---
title: "Top 10 insurance fraud detection platforms in 2026"
description: "A neutral 2026 buyer's roundup of the 10 insurance fraud detection platforms most commonly on US P&C carrier shortlists, with where each wins and where each falls short."
date: "2026-05-13"
lastModified: "2026-11-14"
author: "Nitish Badu"
tags: ["Vendor Comparison"]
canonical: "https://gethesperai.com/blog/top-fraud-detection-platforms-2026/"
---

# Top 10 insurance fraud detection platforms in 2026

> **TL;DR** Verisk, FRISS, and Shift Technology anchor the 2026 detection shortlist; the other seven platforms cover narrower segments. Every vendor on the list flags suspicious claims and hands them to a human investigator. None of them runs the per-claim SIU workflow downstream.
>
> - Verisk's industry-utility data depth (1.8B records, ~95% of US P&C market) is the largest moat in detection and the safest procurement anchor.
> - Shift Technology's move into handler-assist agentic AI is the most consequential category shift since 2024, but it stops at handler assistance.
> - No platform in the top 10 occupies the investigation layer downstream of the flag - that layer is still manual SIU at 14+ days per case.
> - Vendor accuracy claims are not comparable with each other. Deloitte's published baseline - 20-40% detection on soft fraud, 40-80% on hard fraud - is the only industry anchor worth taking into an RFP.

- **$308B** - Annual US insurance fraud loss (Coalition Against Insurance Fraud)
- **~10%** - Of P&C claims involve fraud (Industry baseline)
- **60-85%** - Rules-based false positive rate (Pre-ML detection systems)
- **1.8B** - Records in ISO ClaimSearch (~95% of US P&C market)

The 2026 detection category looks different from the one carriers were buying into in 2023. Three forces have reshaped it: a maturing cross-carrier data utility layer that Verisk continues to compound, the arrival of handler-assist agentic AI from Shift Technology, and a mid-market scoring tier where FRISS and a handful of newer entrants compete on deployment speed. Underneath all three is the same loss-cost math. US insurers lose roughly $308 billion to fraud each year per the [Coalition Against Insurance Fraud](https://insurancefraud.org), and fraud occurs in about 10% of property-casualty insurance losses per the [Coalition's fraud statistics](https://insurancefraud.org/fraud-stats/). This guide ranks the vendors; the methods underneath them - rules, statistical scoring, network analysis, autonomous AI - are mapped in the [complete guide to insurance fraud detection](/blog/insurance-fraud-detection-pillar/).

This is a buyer's-shortlist roundup, not a benchmark. We synthesized public product pages, named carrier deployments, and regulator filings to set the order. We did not run head-to-head precision-recall tests on flagged claims; no public dataset supports that. Hesper AI is not in the top 10. Hesper sits at the investigation layer downstream of detection - a different category from every vendor below - and we cover it separately at the end of the post.

One note on terminology, because it decides the shortlist. Carriers search for insurance fraud detection software as though it names a single product category. The vendors that come back span four: cross-carrier claims data utilities, claim-scoring engines, graph and entity-resolution platforms, and enterprise financial-crime suites built for banking that carry an insurance module on the side. Those four are not substitutes for one another. Put them on one list without marking which lane each sits in and you get the most common procurement error in this category - two contracts solving the same half of the problem and nothing covering the other half. The 10 below are the P&C claims-fraud set; the wider software set further down covers the other lanes and what each one actually displaces.

If you want the full category map across prevention, detection, and investigation, read [the full Pillar 5 comparison of AI fraud platforms](/blog/ai-fraud-platforms-compared-2026-pillar/). This post focuses on the detection layer specifically: who is on most US P&C shortlists in 2026, what each one is actually good at, and where each one stops.

## The 2026 fraud detection landscape

**What does the insurance fraud detection vendor landscape look like in 2026?**

Three layers. A cross-carrier data utility, dominated by Verisk's ISO ClaimSearch at 1.8B records and roughly 95% of the US P&C market. A scoring tier: Verisk ClaimDirector, FRISS, Shift Technology, BAE NetReveal and SAS. And a handler-assist agentic AI tier, opened by Shift Claims in September 2025. Fraud sits in about 10% of P&C losses.

Detection has matured into a layered category. The cross-carrier data utility - what a single carrier cannot see on its own - is dominated by Verisk's ISO ClaimSearch, which holds 1.8B records and serves around 95% of the US P&C market per its [product page](https://www.verisk.com/products/claimsearch/), with 175K daily ingest and 200K daily users. Above that data utility sits a scoring tier: Verisk ClaimDirector with its 0-999 methodology, FRISS, Shift Technology, BAE Systems NetReveal, and SAS Fraud Framework. The frontier in 2026 is agentic AI for handlers - Shift Claims pushed the category here in 2025, and the rest of the field is reacting.

The cost of stillness in this category is high. Rules-only detection systems run 60-85% false positive rates, which jams the SIU queue with claims that should never have been flagged. The carriers winning here are not the ones with the most rules. They are the ones with the deepest contributory data and the cleanest hand-off into the case workflow downstream. That is why Verisk's [Q4 and FY2024 earnings](https://www.verisk.com/company/newsroom/verisk-reports-fourth-quarter-2024-and-full-year-2024-financial-results/) show $2,882M in full-year revenue with the claims sub-segment growing 13% year over year in Q4. The data network compounds; the moat widens.

> **What changed since 2024**
>
> Three shifts define the 2026 evaluation: Shift Technology released Shift Claims (handler-assist agentic AI for adjusters), Verisk continued compounding its data-network advantage with FY2024 revenue of $2,882M, and a new procurement category opened up at the investigation layer downstream of the flag. Carriers who last refreshed in 2023 will see a meaningfully different shortlist today. The consolidation has continued since: [Verisk's Fraud Discovery platform unified five layers of the fraud stack](/blog/verisk-fraud-discovery-vs-autonomous-investigation/) in September 2026, and the investigation itself is still the layer no vendor on this list runs.

## How we ranked the 10

Six criteria, equally weighted. We did not collapse them into a single score because the procurement-relevant trade-offs are between criteria, not within them. A vendor that wins on data depth but loses on deployment speed is a different decision from one that wins on deployment speed but loses on data depth.

1. Detection accuracy: precision and recall on real flagged claims, as reported by the vendor and triangulated against named deployments.
2. Data network depth: how much cross-carrier contributory data the platform draws on. Verisk's ISO ClaimSearch is the bar at 1.8B records.
3. Integration footprint: native connectors to Guidewire ClaimCenter, Duck Creek Claims, Snapsheet, and custom claims systems.
4. Regulatory posture: alignment with NAIC Model Act 680 antifraud-plan filings (adopted in 48 states) and state DOI audit defensibility.
5. Deployment cycle: weeks for cloud-native scoring vendors, quarters or longer for enterprise platforms.
6. Pricing transparency: most are subscription with custom quotes; we flag the few that publish, and the few that bundle into broader claims-tech contracts.

Order in the list below tracks usage depth in US P&C, not alphabetical order and not our preference. Verisk is first because its data utility sits underneath almost every other vendor on the list.

## The 10 platforms

**Which fraud detection platforms should a US P&C carrier shortlist in 2026?**

Verisk, FRISS and Shift Technology carry most US P&C shortlists. BAE NetReveal for organized rings, SAS for carriers already on Viya, CCC for auto-heavy books. Snapsheet, Tractable, Akur8 and Lemonade appear on shortlists but sit adjacent to detection - FNOL workflow, computer vision, pricing AI and a carrier-built stack respectively.

### 1. Verisk (ISO ClaimSearch + ClaimDirector + Xactware)

Verisk Analytics (NASDAQ: VRSK, parent founded 1971) operates the industry's deepest claims data utility. ISO ClaimSearch holds [1.8B claim records covering ~95% of the US P&C market](https://www.verisk.com/products/claimsearch/), ingesting 175K records daily across 200K users. ClaimDirector layers a 0-999 fraud-scoring methodology on top of that data, and [Verisk's own Xactware release](https://www.stocktitan.net/news/VRSK/property-losses-topped-68-billion-in-2020-as-extreme-events-and-uavgd2h1jaxd.html) puts Xactware estimation solutions, Xactimate among them, in 22 of the top 25 US property insurers. The NICB has operated through ISO databases since 1998. Verisk reported $2,882M in FY2024 revenue with a +13% Q4 claims sub-segment.

Where they're weaker: ClaimDirector flags suspicious claims; it does not run per-claim investigation. The SIU still owns the 14+ day manual workflow downstream of the score. Procurement is also the slower path - public-company procurement cycles and custom-quoted subscriptions take quarters, not weeks. For a deeper head-to-head, see [our Verisk comparison](/blog/hesper-vs-verisk/).

### 2. FRISS

FRISS (founded 2006, Netherlands) is the mid-market detection vendor most US carriers shortlist alongside Verisk. The product covers FNOL scoring and claim-lifecycle scoring with a workflow-for-humans interface that SIU directors consistently rate as easier to live with than enterprise platforms. Auto is the strongest line. The 2024 Capgemini investment moved FRISS further into the European mid-market, and its US footprint has grown through Guidewire ClaimCenter integrations.

Where they're weaker: the cross-carrier data network is smaller than Verisk's, so the contributory-data signal is thinner. And like every detection vendor on this list, FRISS flags and hands off; the per-claim investigation is still manual. Pricing is subscription, custom-quoted, with mid-market deals typically in the $300K to $1.5M annual range based on volume.

### 3. Shift Technology

Shift Technology (founded 2013, Paris) is the most active detection vendor in 2026. Their [AXA Switzerland deployment](https://www.shift-technology.com/resources/case-studies/axa-switzerland-insurance-fraud-detection-success) processed over 1M claims and stopped more than €12M in fraud, with Samuel Klaus on record as the program lead. Shift Claims (launched September 16, 2025) is the handler-assist agentic AI tier; per the [launch release](https://www.prnewswire.com/news-releases/shift-technology-launches-shift-claims-to-power-claims-transformation-with-agentic-ai-302557099.html), early adopters report 3% lower claims losses, 30% faster claims handling, a 60% overall automation rate, and greater than 99% accuracy in claims assessment. The Shift IDN cross-carrier network is the cleanest non-Verisk contributory-data play in the field.

> Using Shift Claims Fraud Detection, we are able to consistently identify suspicious activities at FNOL and assign the claim to the appropriate expert for investigation.
>
> - Samuel Klaus, Head of Fraud at AXA Switzerland ([Shift Technology case study](https://www.shift-technology.com/resources/case-studies/axa-switzerland-insurance-fraud-detection-success))

Where they're weaker: Shift Claims is handler-assist, not autonomous end-to-end investigation. The 99% accuracy figure refers to assisted decisions surfaced to a human, not autonomous resolution of a flagged claim. US footprint is thinner than the European one; the strongest named deployments are AXA Switzerland and a handful of European mutuals. The layer distinction is taken apart in full in [Shift Claims vs. Hesper: agentic AI for claims handling vs. autonomous investigation](/blog/shift-claims-agentic-ai-vs-hesper/).

### 4. BAE Systems NetReveal

BAE Systems NetReveal (lineage from 1999) carries financial-crime DNA into insurance. Its strength is network analysis and ring detection - the platform was built for AML before it was tuned for insurance, and that lineage shows up in the graph models. For organized rings in workers comp medical billing or staged-collision auto fraud, NetReveal's ring detection is the strongest match in this roundup, drawing on its AML-graph lineage.

Where they're weaker: enterprise sales cycle, slower deployment, and an interface that SIU directors usually describe as built for analysts rather than investigators. The platform is contracted as an enterprise license, not subscription, so the cost curve looks different from the rest of the field. And the same per-claim investigation gap applies: NetReveal surfaces ring connections; the investigator still does the case work.

### 5. SAS Fraud Framework

SAS Fraud Framework (SAS parent founded 1976) is the rules-plus-ML scoring tier inside the SAS Viya analytics platform. It wins inside carriers that already run SAS Viya as the corporate analytics backbone - the integration cost is near zero, and the rules engine is among the most customizable in the field. Strong fit for technical-buyer audiences with in-house data-science capacity.

Where they're weaker: heavy lift for carriers not already on SAS Viya. Deployment requires SAS-skilled headcount and a multi-quarter implementation cycle. Investigation depth is shallow - same flag-and-hand-off pattern as the rest of the detection tier.

### 6. CCC Intelligent Solutions

CCC Intelligent Solutions (founded 1980, IPO 2021) anchors the auto-claims data network in the US, with deep integrations into OEMs and the repair-shop economy. Detection is part of a broader claims-tech bundle that includes estimating, photo AI, and subrogation. For carriers whose auto book drives the loss ratio, CCC's data network is the most relevant non-Verisk option.

Where they're weaker: CCC is auto-only in any meaningful sense. Property, workers comp, and liability carriers will not find a fit here. And the bundle pricing makes apples-to-apples comparison against pure detection vendors hard - the detection capability is rarely procured stand-alone.

### 7. Snapsheet

Snapsheet (founded 2011, Chicago) is mobile-first FNOL and digital triage; the [platform](https://www.snapsheetclaims.com/) has 170+ customers, runs in 16 of the top 20 US P&C carriers in some capacity, and has processed roughly $6B in appraisals. Detection signals are generated as a byproduct of the digital intake workflow - the platform catches inconsistencies in photo evidence, geolocation, and timeline at FNOL before a claim ever reaches an adjuster.

Where they're weaker: Snapsheet is not a standalone fraud detection platform. It is a claims workflow vendor whose digital intake surfaces some fraud signals as a side effect. Carriers shortlisting Snapsheet for detection alone are evaluating the wrong product; the right framing is FNOL workflow plus a downstream scoring vendor.

### 8. Tractable

Tractable (founded 2014, London) applies computer vision to auto damage assessment and property estimation. Tokio Marine's deployment was reported by [Insurance Journal in April 2020](https://www.insurancejournal.com/news/international/2020/05/11/568090.htm), and the platform now serves auto carriers across Europe, Japan, and the US. Cycle time on visual damage assessment drops from days to minutes for the claim types Tractable's vision models handle.

Where they're weaker: Tractable is not strictly a fraud detection platform. Their computer vision processes legitimate damage assessment; fraud detection is not the layer they sit at. Including Tractable here is a buyer's reality check - it shows up on shortlists because carriers conflate "AI for claims" with "AI for fraud," and the two are different problems with different vendors.

### 9. Akur8

Akur8 (founded 2018, Paris) builds AI for pricing and underwriting, not detection. The platform automates rate-making and risk segmentation at underwriting, which has the effect of preventing some classes of fraud before a claim is filed. Carriers ask about Akur8 in fraud RFPs because the pricing AI conversation has spilled into the detection conversation; the two share "AI" in the headline but solve different problems.

Where they're weaker: Akur8 does not score or investigate claims. If the procurement question is detection, Akur8 is not the answer. We include it because it appears on shortlists, not because it competes head-to-head with the rest of this list.

### 10. Lemonade (illustrative, not a vendor)

Lemonade (founded 2015, public 2020) is a carrier with carrier-built AI, not a vendor selling to other carriers. We include it as a reference point for what carrier-internal fraud AI looks like at scale in 2026. Lemonade's claims-handling AI flags suspicious claims as part of the integrated underwriting-to-claims stack, and the company has published partial details on detection methodology in shareholder letters.

Where they're weaker: not sold to peers. Carriers cannot buy Lemonade's stack. We include it so the shortlist reflects the actual procurement reality: a few large carriers are building, and they are not in the market to buy. The carriers who do buy are evaluating the nine vendors above.

## Side-by-side comparison

**How do the top insurance fraud detection vendors compare side by side?**

On cross-carrier contributory data, only Verisk (industry utility), Shift (IDN) and CCC (auto network) qualify; FRISS is vendor-held and limited, SAS has none. Seven of the ten price as subscription, BAE and SAS as enterprise contracts, and Lemonade is not sold to peers at all. All ten stop at the flag.

The matrix below collapses the profiles to five fields most procurement teams care about: parent-company founding year, tier orientation, whether the platform draws on cross-carrier contributory data, and pricing transparency. Tier is for orientation only - it is not a ranking.

| Vendor | Founded | Tier | Cross-carrier data | Pricing transparency |
| --- | --- | --- | --- | --- |
| Verisk (ClaimSearch + ClaimDirector) | 1971 | Tier 1 incumbent | Yes (industry utility) | Subscription, custom-quoted |
| FRISS | 2006 | Tier 1 detection | Limited (vendor-held) | Subscription, custom-quoted |
| Shift Technology | 2013 | Tier 1 detection | Yes (Shift IDN) | Subscription, custom-quoted |
| BAE Systems NetReveal | 1999 | Tier 1 enterprise | Limited | Enterprise contract |
| SAS Fraud Framework | 1976 | Tier 1 platform-bundled | No | Enterprise license |
| CCC Intelligent Solutions | 1980 | Specialized (auto) | Yes (auto network) | Bundled, custom-quoted |
| Snapsheet | 2011 | Digital-first FNOL | No (workflow-driven) | Subscription |
| Tractable | 2014 | Adjacent (vision) | No | Subscription |
| Akur8 | 2018 | Adjacent (pricing AI) | No | Subscription |
| Lemonade | 2015 | Carrier-built | N/A (carrier) | N/A |

| Coverage of flagged claims: detection vendors stop at the flag | Value | Share |
| --- | --- | --- |
| Flagged by detection vendor | 100% | 100% |
| Reach human SIU investigator | ~95% | 95% |
| Fully investigated within case-cycle window | ~25% | 25% |
| Resolved with audit-ready documentation | ~18% | 18% |

Every vendor in the top 10 contributes to the first bar. None of them contributes to the third or fourth. That is the gap the rest of this post is about.

## The wider insurance fraud detection software set

**What other insurance fraud detection software appears on 2026 shortlists?**

Quantexa, LexisNexis Risk Solutions, DataWalk, NICE Actimize, Thomson Reuters CLEAR, INFORM RiskShield, PwC Risk Detect and Feedzai all surface alongside the top 10. They sort into five lanes: claim scoring, data and identity, graph and entity resolution, enterprise financial crime, and investigation case management. Only the claim-scoring lane competes head-on with the 10 above. The rest are additive, or built for a different industry.

The 10 platforms above are the set US P&C carriers actually shortlist against each other. They are not the whole set of software that comes back when you search the category, and the difference matters before an RFP goes out. Below are the vendors that show up alongside them, sorted by lane rather than ranked, with what each one displaces in a claims operation that already exists. Two deserve flagging straight away. Feedzai is payments-first. And Featurespace has not been an independent vendor since Visa [completed its $946M acquisition in December 2024](https://www.visa.co.uk/about-visa/newsroom/press-releases.3361605.html) and folded it into Visa's Risk and Identity Solutions unit - it still appears on 2026 lists of the best insurance fraud detection software.

| Vendor | Lane | What it displaces | Insurance-native? |
| --- | --- | --- | --- |
| Quantexa | Graph + entity resolution | Manual link analysis across claimants, parties and suppliers | Yes (Claims Accelerator for ClaimCenter) |
| LexisNexis Risk Solutions | Data + identity | Third-party identity, phone and relationship lookups mid-claim | Yes (Claims Clarity, Accurint for Insurance) |
| DataWalk | Graph + investigation analytics | Spreadsheet-and-whiteboard ring analysis | Partly (insurance is one vertical) |
| NICE Actimize | Enterprise financial crime | Bank-grade detection plus case management across modalities | Partly (claims is one modality) |
| Thomson Reuters CLEAR | Data + investigator research | Public-records research inside an open SIU file | Partly (research tool, not a scorer) |
| INFORM RiskShield | Claim scoring | Rules-plus-model claim scoring at FNOL | Yes |
| PwC Risk Detect | Claim + policy analytics | A bespoke in-house analytics build | Partly (consulting-delivered) |
| Feedzai | Enterprise financial crime | Payments and transaction fraud detection | No (payments-first) |
| Featurespace (Visa) | Behavioural payments analytics | Transaction-stream anomaly detection | No (Visa Risk & Identity since Dec 2024) |
| Case IQ / CROSStrax / i-Sight | Investigation case management | Email-and-folder SIU case tracking | Partly (holds the file, does not score) |

The graph lane moved most in 2026. [Quantexa's Claims Accelerator landed on the Guidewire Marketplace on 23 January 2026](https://www.quantexa.com/press/quantexa-claims-accelerator-now-available-on-guidewire-marketplace-bringing-real-time-decision-intelligence-to-insurance-claims/), running entity resolution and contextual analytics inside ClaimCenter on Guidewire Cloud rather than as a separate console; Quantexa cites a Forrester study putting customer ROI at 228% over three years. FRISS reaches ClaimCenter the same way, through a [Claims Analytics accelerator built with Integration Frameworks for Guidewire Cloud](https://marketplace.guidewire.com/product/friss-accelerator-for-claims-analytics-for-claimcenter/01t3n00000T0F8mAAF). If your claims system is ClaimCenter, the marketplace listing is a faster read on integration reality than any vendor datasheet.

The data-and-identity lane is additive rather than competitive. [LexisNexis Risk Solutions ships Claims Clarity, Accurint for Insurance and ThreatMetrix for Insurance](https://risk.lexisnexis.com/insurance/fraud-detection), which validate parties and link relationships during a live claim; Thomson Reuters CLEAR does public-records research for an investigator who already has a file open. Neither scores claims. They feed the people and the systems that do. The case-management lane holds the file and the task list, which is also not detection - and, as the next-to-last section covers, also not the investigation.

## Detection rates: what the accuracy claims actually mean

**How do insurance fraud detection vendors compare on accuracy and false-positive rates?**

Not on their own numbers. Vendor pages quote figures like "up to 90% accuracy" and "false positives as low as 10%" with no dataset, baseline or study behind them. The only published industry baseline is Deloitte's: detection currently runs 20-40% on soft fraud and 40-80% on hard fraud. Compare vendors on referral precision measured against your own flag queue in a paid pilot, at the threshold you would actually run.

This is the question the long-tail searches on this topic keep asking, and the search results answer it badly. DataWalk's product page leads with [up to 90% accuracy on suspicious claims and false-positive rates as low as 10%](https://datawalk.com/insurance-fraud-detection-software/), with no study, dataset or customer attached to either figure. INFORM reports RiskShield taking one insurer's detection rate [from 1% to 3.6% within weeks](https://www.inform-software.com/en/solutions/risk-fraud/fraud-prevention/claims-fraud-detection), which is a far more credible shape of number because it is a lift against a stated baseline rather than an absolute accuracy score. The two cannot be compared, and neither is a benchmark.

There is a published baseline. Deloitte's P&C fraud analysis puts [current detection rates at 20% to 40% for soft fraud and 40% to 80% for hard fraud](https://www.deloitte.com/us/en/insights/industry/financial-services/financial-services-industry-predictions/2025/ai-to-fight-insurance-fraud.html), against $122B of annual P&C fraud loss, which Deloitte puts at 40% of the insurance industry's total fraud losses. Its forecast is that AI applied across the claims lifecycle could let P&C insurers avoid $80B to $160B in fraudulent claims by 2032. Those are the numbers to anchor an evaluation on, because they describe the problem rather than a vendor's self-report.

**Figure 1 (Detection rates) - Vendor accuracy is a claim. Deloitte's 20-40% and 40-80% are the baseline.** Only a published study, or a lift against a stated baseline, is a number worth carrying into an RFP.

**The accuracy numbers in circulation**

| Figure | Value | Basis |
| --- | --- | --- |
| Deloitte: detection rate, soft fraud | 20-40% | Published |
| Deloitte: detection rate, hard fraud | 40-80% | Published |
| INFORM RiskShield: one insurer's detection rate | 1% to 3.6% | vs a baseline |
| DataWalk: accuracy on suspicious claims | up to 90% | No study named |
| DataWalk: false-positive rate | as low as 10% | No study named |

*Neither DataWalk figure names a study, dataset or customer.*

**Takeaway:** Ask for referral precision on your own flag queue, in a paid pilot, at the threshold you would actually run.

Sources: Deloitte P&C fraud analysis (20-40% detection on soft fraud, 40-80% on hard fraud); INFORM on RiskShield; DataWalk product page; all as cited in this article. Vendor figures are self-reported and not comparable with each other.

What to ask instead of "what is your accuracy": what share of the claims you flag do my investigators confirm, measured on my book, in a paid pilot, against my current baseline. Ask for referral precision and the false-positive rate at the operating threshold you would actually run, not at the vendor's best point on the curve, and ask what recall was given up to get there. Rules-only systems run 60-85% false positives; a single-digit claim is a claim about one threshold on one portfolio, not a property of the software.

## What insurance fraud detection software costs in 2026

**How much does insurance fraud detection software cost, and what does ISO ClaimSearch cost?**

No major vendor publishes a rate card, Verisk included. ClaimSearch is quoted on claim volume under a participation contract. What Verisk does publish is the packaging: Claim Scoring, Research Hub, Digital Media Forensics and Compliance Solutions sit in ClaimSearch Essentials, while Provider Scoring, Claims Coverage Identifier and Networking Analysis are Complete-only. Ring detection lives in the upper tier. Treat any directory price without a source as a guess.

How much ISO ClaimSearch costs is one of the most-searched questions that lands on this page, and the honest answer is that Verisk does not publish a price. What Verisk does publish is the packaging, which is more useful than a number anyway. ClaimSearch splits into [ClaimSearch Essentials and ClaimSearch Complete](https://www.verisk.com/products/claimsearch-essentials/): Claim Scoring, Research Hub, Digital Media Forensics and Compliance Solutions are included in both tiers, while Provider Scoring, Claims Coverage Identifier and Networking Analysis are in Complete only. Networking Analysis is the capability that finds organized rings. If ring detection is the reason you are buying, you are buying the upper tier, and that is the most useful thing to know before the first pricing call.

The rest of the category prices the same way: custom-quoted annual subscription, scaled on claim volume, lines of business and module count. Enterprise platform licences - BAE Systems NetReveal, SAS - are negotiated as multi-year enterprise contracts and are usually the slowest to close. Nobody on this list publishes a per-claim rate, and nobody publishes a floor. For scale on what the category is collectively charging, Mordor Intelligence sizes [the insurance fraud detection market at $8.52B in 2026, reaching $20.22B by 2031 at an 18.87% CAGR](https://www.mordorintelligence.com/industry-reports/insurance-fraud-detection-market).

Be careful with the pricing figures that circulate in the ranked roundups. One widely-cited 2026 roundup runs a pricing column in which Verisk is listed at $850 per three months - that is the Xactimate estimation subscription, a different product from ClaimSearch or ClaimDirector - and IBM is listed at $10 per TB per month, which is cloud object storage. The scored directory pages quote "from $100K+", "$200K+" and "$500K+ annually" with no source and no vendor attribution at all. None of those numbers should enter a budget model.

### AI governance is now a procurement line item

The thing that changed in 2026 is not the detection math. It is that buying a fraud model now carries a documented-governance obligation across most of the country. The NAIC adopted its Model Bulletin on the Use of Artificial Intelligence Systems by Insurers in December 2023, and it had been picked up by [29 jurisdictions as of the NAIC's 8 October 2026 adoption map](https://content.naic.org/sites/default/files/legal-adoption-map-ai-model-bulletin.pdf) - 27 states plus the District of Columbia and Puerto Rico - with California, Colorado, New York and Texas running their own insurance-specific AI guidance alongside it. North Dakota adopted on 5 October 2026 and South Carolina on 24 September 2026, so the count is still moving.

The procurement consequence is specific: the bulletin leaves responsibility for a third-party AI system's behaviour with the insurer using it, not the vendor that built it. A fraud-detection RFP in 2026 therefore has to ask for model documentation, validation and bias-testing evidence, the override and human-review design, and the vendor's willingness to stand behind a DOI examination - on top of the usual security and data-handling review. That is a materially different question set from a 2023 RFP, and it is the dimension none of the ranked best-software lists score vendors on. Our [SOC 2 and compliance walkthrough for AI fraud work](/blog/soc2-compliance-ai-fraud-investigation/) covers the security and data-handling half of the same review.

## How to read the ranked top-10 lists

**Are the ranked insurance fraud detection software lists reliable?**

Several are not. Two sites publishing separate "Top 10 Best Insurance Fraud Detection Software of 2026" guides return an identical vendor order with identical scores on an identical 40/30/30 Features/Ease/Value weighting, and a third uses the same weighting and nearly the same score ladder in a different order - which moves Verisk from 8.8 to 7.3. None discloses a review count or a named data source. Use these pages to build a candidate set, never to rank one.

If you searched this category you met a layer of pages that publish a decimal score per vendor. They look like evaluation data. Checked against each other, they stop looking that way.

worldmetrics.org and wifitalents.com each publish a Top 10 Best Insurance Fraud Detection Software of 2026 ranking. Both return the same order and the same scores: Quantexa 9.1, Verisk 8.8, FRISS 8.5, Shift Technology 8.2, SAS 7.9, NICE Actimize 7.6, Featurespace 7.2, LexisNexis 6.9, TransUnion 6.6, Socure 6.3. Both describe the composite as roughly 40% Features, 30% Ease of use, 30% Value. zipdo.co publishes the same title with the same 40/30/30 weighting and a near-identical ladder - 9.2, 8.8, 8.5, 8.2, 7.9, 7.6, 7.3, 6.9, 6.6, 6.3 - but a different vendor order, which puts Verisk at 7.3 instead of 8.8 and Shift at 9.2 instead of 8.2. None of the three discloses a review count or a named source, and all three still rank Featurespace as a standalone insurance vendor. A score that swings 1.5 points depending on which copy of the template you open is not a measurement.

**Figure 2 (Ranked top-10 lists) - One template, three copies: Verisk scores 8.8 or 7.3 depending on the copy.** Three sites publish the same Top 10 Best Insurance Fraud Detection Software of 2026 title with the same weighting; one reshuffles the order.

**Verisk's composite score**

- worldmetrics.org: 8.8
- wifitalents.com: 8.8
- zipdo.co: **7.3**

**Shift Technology's composite score**

- worldmetrics.org: 8.2
- wifitalents.com: 8.2
- zipdo.co: **9.2**

**Same weights:** All three describe the composite as roughly 40% Features, 30% Ease of use and 30% Value.

**Takeaway:** Use these pages to build a candidate set, never to rank one.

Sources: worldmetrics.org, wifitalents.com and zipdo.co Top 10 Best Insurance Fraud Detection Software of 2026 pages, as reviewed for this article. None discloses a review count or a named data source.

The editorial roundups are better, but still worth checking on lane. One widely-cited top-10 of claims fraud detection tools lists FICO Insurance Fraud Manager, which is [FICO's health-care payer fraud, waste and abuse product](https://www.fico.com/en/newsroom/fico-releases-first-health-care-insurance-fraud-solution-scores-claims-independent-testing), next to Feedzai and NICE Actimize, which are payments and bank financial-crime platforms, and a Guidewire Fraud Analytics entry - when the P&C fraud scoring that actually reaches ClaimCenter arrives through the FRISS and Shift accelerators on the Guidewire Marketplace. Its own Rating column reads N/A on all ten rows. Use the lists to assemble a candidate set, then sort the candidates by lane yourself before anyone books a demo.

## Detection vs investigation - the layer no platform on this list occupies

**What is the difference between fraud detection and fraud investigation in insurance?**

Detection scores a claim and raises a flag. Investigation is everything after: document forensics, OSINT, statement cross-reference, timeline reconstruction. At the median US P&C carrier that is manual SIU at 14+ days per case, covering roughly 25% of the flag queue. No detection vendor ships it - see [the three-layer model](/blog/insurance-fraud-prevention-vs-detection-vs-investigation/).

Detection is upstream; investigation is downstream. Every vendor above ends its job at the flag. The per-claim investigation that follows - document forensics, OSINT, statement cross-reference, timeline reconstruction, financial pattern review - is the manual SIU workflow at 14+ days per case. Coverage sits at roughly 25% of flagged claims at the median US P&C carrier; the rest are paid, denied without full work, or queued indefinitely. Data vendors have started wiring into that handoff - Carpe made its claims reports agent-orderable over MCP in July 2026 - but as the [Carpe Data alternatives](/blog/carpe-data-alternatives/) comparison lays out, ordering an investigation is not running one.

2026 is the year the detection vendors started moving toward this layer without entering it. Verisk's own product timeline dates the launch of [AI-powered ClaimSearch Assistant to 2026](https://www.verisk.com/products/claimsearch/), described as AI built for investigations that synthesizes insights, reduces manual review and prioritizes next steps inside ClaimSearch. Quantexa's Claims Accelerator does the analogous thing inside ClaimCenter. Both are real improvements, and both are assistance: they make a human investigator faster on a case that the human still runs end to end. The 14+ day cycle and the roughly 25% coverage are products of how many cases one investigator can personally carry - 200+ at a typical carrier - and an assistant that speeds up each step does not change that arithmetic the way removing the per-case human bottleneck does.

**Figure 3 (After the flag) - Assistants speed up the investigator. Agents remove the per-case bottleneck.** The 14+ day cycle and ~25% coverage come from how many cases one investigator can carry. Faster steps do not change that arithmetic.

|  | Who works the case | Cycle per case | Flag coverage |
| --- | --- | --- | --- |
| Manual SIU (Hesper internal benchmark) | Investigator, end to end, 200+ cases each | 14+ days per case | ~25% of flagged claims |
| With an assistant (e.g. ClaimSearch Assistant) | Investigator, still end to end | Faster steps, same human per case | Still bound by caseload |
| With agents | 15+ phases in parallel, SIU lead reviews | minutes, not weeks | 100% of flagged claims |

**Complementary:** FRISS, Shift or Verisk keep flagging; the investigation layer takes the flag and resolves it.

Sources: Hesper internal benchmark (manual SIU at 14+ days per case, 200+ cases per investigator and ~25% coverage of flagged claims; 15+ phases); Verisk product timeline (ClaimSearch Assistant, 2026); Quantexa Claims Accelerator on the Guidewire Marketplace, 23 January 2026.

Hesper AI builds the investigation-layer agent that runs end-to-end on every flagged claim. The agent runs 15+ investigation phases in parallel - the parallelism is structural, not optional, because a software agent's per-case attention is not a constraint the way a human investigator's is. The cycle compresses from 14+ days to minutes, and coverage shifts from ~25% to 100% of flagged claims. The output is an audit-ready report a human SIU lead reviews, not a score that hands off to an investigator who then does the work.

Hesper is complementary to every vendor in the list above. Not a replacement. A carrier running FRISS, Shift, or Verisk continues to flag with that vendor; Hesper takes the flag and resolves it. Read [the three-layer model](/blog/insurance-fraud-prevention-vs-detection-vs-investigation/) for the full breakdown of where each layer ends, and [the Verisk head-to-head](/blog/hesper-vs-verisk/) for how this looks against the largest data utility specifically.

## How to pick: a decision framework for Top 50 P&C carriers

**How should a Top 50 P&C carrier choose a fraud detection platform?**

By tier. National Top-25: Verisk for the 1.8B-record utility, plus FRISS or Shift tuned to your largest line. Mid-market multi-state: FRISS or Shift as primary scorer over Verisk ClaimSearch. Regional digital-first: Snapsheet FNOL signals plus ClaimSearch. In all three, the open 2026 question is the investigation layer downstream.

Three tiers of carrier need; three different shortlists. The wrong move in this category is to copy a peer carrier's stack without checking whether your loss ratio is concentrated in the same lines theirs is.

### Tier A - National Top-25, complex multi-line

Anchor on Verisk for cross-carrier data (the 1.8B-record utility is not optional at this scale), layer FRISS or Shift Technology for scoring tuned to your strongest line, and add an investigation-layer vendor downstream. The CIO conversation will center on Guidewire ClaimCenter or Duck Creek Claims integration; the Compliance Officer will want documented-decision posture per NAIC Model Act 680.

### Tier B - Mid-market multi-state

FRISS or Shift Technology as the primary scoring vendor, with Verisk ClaimSearch as the cross-carrier data utility underneath. The investigation-layer decision is the new 2026 procurement question for this segment; the math works because flagged-claim coverage at mid-market carriers is typically the lowest, and the cost-per-case delta from manual to AI is the largest.

### Tier C - Regional, digital-first

Snapsheet or carrier-built AI for FNOL detection signals, plus Verisk ClaimSearch for cross-carrier match. The investigation layer often stays manual at this scale, but that is changing - the per-case cost curve makes investigation-layer agents accessible to regional carriers as the volume threshold drops. For the full stack picture, see [the 2026 SIU technology stack](/blog/siu-technology-stack-2026/) and [the claims management systems comparison](/blog/claims-management-systems-comparison/).

## Key takeaways

- The 2026 US P&C detection shortlist is dominated by Verisk, FRISS, and Shift Technology; the other seven platforms address narrower segments and should be evaluated against specific line-of-business needs.
- Verisk's industry-utility data depth (1.8B records, ~95% of US P&C market, FY2024 revenue of $2,882M) is the largest moat in detection and the safest procurement anchor for any carrier.
- Shift Technology's move into handler-assist agentic AI is the most consequential category shift since 2024, but Shift Claims is detection-and-assist, not autonomous investigation.
- Every detection vendor in the list flags claims and hands them to a human investigator; the per-claim investigation downstream is the manual SIU workflow at 14+ days per case and roughly 25% coverage.
- The 2026 procurement decision for most Top-50 P&C carriers is not which detection vendor to buy - it is what to add at the investigation layer downstream of the detection vendor they already run.
- Searching for insurance fraud detection software returns four different product categories - cross-carrier data utilities, claim scoring, graph and entity resolution, and bank-built financial-crime suites. Sort candidates by lane before ranking them, or you will price two products that solve the same half of the problem.
- Vendor accuracy numbers are not comparable. Anchor on Deloitte's published baseline of 20-40% detection on soft fraud and 40-80% on hard fraud, then measure referral precision on your own flag queue in a paid pilot.
- AI governance is now a procurement line item: the NAIC AI model bulletin reached 29 jurisdictions by 8 October 2026 and leaves responsibility for a third-party model's behaviour with the insurer, which makes model documentation and validation evidence an RFP requirement.

## Frequently asked questions

### Do I need more than one fraud detection platform?

Most US P&C carriers above 100K annual claims run more than one. The pattern is a cross-carrier data utility (Verisk ClaimSearch is the de facto standard at ~95% of US P&C market) layered with a scoring or agentic-AI vendor (FRISS or Shift Technology) tuned to the lines the carrier writes most. The data-utility layer surfaces history a single carrier cannot see on its own; the scoring layer rank-orders new claims against that history plus carrier-specific patterns. Running both is not redundant; they answer different questions. The cost question is whether the scoring vendor's incremental lift over Verisk's native scoring justifies the second contract. For most multi-state mid-market carriers, the answer is yes.

### How do these platforms differ from claims management systems?

Claims management systems (Guidewire ClaimCenter, Duck Creek Claims, Majesco ClaimVantage, Snapsheet) are the workflow backbone. They hold the case, route it, and store the file of record. Detection platforms sit alongside the claims management system and feed it scores or flags. Guidewire's [Celent 2024 Luminary status across all four regional P&C claims-system reports](https://www.guidewire.com/about/press-center/press-releases/20240409/guidewire-wins-seven-awards-and-luminary-distinction-in-celents-regional) reflects workflow dominance, not detection capability. A carrier running Guidewire still needs Verisk, FRISS, or Shift Technology for fraud scoring, and the flagged-claim file lives in ClaimCenter while the score travels in from outside. See the side-by-side in our [claims management systems comparison](/blog/claims-management-systems-comparison/).

### What happens to flagged claims after detection?

Most flagged claims are handed to a human investigator inside the carrier's Special Investigations Unit. The SIU runs the manual workflow: document forensics, OSINT, statement comparison, timeline reconstruction, financial pattern review. That workflow takes 14+ days per case at the median US P&C carrier, and the team gets through roughly 25% of the flag queue before the rest are paid, denied without full work, or queued indefinitely. The detection platform's job ends at the flag. The investigation is a separate layer with separate vendors - or, much more commonly today, no vendor at all and just headcount. See [the three-layer model](/blog/insurance-fraud-prevention-vs-detection-vs-investigation/) for the full breakdown.

### Are open-source fraud detection options viable for mid-market carriers?

For a single line of business with strong in-house data-science capacity, yes. Open-source scoring models (gradient-boosted trees, graph-based ring detection) can deliver detection lift comparable to the lower tier of commercial platforms. The trade-off is regulatory posture. State DOI antifraud-plan filings under NAIC Model Act 680 - [adopted, in the model or a variation of it, by 48 states](https://www.irmi.com/articles/expert-commentary/fraud-and-abuse-or-abuse-of-fraud-do-insurance-antifraud-laws-tip-the-scales-against-insureds) - require documented anti-fraud methodology. Commercial vendors carry that documentation and audit defense as part of the contract. An in-house open-source stack has to build it. For carriers with a Compliance Officer reviewing the antifraud-plan filing, the documentation overhead usually pushes the decision back toward a commercial vendor.

### How often should we re-evaluate our detection stack?

A full re-evaluation every 24 to 36 months tracks how fast the category is moving. Three triggers should pull it forward: a state DOI antifraud-plan filing flagged for review; a loss-ratio deterioration on a specific line where fraud is concentrated (workers comp, auto BI, property); or a new CCO, CFO, or Claims VP reviewing fraud-tech as part of a 100-day plan. The detection category itself is shifting from rules and scoring toward agentic AI; Shift Technology's handler-assist agents and the emergence of investigation-layer vendors are the two largest 2026 changes. Carriers who last re-evaluated in 2023 will see a meaningfully different shortlist today.

### Which platform fits workers comp, auto, or property?

There is no single answer. Verisk ClaimSearch and ClaimDirector cover all three with depth, anchored on the industry-utility data set, and Xactimate adds property-specific estimation across 22 of the top 25 US property insurers. FRISS has historically been strongest on auto scoring. Shift Technology covers auto and property with European-deployment depth, including the AXA Switzerland deployment that processed over 1M claims and stopped more than €12M in fraud. BAE NetReveal's ring-detection lineage fits organized workers-comp medical billing rings. CCC anchors auto specifically. The right answer depends on which line drives the largest share of your loss ratio.

### How much does ISO ClaimSearch cost?

Verisk does not publish a rate card for ClaimSearch, and no reliable public figure exists. It is quoted per carrier under a participation contract, scaled on claim volume rather than charged per click or per search. What Verisk does publish is the packaging, which is the more useful input to a budget conversation: [ClaimSearch Essentials and ClaimSearch Complete](https://www.verisk.com/products/claimsearch-essentials/). Claim Scoring, Research Hub, Digital Media Forensics and Compliance Solutions are included in both tiers. Provider Scoring, Claims Coverage Identifier and Networking Analysis are Complete-only. Networking Analysis is the ring-detection capability, so carriers buying ClaimSearch specifically to find organized fraud are buying the upper tier. Treat the "from $100K+" and "$500K+ annually" figures in the scored directory listicles as unsourced guesses; none of them attributes a number to a vendor.

### Which insurance fraud detection software should a fraud team shortlist first?

Start by sorting candidates into lanes, not by ranking them. Lane one is the cross-carrier data utility, where Verisk's ISO ClaimSearch is the de facto standard at 1.8B records and roughly 95% of the US P&C market. Lane two is claim scoring: FRISS, Shift Technology, Verisk ClaimDirector, INFORM RiskShield, SAS, BAE NetReveal. Lane three is graph and entity resolution, where Quantexa and DataWalk sit. Lane four is data and identity enrichment - LexisNexis Risk Solutions, Thomson Reuters CLEAR - which is additive to any scorer. Lane five is bank-built enterprise financial crime, NICE Actimize and Feedzai, where insurance claims is one modality among payments and cards. A first shortlist for a US P&C carrier is usually one vendor from lane one, one from lane two, and an explicit decision about whether lane three is a 2026 purchase or a 2027 one.

### How do fraud detection vendors compare on false-positive rates?

They cannot be compared on published numbers, because the numbers are not measured the same way. DataWalk advertises false-positive rates as low as 10%; rules-only legacy systems run 60-85%. Both figures are real descriptions of some configuration on some portfolio at some threshold, and neither is a property of the software you would buy. The comparable measurement is referral precision on your own flag queue: of the claims the system flags, what share do your investigators confirm. Run it as a paid pilot on historical claims with known outcomes, fix the operating threshold at the volume your SIU can actually absorb, and ask each vendor what recall they gave up to reach the precision they are quoting. A vendor that will not state the threshold behind a false-positive number is quoting marketing, not performance.

### What is the difference between fraud detection software and fraud prevention software?

Prevention acts before the loss or before the policy is bound: identity verification at quote, document authenticity checks at application, premium-fraud and rate-evasion screening at underwriting. Detection acts on a claim that already exists, scoring it and raising a flag. Investigation is the third layer, downstream of the flag, where the claim is actually worked. Vendors market across the boundary constantly, which is why prevention and detection searches return overlapping vendor sets, so the practical test is where in the lifecycle the software makes its decision. See [the three-layer model](/blog/insurance-fraud-prevention-vs-detection-vs-investigation/) for the full breakdown of where each layer ends.

### Does fraud detection software have to comply with the NAIC AI model bulletin?

The obligation sits with the insurer, not the software. The NAIC's Model Bulletin on the Use of Artificial Intelligence Systems by Insurers had been adopted by 29 jurisdictions as of the NAIC's 8 October 2026 adoption map - 27 states plus the District of Columbia and Puerto Rico - with California, Colorado, New York and Texas running their own insurance-specific AI guidance. It calls for a written AI systems programme with senior-management accountability, model validation and testing, and oversight of third-party AI tools where the insurer remains responsible for the tool's behaviour. In practice that means a 2026 fraud-detection RFP needs model documentation, validation and bias-testing evidence, a described override and human-review path, and a vendor commitment to support a DOI examination. None of the ranked best-software lists score vendors on any of it.

### Where does an investigation-layer vendor fit in this stack?

Downstream of detection. Every platform in this roundup ends its job at the flag; the per-claim investigation that follows is the manual SIU workflow at 14+ days per case and roughly 25% coverage of the flag queue. An investigation-layer vendor - Hesper AI is purpose-built here - takes flagged claims from FRISS, Shift, or Verisk and runs the SIU playbook autonomously, compressing the cycle to minutes and lifting coverage to 100% of flagged claims. The model is not replacement of any detection vendor on this list. It is the layer none of them ship. See [how this plays out against Verisk specifically](/blog/hesper-vs-verisk/).
