, is the largest. Per [Verisk's own description of ClaimSearch as the backbone of the P&C claims ecosystem](https://www.verisk.com/blog/verisk-claimsearch-the-backbone-of-the-pc-claims-ecosystem/), it holds over 1.8 billion US claims records, represents roughly 95% of the US P&C market, draws from 2,800-plus contributors, and is queried by around 200,000 claims professionals every day.

NICB, the National Insurance Crime Bureau, is a non-profit consortium, not a commercial database. Per [public NICB reference material](https://en.wikipedia.org/wiki/National_Insurance_Crime_Bureau), it is supported by more than 1,100 property-casualty and self-insured companies. Its member materials describe coverage of roughly 96% of personal-auto and about 82% of all P&C premiums. Its value is referrals, ring intelligence, and law-enforcement liaison.

Shift Technology's IDN (Insurance Data Network) is the newest of the three - a cross-carrier intelligence network that surfaces connections across participating carriers' claims. All three occupy the detection and matching layer. The investigation layer beneath them is still manual SIU.

**Figure 1 (The matching layer) - Three networks, one question: have we seen this before?** Each checks an incoming claim against shared claim history and returns a match. None of them works the match.

**The scale of shared history**

- **1.8B+** - US claims records in ISO ClaimSearch
- **~95%** - of the US P&C market represented in ClaimSearch
- **2,800+** - contributors to ClaimSearch
- **~200,000** - claims professionals query ClaimSearch every day
- **1,100+** - property-casualty and self-insured companies support NICB
- **~96%** - of personal-auto premiums covered by NICB members

**What a match returns, and where it goes**

1. Incoming claim matched against shared claim history
2. Prior-loss, duplicate or watchlist hit returned
3. The flag lands on a human SIU investigator's desk

**Takeaway:** All three occupy the detection layer. None occupies the investigation layer beneath them.

Sources: Verisk's description of ISO ClaimSearch (records, market share, contributors, daily users); NICB public reference and member materials (members, personal-auto premium coverage). Figures as cited in this article, July 2026.

## What these networks catch well

Cross-carrier networks are strongest at anything with a prior record. Where a claim resembles something already in the system, a match engine is fast, cheap, and hard to beat. They reliably catch:

- Duplicate and double-dipped claims filed across carriers
- Prior-loss history
- Salvage and theft records
- Hits against known fraud rings and provider watchlists

The scale is the point. A carrier querying ClaimSearch checks a single claim against 1.8 billion records from 95% of the market in seconds. No manual process reaches that, which is why every serious SIU runs these queries daily.

NICB's vehicle-crime intelligence shows the category done well. VINCheck lets an investigator confirm whether a vehicle was reported stolen or declared salvage before a claim is paid. And one carrier's questionable-claim referral becomes intelligence the whole membership can use.

> **Detection and investigation are different questions**
>
> A network answers 'have we seen this claim, person, VIN, or provider before?' Investigation answers 'is this specific claim fraudulent, and here is the documented evidence.' A network is excellent at the first and does not attempt the second. Treating a match report as a resolved case is the error.

## What they structurally miss

The limits of a cross-carrier network are structural, not a data-quality problem you can fix with more contributors. A match engine can only recognize what already exists in shared history. That single design fact produces four gaps.

### They flag; they do not investigate

A match is where the work starts, not where it ends. Someone still has to run the investigation and produce a documented finding. Manual SIU investigation takes 14+ days per case, and one investigator handles 200+ cases. Across US P&C carriers, only about 25% of flagged claims are fully investigated. The coverage math is in [why most flagged insurance claims are never fully investigated](/blog/why-flagged-insurance-claims-never-investigated/).

### Coverage gaps and data latency

A contributory network only sees what its members contribute. It is blind to anything filed with a carrier that does not participate. Refresh cadence also adds latency: a claim filed elsewhere last week may not yet be matchable this week. And a match is only as good as the record it matches against.

### Novel and first-time fraud with no prior match

This is the gap that is growing fastest. First-time claims, synthetic identities, and AI-generated photos or documents have no prior record, so they pass the network clean. Per [Claims Journal, citing Reinsurance Group of America](https://www.claimsjournal.com/news/national/2026/05/15/337321.htm), synthetic identity fraud grew from roughly $8 billion in 2020 to more than $30 billion by mid-2025.

### Organized fraud that stays below the match threshold

Sophisticated rings understand how matching works and engineer around it. By varying identities, addresses, phone numbers, and treating providers just enough, a ring can keep any single claim below the hard-match threshold. A fuzzy near-match may still surface, but it is a lower-confidence lead a human has to confirm - right back into the same capacity wall.

*Figure: A cross-carrier network answers 'seen this before?' and stops. First-time fraud, synthetic identities, and AI-generated evidence have no prior record to match, so they pass clean - and every real match still lands on a human investigator who can work only about a quarter of the flags. The gap is not data quality; it is the investigation layer beneath the flag.*

## Detection network vs investigation layer

The network's job and the investigation layer's job are different questions. A network answers whether a claim has been seen before and returns a match instantly. The investigation layer answers whether a claim is fraudulent and returns documented evidence. Buying more matching capacity does not add investigation capacity.

| Question | Cross-carrier data network | Investigation layer (Hesper) |
| --- | --- | --- |
| Have we seen this claim / person / VIN / provider before? | Yes - core function (1.8B records, ~95% of market) | Uses the match as one input among 15+ phases |
| Is THIS claim fraudulent, with documented evidence? | No - raises a lead only | Yes - audit-ready finding per claim |
| Time to a worked answer | Instant match, then 14+ days manual to resolve | Minutes per case |
| Coverage of flagged claims actually resolved | ~25% (human capacity limit) | 100% |
| Catches first-time / no-prior-record fraud | No - nothing to match against | Yes - evidence-based, not match-based |
| Cost per investigated case | Match is cheap; working it manually is expensive | A fraction of the manual cost |

The left column is where a match report leaves you. The right column is where a case has to end up before it is defensible. The distance between them is the investigation. The same limit shows up in each of the three networks.

| Network | Operator | Model | Strongest at | Structural limit |
| --- | --- | --- | --- | --- |
| ISO ClaimSearch | Verisk | Contributory database, ~95% P&C, 1.8B records | Duplicate and prior-loss matching at scale | Flags; does not investigate |
| NICB | Non-profit consortium (1,100+ members) | Referrals, VIN/theft data, ring intel | Vehicle theft, organized rings, LE liaison | Referral lands on a human's desk |
| Shift IDN | Shift Technology | Cross-carrier intelligence network | Surfacing cross-carrier connections | Per-claim depth still manual |

Each of these is essential and none is a competitor to investigation. They share one structural limit: they raise a flag, and the flag has to be worked by hand. For the contrast against the vendor behind ClaimSearch, see [Hesper AI vs. Verisk](/blog/hesper-vs-verisk/).

## A better network raises more flags - which widens the gap without an investigation layer

A better detection network surfaces more matches, which means a larger flag pile. If investigation throughput is fixed at about 25% coverage, more and better flags produce more unworked leads, not less loss. This is the direct read on Shift IDN's 2026 expansion: real progress on detection, but the bottleneck moved downstream years ago.

Per the [Coalition Against Insurance Fraud](https://insurancefraud.org/fraud-stats/), insurance fraud steals at least $308.6 billion every year from American consumers. Per the [Insurance Information Institute](https://www.iii.org/article/background-on-insurance-fraud), citing the same study, that includes roughly $74.7 billion in life insurance, $45 billion in property-casualty, $34 billion in workers' compensation, and $7.4 billion in auto theft. Fraud is distributed across lines, not concentrated where a match engine is strongest.

**Figure 2 (Where the fraud sits) - Fraud is spread across lines, not pooled where matching is strongest.** Four lines inside the $308.6 billion a year: the slice that VIN, theft and salvage records see best is the smallest.

**Inside $308.6B a year, by line**

- Life insurance: $74.7B
- Property-casualty: $45B
- Workers' compensation: $34B
- Auto theft: **$7.4B** - Where VIN, theft and salvage matching is strongest

**More flags, same investigation capacity**

1. A better network surfaces more good matches
2. Investigation coverage stays where it was
3. The extra flags are paid, denied without full work, or queued

**Takeaway:** The loss-ratio lever is the share of flags resolved, not the quality of the match.

Sources: Coalition Against Insurance Fraud (at least $308.6B a year from American consumers); Insurance Information Institute, citing the same study, for the breakdown by line. The lines shown are not exhaustive and do not sum to the total.

For a claims VP evaluating the stack, the loss-ratio lever is not the quality of the match. It is the share of matches that get resolved. If detection improves while investigation coverage stays at 25%, the extra flags are paid, denied without full work, or queued. That is a downstream capacity failure, not a detection failure.

> The bottleneck in fraud stopped being detection years ago. Carriers can already flag far more suspicious claims than they can investigate. Adding a better matching network raises the number of leads; it does not raise the number of leads that become resolved cases. The constraint is investigation throughput, and no amount of additional matching relieves it.
>
> - Hesper AI product research

| Flagged-claim coverage: manual SIU vs an investigation layer (Hesper internal benchmark) | Value | Share |
| --- | --- | --- |
| Manual SIU coverage of flags | ~25% | 25% |
| Investigation-layer coverage | 100% | 100% |

## Where an AI investigation layer fits

An AI investigation layer sits directly downstream of the networks: the network raises a flag, and the investigation layer resolves that flag end-to-end with an audit-ready file. It is complementary to ISO ClaimSearch, NICB, and Shift IDN - not a replacement for any of them - and it can also run standalone.

Hesper AI takes a flagged claim - from a ClaimSearch match, an NICB referral, an IDN connection, or a FRISS or Shift score - and runs 15+ investigation phases in parallel on it. It completes a case in minutes rather than 14+ days. Because every flagged claim can be worked, coverage moves from about 25% to 100% (Hesper internal benchmark).

**Figure 3 (Downstream of the flag) - The network raises the flag. The investigation layer works it.** A flag from any source becomes one evidence file; the SIU lead keeps the finding.

Input: A flagged claim, from any source (ClaimSearch_match.pdf, NICB_referral.pdf, IDN_connection.pdf)

**Agent legwork** (Logged with sources, reasoning, timestamps)

1. Document forensics on every submitted document
2. Statement and EUO cross-reference
3. Timeline reconstruction
4. Financial pattern and network analysis
5. Case narrative and report assembly

15+ investigation phases in parallel

Hand-off: Evidence file

**SIU lead decides** (Licensed human, every decision)

- The fraud finding
- Referral, denial or payment
- NICB or law-enforcement handoff

Agent produces evidence. Human adjudicates.

|  | Time to a worked answer | Caseload ceiling |
| --- | --- | --- |
| Manual baseline (Hesper internal benchmark) | 14+ days per case | 200+ cases per investigator |
| With agents | minutes, not weeks | Every flagged claim worked |

Source: Illustrative decision-rights map. Agent and human roles per Hesper product scope. Benchmark strip: Hesper internal benchmarks, not industry figures.

(Illustrative example)

This also covers what a network cannot. An investigation layer works from the actual documents, metadata, and story of the claim, so it does not depend on a match existing at all. It catches first-time and synthetic fraud not because it has more data, but because it asks a different question about each claim.

For a compliance officer, the defensibility is the reason this layer clears review. Every phase Hesper runs is logged with its sources, reasoning, and timestamps, so each finding is produced as an audit-trail-native record that satisfies California 10 CCR 2698.36's documented-decision requirement and the antifraud-plan obligations under NAIC Model Act 680. The agent produces the evidence, the human adjudicates. The layered model is covered in [prevention vs. detection vs. investigation](/blog/insurance-fraud-prevention-vs-detection-vs-investigation/).

## Key takeaways

- A cross-carrier fraud data network answers one question - have we seen this claim, person, VIN, or provider before - and that match is a lead, not a verdict.
- These networks catch anything with a prior record extremely well: ClaimSearch matches a claim against 1.8 billion records from about 95% of the US P&C market.
- The limits are structural: a match engine is blind to first-time and synthetic fraud with no prior record - a category that grew from about $8 billion in 2020 to more than $30 billion by mid-2025.
- A better network raises more flags. If investigation coverage stays at about 25%, the loss-ratio lever is the share of flags resolved, not the quality of the match.
- An AI investigation layer is complementary to these networks and standalone-capable: it works each flagged claim in minutes instead of 14+ days, lifting flagged-claim coverage from about 25% to 100%.

## Frequently asked questions

### What is a cross-carrier fraud data network?

A cross-carrier fraud data network is a shared or consortium database that checks an incoming insurance claim against claim history from other carriers to surface matches - duplicate claims, prior losses, known fraud rings, or flagged vehicles and providers. The three most common in US P&C are ISO ClaimSearch (Verisk), NICB, and Shift Technology's IDN. That match is a lead. A human investigator still works the flag downstream.

### What does ISO ClaimSearch catch, and what does it miss?

ISO ClaimSearch is strongest at anything with a prior record: duplicate or double-dipped claims filed across carriers, prior-loss history, salvage and theft records, and matches to known fraud rings and provider watchlists. It cannot see fraud in non-contributing carriers, it lags on data refresh, and it cannot flag first-time or synthetic fraud that has no prior record. It also stops at the flag.

### If a cross-carrier network flags a claim, why isn't the fraud caught?

Because a flag is where the work starts, not where it ends. Someone still has to run the investigation and produce a documented finding. Manual SIU investigation takes 14+ days per case, and US carriers fully investigate only about 25% of flagged claims. If investigation throughput is fixed, more flags means more unworked leads - not less loss.

### Can a cross-carrier data network detect new or AI-generated fraud?

Not well, by design. A match-based network can only recognize what already exists in shared history. First-time fraud, synthetic identities, and AI-generated claim photos or documents have no prior record to match against, so they pass the network clean. Catching novel fraud requires evidence-based investigation - examining the actual documents, metadata, and story of the claim.

### Does Hesper AI replace ISO ClaimSearch, NICB, or Shift IDN?

No. Hesper sits downstream of them and is complementary, not a replacement. Cross-carrier networks are the industry's detection and matching layer. Hesper occupies the investigation layer beneath them, taking a flagged claim and resolving it end-to-end. A carrier keeps its ClaimSearch contribution, NICB membership, and any intelligence network, and adds Hesper to investigate the flags those systems raise.
