Hesper AI vs. Snapsheet: a faster claim does not buy a better investigation

Most people arrive at this comparison from one of two searches. One is a replacement search: Snapsheet alternatives, because a contract is up or a core-system decision is open. The other is a gap search: what Snapsheet does not do, because a flagged claim went out the door with nothing behind it. The ranked alternatives lists answer the first search and are silent on the second, which is awkward, because the second is where the loss dollars sit. This post answers both, and it starts with the one the lists skip. Take five days out of your average auto claim and you have done something genuinely hard. You have also done nothing to the flagged claim sitting three queues over, because cycle time and investigation capacity are independent variables. Snapsheet compresses the time it takes to settle a claim that is going to be paid. Hesper AI develops the claim that should not be paid until somebody looks at it properly. A carrier can take days out of its average claim and still close the same share of flagged files with no investigation behind them, because the second number was never a function of the first.

Snapsheet is good at the layer it occupies, and that is worth saying before anything else. Virtual auto appraisal at scale is hard to build, and Snapsheet built it starting in 2011, before most of the market believed the model would work. Its homepage states "Trusted by 170+ customers & investors, including 16 of the top 20 P&C carriers," and its appraisals page publishes a "3-5 Day average reduction in cycle time." Nothing in this post argues that a carrier should spend less on that work or run less of it.

One correction first, because it belongs to us. Our May 2026 four-vendor claims management systems comparison said Snapsheet's product story does not mention fraud or SIU capability anywhere on its public pages. That statement was too broad and this post supersedes it. Snapsheet ships OFAC and ISO compliance screening inside Payments, publishes a partner directory category headed "AI, Analytics & Fraud," integrated Shift Technology's Force fraud detection in May 2020, and announced a crash-severity and injury-analysis partnership with Talem AI in April 2024. All of that is real and it should be credited. The sharper point survives it: partner-delivered detection is still detection, and detection is not investigation.

What follows is the alternatives set sorted by what each vendor actually replaces, what the analysts scored and the dimension they left out, the product surface as re-checked on October 8, 2026, the fraud posture stated accurately, where the software directories get that posture wrong, the regulatory clock a flagged claim has to fit inside, the two curves that are moving at different rates, the California provision that indexes SIU staffing adequacy to detection volume, the two audit trails a carrier has to be able to produce, and the question sets for each RFP. For where Snapsheet sits by carrier tier against the rest of the shortlist, see the top claims management systems for P&C carriers in 2026. For the layer map across the whole fraud stack, the AI fraud platforms buyer's guide is the pillar.

The alternatives pages that rank for this query are shopping pages, and they are built the way shopping pages are built: a ranked grid, a star rating, a short blurb, a price column that almost always reads "price on request." The deepest of them, rfp.wiki, ranks 21 vendors with composite scores and puts CCC Intelligent Solutions and Shift Technology at the top on 4.4, then Cloud Claims, SimpleSolve, FileHandler Enterprise, Guidewire InsuranceSuite and Origami Risk at 3.7, down to NettShield ClaimsPoint at 2.6. It is a careful page. It is also a single flat list containing a claims core, an estimating platform and a fraud detection engine, scored against each other as though a carrier chooses one of the three.

The shorter lists have the same shape. One seven-item list names CCC, Mitchell, Tractable, Solera Audatex, Claim Genius, WeGoLook and HOVER, which is an appraisal and inspection list rather than a claims platform list. Another five-item list names Five Sigma, RightIndem, ClaimGenius, Tractable and Shift Technology, and reaches the sensible conclusion that Five Sigma is the closest functional replacement. A third returns YoroClaims, ClaimBook, Perfios Acclaim, Guidewire ClaimCenter, FileHandler Enterprise, Reimbursify and ProClaimz, several of which serve a different market entirely. A buyer reading all three gets 25 names and no map.

So here is the map. Snapsheet occupies four lanes at once, which is why the alternatives question has four answers. Sort the names by the lane each one would actually replace in a Snapsheet footprint, and the list stops being a ranking and starts being a decision.

Read the bottom row as the point of the table. A carrier can replace Snapsheet lane by lane, or keep it and bolt a second vendor onto each lane, and in neither case does anything on those lists produce the artifact that 10 CCR 2698.36(a)(5) asks for. The detection lane comes closest and still stops short: a score, a flag, a case record, then a person. That is the gap this post is about, and it is the same gap the broader market scan finds in the four claims cores, which is worked through in the top claims management systems for P&C carriers in 2026.

What Snapsheet costs, as far as anyone publishes it

Pricing transparency is a stated evaluation criterion on at least three of the pages ranking for this query, and not one of them resolves it, so the honest version is worth writing down. Snapsheet publishes no price. There is no rate card on snapsheetclaims.com, no pricing page in the navigation and no public tier structure. SelectHub lists the model as monthly and quote-based with no free trial and prints a starting range of $100 to $500, which is a directory estimate rather than a vendor figure and carries no stated unit. Another review states plainly that costs depend on company size, selected solutions, integrations and implementation requirements, which is the accurate answer. Per-claim figures circulate in summaries of this category and none of them trace to a named source, so they do not belong in a business case.

Which makes the pricing question a structural one rather than a number hunt. Four questions get a usable answer out of any claims platform quote: is the unit a claim, a seat, a transaction or the book; what happens to the unit price when volume moves 30% in either direction; which modules are inside the base and which are separate line items, specifically payments, appraisal and the AI actions; and what is billed at implementation versus at renewal. Ask the same four of the investigation layer. Hesper charges a flat per-case rate and takes no contingent or recovery-share compensation, so the first question has a one-word answer and the second does not move the unit price.

Every page ranking for this query uses crowdsourced star ratings, and none of them mentions the one evaluation in this category that is done by analysts against a published framework. Celent's P&C Claims Systems: 2026 North America Edition, published on June 30, 2026 by Nathan Golia, Harry Huberty, Karun Arathil, Fabio Sarrico and Karlyn Carnahan, profiles the major P&C claims administration systems in North America. Snapsheet's own announcement of the result, dated July 7, 2026, puts the evaluated set at 66 claims systems sorted into five tiers: Luminary, Technology Standout, Functionality Standout, Noteworthy Solution and Developing Solution. Five Sigma's release adds the count that makes the tier mean something: six systems out of 66 are Luminaries.

Six slots out of 66 is a real shortlist, and it is a better starting point for a replacement decision than a 21-row grid of composite scores. It is also not published as a list anywhere, because Celent sells the report and each vendor announces only its own result. Assembled from the vendor announcements, with each row stating exactly what that vendor claimed and when, the tier looks like this.

Two rows are deliberately hedged, because the vendor releases do not all use the same report title and inventing agreement between them would be dishonest. Four vendors use the words Luminary and North America about the same edition. Duck Creek announces the top rating in the same matrix on September 9, 2026 under a different title, and Origami Risk announces Luminary status on July 1, 2026 under another. A buyer should treat the first four as confirmed and ask Celent about the last two.

Now the part that matters more than the membership list. The three dimensions Celent scores are Advanced Technology, Breadth of Functionality, and Customer Base and Support, and the XCelent award names confirm those are the scored axes rather than a summary of them. Investigation is not one of them. Fraud is not one of them. The Celent abstract notes that insurers are trying to take friction out of claims "while still being vigilant about potential leakage and fraud," which states the tension and then scores neither side of it.

That is not a criticism of the framework. A claims administration system should be scored on claims administration, and sixty-six of them were. It is a precise statement about what a Luminary badge certifies: that the system is technically modern, functionally broad and well supported. It certifies nothing about whether a flagged claim leaving that system carries a developed investigation behind it, because that was never on the scorecard. A carrier that shortlists from the Luminary tier has picked a good claims core and has not yet made a decision about the investigation lane at all.

Snapsheet and Hesper AI are bought against two different files. Snapsheet optimizes the claim that is going to be paid, and it is scored on cycle time, appraisal accuracy and cost to handle. Hesper works the claim that has been flagged, and it is scored on how many flagged claims get fully developed and how defensible the finding is.

Most claims are legitimate, and the operating model that serves them is the one Snapsheet built. A policyholder submits photos through a guided flow, the file routes by skill, licensing and geography, an estimate is written remotely, and payment goes out digitally. Every day removed from that path is a day of loss adjustment expense removed and a point of policyholder satisfaction gained. That is a real budget line with a real owner, and it has been the priority of most claims organizations for a decade.

The other file behaves nothing like it, and it is expensive. In a November 18, 2024 presentation to the NAIC Antifraud (D) Task Force, Rich Gibson, MAAA, FCAS, Senior Casualty Fellow at the American Academy of Actuaries put annual property and casualty fraud at $45.0 billion, workers compensation at $34.0 billion and premium evasion at $35.1 billion. The presentation adds the framing an actuary would reach for: property and casualty, workers compensation and auto premium evasion together come to roughly $90 billion on approximately $700 billion of premium, which is "12% or more of all premiums go to fraud."

The Insurance Information Institute puts the whole-market number alongside it. Its fraud facts page, citing the Coalition Against Insurance Fraud, states that insurance fraud costs the U.S. $308.6 billion annually. The background page, citing a 2017 Verisk study, puts auto premium leakage at "at least $29 billion a year," described as "as much as 14 percent of all personal auto premiums." None of those dollars sit in the intake step or the estimate step. They sit in the files that got flagged and then did not get worked.

Here is the arithmetic a claims VP should carry into both vendor conversations. Manual SIU capacity reaches roughly 25% of what a carrier flags. That figure is a Hesper internal benchmark rather than a published industry statistic, and it is labelled that way everywhere it appears in this post. Read the other way round, it is a permitted derivation from the same benchmark: roughly three of every four flagged claims close without a full investigation. That ratio does not move when the front end gets faster, for the reason this post keeps returning to. The mechanics of why the ratio sits where it does are in why 75% of flagged insurance claims are never fully investigated.

Two budget lines, then, not one. The cycle-time line is scored in loss adjustment expense and in policyholder satisfaction, and it belongs to the claims operation. The loss-cost line is scored in basis points of loss ratio and in what a state department of insurance sees during an antifraud plan review, and it belongs to the same executive wearing a different hat. Snapsheet moves the first and does not claim to move the second. Hesper moves the second and does not touch the first.

The reason the two lines get confused is that they run through the same claim. A flagged claim is still a claim: it has an acknowledgment clock, an accept-or-deny clock, a payment clock and a policyholder waiting on the other end of all three. Speeding up the handling of that claim and developing the suspicion inside it are separate pieces of work performed by separate parts of the organization on separate evidence. The stage-by-stage version of that map, showing where intake, appraisal, payment and investigation each sit in the lifecycle, is in the claims process end to end. Which of those stages software now automates, and which it does not, is the subject of the complete guide to automating the claims lifecycle.

Snapsheet is a cloud-native claims management and virtual appraisal platform founded in Chicago in 2011. Its published product surface covers a claims platform, no-code workflows, integrations and APIs, analytics and oversight, payments, an AI module stated as launching in 2026, virtual vehicle appraisals across seven vehicle categories, and total loss settlement. It sells to carriers, MGAs, TPAs and self-insured entities.

The eight modules, as published

The product navigation on snapsheetclaims.com, checked on August 21, 2026, lists the following. Each description below stays inside what the vendor publishes.

• Claims Platform, the claims management engine, with a unified claim view, assignment, work queues, financials and dashboards.
• No-Code Workflows, point-and-click configuration of how a claim is handled.
• Integrations & APIs, connectors to external systems and third-party platforms.
• Analytics & Oversight, metrics and claim performance tracking.
• Payments, digital disbursement, headlined on the product page as "Deliver secure, digital disbursements in minutes, not days."
• Snapsheet AI, six pre-built AI actions, stated on the product page as launching in 2026.
• Virtual Vehicle Appraisals, covering private passenger, boat and marine, RV, motorcycles, medium and heavy duty trucks, utility and exotic vehicles.
• Total Loss Settlement, the end-of-life vehicle claim.

The performance figures Snapsheet publishes are vendor-published and have not been independently audited, so read them as the company's own claims about its own product. The claims platform page states "10 Million+ monthly automated actions," "$75 Billion+ in premiums supported" and "Implementation as fast as 12 weeks." The insurance appraisals page states a "3-5 Day average reduction in cycle time," a "+20% improvement in appraisal accuracy," and "Trusted by 15 of the top 20 P&C Carriers."

The customer count differs by page, and averaging them would be dishonest, so here they are as published: the homepage says 170+ customers and investors including 16 of the top 20 P&C carriers, and the appraisals page says 15 of the top 20. Both are the vendor's own count on the vendor's own page as of August 21, 2026.

Funding, backing and a dated third-party implementation record

Snapsheet's own March 25, 2021 Series E2 announcement describes the company as "a pioneer in virtual appraisals and a leader in cloud-native claims management software," and puts total capital raised at more than $100 million. The $30 million round was led by Ping An Global Voyager Fund and Pivot Investment Partners, with participation from Nationwide, Liberty Mutual, Intact Ventures, Tola Capital, Commerce Ventures and F-Prime Capital. Carriers investing in a claims platform is a meaningful signal about the platform.

State Farm Ventures added a strategic investment on July 17, 2023. That announcement describes Snapsheet as partnering with "over 140 insurance carriers, Managing General Agents (MGAs), Third-Party Administrators (TPAs), and self-insured entities in the United States, Canada, and Europe," operating primarily in auto, property and commercial lines. Michael Remmes, VP of State Farm Ventures, said State Farm Ventures "has been tracking Snapsheet for a while and is pleased to invest in them as a differentiated claims technology company."

Implementation speed is the part of the story with a dated third-party record behind it. On November 17, 2025, Aspire General Insurance announced it had selected Snapsheet as its claims management system, and the release states that "From initial kickoff to the first claims processed, their full implementation onto the Snapsheet platform took only 90 days." Sam Rea, CTO at Aspire, said: "When we were evaluating whether to build or buy, Snapsheet stood out as the clear choice." Mary Zelaya, Senior Claims Manager, said: "Before Snapsheet, our claims handling was as tedious as pen and paper. Now, tasks that used to take an hour are done in minutes."

Snapsheet AI, and what its six actions are scoped to do

The newest piece is Snapsheet AI, headlined "Take AI out of the black box" and stated on the page as "Launching in 2026." Six actions are named: "Claims Summarization - From high-level to highly specific use cases"; "Auditing and QA - General handling and SOPs specific to you"; "Communication Drafts - Across SMS, email, letters, and file notes"; "Recommendations - Real-time guidance for any claim type"; "Key Signal Identification - Including triage, subrogation, and litigation"; and "Photos and Documents - Tagging, identification, and summarization." The page also states that "Our platform provides the open APIs and development tools you need to bring your AI solutions to life."

Refusing to ship a black box is the right instinct, and it happens to be the same instinct that makes an investigation record defensible. The enumeration is what it is: six actions, scoped precisely, aimed at handling. Fraud and SIU are not among them, and there is nothing to read into that beyond what the page says. A company that scopes its AI page carefully is a company that is not overclaiming, which is more than can be said for a good deal of this market.

That is a strong product line and a serious company. Snapsheet built the virtual appraisal category, it has carriers on its cap table, it ships a configurable workflow engine and a payments rail, and it can put a mid-sized carrier live in about a quarter. Nothing that follows takes any of that back.

Snapsheet has a fraud posture and it is partner-delivered and screening-based by design. Sanctions and contributory-database screening ship inside Payments. A partner directory category titled "AI, Analytics & Fraud" supplies predictive fraud analytics. A May 2020 integration delivers a detection vendor's alerts into the platform. Fraud signal reaches Snapsheet. Developing the flagged file is a different layer.

The partner directory names the category

The clearest statement Snapsheet makes about its own architecture is its partnerships page. It organizes partners under seven category headings, and the first of them is "AI, Analytics & Fraud." Inside that category, in Snapsheet's own descriptions: ClaimSearch, which automates "data reporting and alert generation, streamlining claims processing, meeting reporting obligations, and mitigating risk"; OFAC, "Integrated sanctions screening for compliant claim disbursements"; Charlee.ai, "AI based predictive analytics related to fraud, litigation, severity, and reserving"; and CLARA Analytics, "AI-driven insights that precisely assess claims severity, detect potential fraud, and streamline decision-making." Foundation AI, Claimtouch, CCC Intelligent Solutions, Agentech, Liberate, Reserv and others fill out the same directory.

That is a real fraud posture, credited to the vendors that supply it. It is also a precise architectural statement. Detection arrives through partners. The platform is the surface that receives it. Nobody is hiding anything, and a carrier reading that page gets an accurate picture of what it is buying and from whom.

The Payments module goes further and ships two controls natively, listed on the page as "OFAC & ISO Compliance," inside pre-engineered payment workflows that "apply checks, validations, and audit trails automatically." Sanctions screening at disbursement is a genuine antifraud control, and reporting a claim into a contributory database is a real obligation discharged. Both are matching and reporting functions. Screening a payee against a list answers whether this party may lawfully be paid. Reporting a claim to a contributory database answers whether this loss has been seen elsewhere. Neither answers whether the loss happened the way it was described.

The 2020 Shift Force integration, in the CEO's own words

In May 2020, Shift Technology and Snapsheet announced a partnership under which "Users of Snapsheet's claims management platform can receive notification of suspicious claims directly from Shift's Force fraud detection technology." Brad Weisberg, CEO and Founder of Snapsheet, said: "With our API-driven platform, incorporating fraud detection from Shift is only a click away, and is another key way in which we can give claims professionals the tools needed to do their job as effectively as possible." Jeremy Jawish, CEO and Co-founder of Shift Technology, said the partnership gives joint customers "the power to uncover anomalies and identify potentially fraudulent claims within the claims management platform they're already familiar with, speeding up the overall process."

Weisberg is describing his architecture correctly, and that is exactly why the quote is useful and not something to argue with. A click that brings in a flag is upstream of every hour of work the flag creates. Once an anomaly surfaces inside the platform, somebody still has to pull the prior loss history, examine the submitted documents for signs of alteration, cross-reference the recorded statement against the physical evidence, reconstruct the timeline, research the involved parties, and then write down what they found and why they believe it. That sequence is the investigation, and no notification produces it.

Talem AI, the most fraud-adjacent thing on the surface

On April 2, 2024, Snapsheet and Talem AI announced a partnership bringing Talem's "AI-driven Delta-V and injury analysis capabilities" and "upfront crash and injury information" into the Snapsheet platform. Matt Kay, COO at Talem AI, said the collaboration "represents a significant step forward in enhancing the efficiency and accuracy of auto insurance claims management." David Beck, Business Operations Manager at Snapsheet, said that "within Snapsheet's claims platform, insurers can apply a proactive, guided, and dynamic approach that properly advances the claim and generates a wide ripple effect of benefits for the insurer and customer."

Give this one full credit, because Delta-V and biomechanical review are the classic evidentiary tools for low-impact soft-tissue questions and staged-collision questions. Crash severity against claimed injury is one of the oldest and most useful tests an investigator has. Having that input arrive automatically on the claim is a genuine improvement over a handler eyeballing a photo of a bumper. It is also an input, not a determination. Snapsheet does not market it as an SIU product, and it would be unfair to either company to describe it as one.

One more item belongs in Snapsheet's column. The company has published its own article on photo-based solutions and auto insurance fraud. It engages the topic seriously and makes no product capability claim on the back of it. Vendors overclaim on fraud constantly. Snapsheet engaged the subject and did not, and that restraint is worth naming.

So the honest summary is a layered one. Screening sits at the compliance-control layer. Partner analytics and Force alerts sit at the detection layer. Talem's crash-severity inputs sit at the evidence-supply layer. The investigation layer, where a flagged file is developed into a determination somebody signs, is unfilled. That is not a gap in Snapsheet's marketing. It is a gap in the carrier's stack that Snapsheet's marketing correctly does not pretend to fill.

Hesper is built for that layer, and it is built with detection inside it rather than only downstream of somebody else's. Hesper has built-in fraud detection plus full investigation: from fraud detection to fraud resolution. A carrier already running Shift Force, Charlee.ai, CLARA Analytics or ISO ClaimSearch through Snapsheet keeps all of it, because Hesper is complementary to FRISS, Shift Technology, and Verisk - not a replacement. The equivalent head-to-head on the estimating layer, where the same two-layer logic applies to a different product, is Hesper AI vs. CCC Intelligent Solutions.

This section exists because the section above it took work, and the pages ranking for this query did not do that work. SelectHub's Snapsheet profile lists fraud detection as an active capability and describes it as "Built-in fraud detection capabilities help identify suspicious claims early." That sentence is wrong in the one word that matters. Built-in is a claim about architecture, and Snapsheet's architecture puts detection in a partner category, which Snapsheet itself documents honestly on its own partnerships page under the heading "AI, Analytics & Fraud." The error is the directory's, not the vendor's.

It is worth being clear about why that matters more than a typo. The partner-delivered architecture is a defensible design choice and this post credits it twice. But the two architectures behave differently in procurement: partner-delivered detection means a second contract, a second data agreement, a second integration to configure and a second vendor to name in a market conduct response. A buyer told the capability is built in budgets for none of those and discovers all four at implementation. The same buyer then has no reason to ask the question that actually decides the outcome, which is what happens after the flag fires.

The ratings underneath these grids deserve the same scrutiny, and the directories are usually transparent about it if you read the small print. The same profile reports an 82% user satisfaction rating and states its basis: 14 user reviews across two review sites. Fourteen reviews is a signal and it is not a benchmark, and it is the number sitting behind a score presented next to a 21-vendor ranking. The listed cons are similarly honest about their own limits, one of them reading that reviews "don't mention" whether customization is flexible, which is an absence of data rendered as a weakness. Meanwhile the most thorough of the grids notes its own gaps plainly: fraud and SIU get limited direct discussion, and regulatory audit requirements are not addressed at all.

Three questions take a directory shortlist and turn it into something a claims committee can use. First, for each vendor on the list, is the fraud capability native or partner-delivered, and if partner-delivered, which partner and under whose contract? Second, how many reviews is each score built on, and over what period? Third, and this is the one no grid answers, what does each vendor produce after a flag fires: a score, a case record, or a written finding with its sources attached? The third question is the one that separates the lanes in the table above, and it is answerable by a demo rather than by a star rating.

A fraud investigation does not suspend the claim clock. It replaces a fixed day count with a reasonable time for full investigation, and only where the insurer already holds a reasonable basis supported by specific information available for review by the regulator. Investigation buys the extra time. The extra time does not buy the investigation.

Start with the speed side, because that is the side Snapsheet serves. The NAIC Unfair Property/Casualty Claims Settlement Practices Model Regulation #902 sets the pattern most states follow. Section 6.A: "Every insurer, upon receiving notification of a claim shall, within fifteen (15) days, acknowledge the receipt of such notice unless payment is made within that period of time. If an acknowledgement is made by means other than writing, an appropriate notation of the acknowledgement shall be made in the claim file of the insurer and dated. Notification given to an agent of an insurer shall be notification to the insurer."

Section 6.B adds the regulator's own clock: "Every insurer, upon receipt of any inquiry from the insurance department respecting a claim shall, within twenty-one (21) days of receipt of such inquiry, furnish the department with an adequate response to the inquiry in duplicate." Section 7.F sets the payment clock: "The insurer shall affirm or deny liability on claims within a reasonable time and shall tender payment within thirty (30) days of affirmation of liability, if the amount of the claim is determined and not in dispute."

Section 7.A carries the accept-or-deny obligation: "Within twenty-one (21) days after receipt by the insurer of properly executed proofs of loss, the first party claimant shall be advised of the acceptance or denial of the claim by the insurer. No insurer shall deny a claim on the grounds of a specific policy provision, condition or exclusion unless reference to such provision, condition, or exclusion is included in the denial. The denial must be given to the claimant in writing and the claim file of the insurer shall contain documentation of the denial as required by Section 4."

Section 7.B is what happens when the file stays open: "If the insurer needs more time to determine whether a first party claim should be accepted or denied, it shall so notify the first party claimant within twenty-one (21) days after receipt of the proofs of loss, giving the reasons more time is needed. If the investigation remains incomplete, the insurer shall, forty-five (45) days from the initial notification and every forty-five (45) days thereafter, send to the claimant a letter setting forth the reasons additional time is needed for investigation." Every open investigation past that mark is a recurring administrative obligation with a compliance tail on it.

The carve-out appears twice, and both times it is conditioned

Now the hinge, and it is worth having both halves side by side because having them together is what proves the rule, not an exception. Section 7.A, second paragraph: "Where there is a reasonable basis supported by specific information available for review by the insurance regulatory authority that the first party claimant has fraudulently caused or contributed to the loss, the insurer is relieved from the requirements of this subsection; provided, however, that the claimant shall be advised of the acceptance or denial of the claim within a reasonable time for full investigation after receipt by the insurer of a properly executed proof of loss."

Section 7.B, second paragraph, says the same thing about the extended-investigation clock: "Where there is a reasonable basis supported by specific information available for review by the insurance regulatory authority for suspecting that the first party claimant has fraudulently caused or contributed to the loss, the insurer is relieved from the requirements of this subsection; provided, however, that the claimant shall be advised of the acceptance or denial of the claim by the insurer within a reasonable time for full investigation after receipt by the insurer of a properly executed proof of loss."

Read the two together and the structure is unmistakable. Relief from the day count is available twice, and both times it is conditioned on specific information the regulator can pick up and review, and both times the fixed count is replaced by a reasonable time for full investigation rather than by no obligation at all. Suspicion alone does not open the valve. Developed information does.

California puts numbers on the same structure

California's 10 CCR 2695.7 tightens the middle number and then repeats the same conditional. Subsection (b): "Upon receiving proof of claim, every insurer, except as specified in subsection 2695.7(b)(4) below, shall immediately, but in no event more than forty (40) calendar days later, accept or deny the claim, in whole or in part." Subsection (h) sets the payment clock: upon acceptance, "every insurer, except as specified in subsection 2695.7(h)(1) and (2) below, shall immediately, but in no event more than thirty (30) calendar days later, tender payment or otherwise take action to perform its claim obligation."

Subsection (d) states the investigation duty in the same body of regulation that states the speed duty: "Every insurer shall conduct and diligently pursue a thorough, fair and objective investigation and shall not persist in seeking information not reasonably required for or material to the resolution of a claim dispute." Subsection (c)(1) covers the extension request: "If more time is required than is allotted in subsection 2695.7(b) to determine whether a claim should be accepted and/or denied in whole or in part, every insurer shall provide the claimant, within the time frame specified in subsection 2695.7(b), with written notice of the need for additional time."

And subsection (k) is the California version of the hinge, quoted in full because the conditions are the whole point: "Subject to the provisions of subsection 2695.7(c), where there is a reasonable basis, supported by specific information available for review by the California Department of Insurance, for the belief that the claimant has submitted or caused to be submitted to an insurer a suspected false or fraudulent claim as specified in California Penal Code Section 550 or California Insurance Code Section 1871.4(a), the number of calendar days specified in subsection 2695.7(b) shall be: (1) increased to eighty (80) calendar days; or, (2) suspended until otherwise ordered by the Commissioner, provided the insurer has complied with California Insurance Code Section 1872.4 and the insurer can demonstrate to the Commissioner that it has made a diligent attempt to determine whether the subject claim is false or fraudulent within the eighty day period specified by subsection 2695.7(k)(1)." Read the opening clause of (k) rather than skipping to the fraud language. The extension is subject to subsection (c), so the written notice of additional time is still owed even when the eighty days are running.

One provision cuts in the carrier's favor and is worth knowing, because it is the reason a quiet investigation is contemplated by the rules, not merely tolerated by them. Subsection (b)(1) requires an insurer denying a first party claim to provide "a statement listing all bases for such rejection or denial and the factual and legal bases for each reason given." Subsection (b)(2) then provides: "Subject to the provisions of subsection 2695.7(k), nothing contained in subsection 2695.7(b)(1) shall require an insurer to disclose any information that could reasonably be expected to alert a claimant to the fact that the subject claim is being investigated as a suspected fraudulent claim." The regulation anticipates an investigation running underneath the clock. It expects the investigation to actually happen.

Now map the manual baseline onto those numbers. A manual SIU investigation runs 14+ days per case, and 14 days is where the work starts rather than where it lands, because one investigator is carrying 200+ open cases. Both are Hesper internal benchmarks. Against a 21-day accept-or-deny clock and a 40-day California clock, a file opened on day three of the claim and worked for a fortnight or more is routinely still open when the answer is due. That is how the 45-day letter cadence starts, and it is also the mechanism behind the coverage number: when capacity is the constraint, most flagged files never get the full workup, so the carve-out is unavailable in practice because there is nothing to put in front of the regulator.

The front end of the claim is measurably faster and measurably more digital, and investigation capacity is not moving at the same rate. J.D. Power measured average property repair completion at 29.6 days in its 2026 study, down 2.8 days year over year. Over a comparable horizon, SIU staff at participating insurers grew 1.4%. Two curves, two very different slopes.

The front-end numbers are current and well attributed. The 2026 U.S. Property Claims Satisfaction Study, reported by Claims Journal on March 18, 2026, was fielded from December 2024 through December 2025 and is based on 5,093 homeowner responses. Average repair completion came in at 29.6 days, down 2.8 days year over year. Average time to final payment came in at 40.7 days, down 3.4 days. Mark Garrett, director of insurance intelligence at J.D. Power, attributed the movement to "investments made over the past several years in digital channels that make it faster and easier to communicate with customers throughout the claims process."

One precision note before anyone puts those two numbers next to a regulation. The 40.7-day figure is time to final payment, which is downstream of acceptance. California's 40-day clock under 2695.7(b) is an accept-or-deny clock. The digits nearly match and the measures do not. Treating them as the same number would be a factual error dressed up as a coincidence.

The same study measured how the claim is now being submitted, and this is where the shape of the evidence changed. Thirty-eight percent of homeowners filed first notice of loss digitally. Forty-nine percent submitted photos for estimates or payments. Forty-five percent received updates digitally. Nearly half of property claimants are now supplying photographic evidence through a channel, which is exactly the model Snapsheet pioneered in auto and exactly the model that has become mainstream in property.

Now the other curve. The Insurance Information Institute, citing the Coalition Against Insurance Fraud's Insurer SIU Benchmarking Study, reports that study participants experienced "an increase in SIU staff at 1.4 percent from 2021 to 2022 lower than the 2.5 percent growth rates from the two previous studies." Growth in the low single digits, decelerating. The same III page reports that "80 percent of respondents currently use predictive modeling to detect fraud, up from 55 percent in 2018." Detection adoption climbing steeply; investigation headcount inching.

There is one more data point worth carrying, with its age stated plainly. In the last iteration of the SIU benchmark that has been reported publicly, covering 2019, Insurance Journal reported that "the percentage of referrals from automated systems accepted by special investigation units has declined to 15 percent from 22 percent." That is 2019 data published in 2020, older than we would normally cite, and it is included for one reason: it is direct evidence that adding automated detection signal did not, by itself, produce more accepted investigations. The signal went up. The acceptance rate went down.

The reason those two curves move at different rates is structural, not cultural. Front-end cycle time responds to software: a guided capture flow, a routing rule, a payment rail. Investigation capacity has historically responded only to headcount, because the unit of investigation was a person's attention, one file at a time. A manual investigator completes roughly 10 investigations a month and carries 200+ open cases, both Hesper internal benchmarks. Doubling detection volume against that denominator does not double investigations. It doubles the queue.

This is why a faster front end raises the pressure on the flagged lane rather than relieving it. The same regulatory clock now has less slack in it, because the rest of the claim moves faster and the flagged file still takes 14+ days to develop. The gap between how fast the clean claim resolves and how long the flagged one waits widens with every cycle-time improvement, and every one of those waiting days is a day the file sits inside a 21-day or 40-day window with nothing in it a regulator could review.

Autonomous investigation changes the denominator instead of the numerator. Hesper runs 15+ investigation phases in parallel on every flagged claim rather than sequentially, so the constraint stops being one person's attention. Throughput moves from roughly 10 investigations per investigator per month to 800+, and coverage moves from ~25% of flagged claims to 100%. Both are Hesper internal benchmarks. Cost per case falls with them, far enough below what a manual file costs that the decision to work one more flagged claim stops turning on what it costs. The investigator's role shifts from execution to decision-making: which findings hold, which need a human interview, which go to referral.

California measures the adequacy of an insurer's SIU staffing partly against the insurer's own detection volume. 10 CCR 2698.32(a) lists the volume of suspected fraudulent California claims currently being detected among the factors that may be considered in staffing the SIU. Adding detection surface without adding investigation capacity moves the standard the carrier is measured against.

The provision reads, in full: "The adequacy of an insurer's SIU staffing shall be determined by its demonstrated ability to establish, operate and maintain an SIU that is in compliance with these regulations. Factors that may be considered in staffing the SIU include, but not limited to, the number of policies written and individuals insured in California, number of claims received with respect to California insureds on an annual basis, volume of suspected fraudulent California claims currently being detected and other factors relating to the vulnerability of the insurer to insurance fraud." (10 CCR 2698.32)

Read that last clause slowly, because it is the whole section. Staffing adequacy is indexed, among other things, to the volume of suspected fraudulent claims currently being detected. A carrier that bolts a partner detection surface onto a fast workflow platform has increased the numerator of that ratio without touching the denominator. It has not become non-compliant by doing so. It has moved the bar it will be measured against, and it did so as a side effect of a purchase that was made for entirely different reasons.

The coverage obligation is the companion provision, and its clause about automated referrals is the one that matters in a platform stack. 10 CCR 2698.36(c): "The SIU shall investigate each credible referral of suspected insurance fraud that it receives from integral anti-fraud personnel, including automated or system-generated referrals. A credible referral of suspected insurance fraud is one that includes a red flag or red flags. However, the first sentence of this subdivision (c) notwithstanding, in the event that upon a preliminary review the SIU determines that it is reasonably clear that the red flag or red flags contained in the referral is not or are not the result of suspected insurance fraud, the SIU need not open an investigation. In the event that the SIU refrains from opening an investigation pursuant to the immediately preceding sentence, the SIU shall document in the claim file or SIU investigation file the reasons supporting its conclusion that the red flag or red flags contained in the referral is not or are not the result of suspected insurance fraud."

Four words in that provision do the work: including automated or system-generated referrals. A flag arriving from a partner detection engine inside a claims platform is not a lesser class of referral. If it carries a red flag it is a credible referral, and it produces either an investigation or a documented reason there was not one. Neither of those is free, and both scale with detection volume rather than with headcount. That is the arithmetic a carrier inherits when it turns on a detection partner.

The supporting definitions are broad on purpose. 10 CCR 2698.30(n) defines a red flag as "facts, circumstances, or events which, singly or in combination, support(s) an inference that insurance fraud may have been committed, and includes patterns or trends that may indicate fraud, facts or circumstances present on a claim, and behavior or history of person(s) submitting a claim or application." Subsection (l) defines integral anti-fraud personnel to include "insurer personnel who the insurer has not identified as being directly assigned to its SIU but whose duties may include the processing, investigating, or litigation pertaining to payment or denial of a claim or application for adjudication of claim or application for insurance." The duty attaches to the work rather than to the job title.

Subsection (q) names the evidence categories directly, and it is the paragraph a photo-first pipeline should read twice. "Suspected insurance fraud" includes "any misrepresentation of fact or omission of fact pertaining to a transaction of insurance including claims, premium, and application fraud. These facts may include evidence of doctoring, altering or destroying forms; prior history of the claimant, policy holder, applicant, or provider; receipts; estimates; explanations of benefits (EOBs), medical evaluations or billings; ... An identifiable pattern in a claim history may also suggest the possibility of suspected fraudulent claims activity. A claim may contain evidence of suspected insurance fraud regardless of the payment status."

Receipts and estimates are named in the regulation, and a virtual appraisal pipeline runs on photos, receipts and estimates by design. That is not a criticism of the model; guided photo capture is a better evidence pipeline than a clipboard, and Snapsheet was right to build it. It does mean the flagged file needs corroboration and not only inspection, which is a full subject rather than a paragraph, and it is covered in the AI-generated evidence investigation playbook. Snapsheet itself has written on photo-based solutions and auto fraud without claiming a product capability from it, which is the honest posture.

The profession is already asking the coverage question in its own words. Gibson's presentation to the NAIC Antifraud (D) Task Force names exactly three referral mechanisms: adjuster judgement; a check-list approach where claim data is compared against set criteria and "If a criterion (often called a 'flag') corresponds to the claim, it is referred to the SIU"; and "A computer model, developed internally or externally, creates a fraud score for claims and claims above a predetermined threshold are referred to the SIU." A later slide asks what actuaries consider, and two of the questions are "What proportion of claims are referred to the SIU and what criteria led to their referral?" and "What proportion of those claims referred to SIU resulted in prevention of paying fraudulent claims?"

That second question is the coverage question, posed by a credentialed actuary to insurance regulators rather than by a vendor to a prospect. A carrier that has bought detection and not investigation has a good answer to the first question and no answer to the second. Make every flagged claim investigable is the operational form of an answer to it. The section-by-section walkthrough of the California regime behind these citations is in the 10 CCR 2698 SIU compliance guide.

A carrier has to produce two different records, and they are not the same artifact. NAIC Model Regulation #902 section 4.B asks for documentation sufficient to permit reconstruction of the insurer's activities on each claim. California 10 CCR 2698.36(a)(5) asks for a summary of the investigator's findings and the basis for them.

The first record is short to state. NAIC #902 section 4.B: "Detailed documentation shall be contained in each claim file in order to permit reconstruction of the insurer's activities relative to each claim." That is an activity log, and the scope of what counts as an activity is wide, because section 3.G defines investigation as "all activities of an insurer directly or indirectly related to the determination of liabilities under coverages afforded by an insurance policy or insurance contract."

Snapsheet ships exactly the right artifact for that provision, and names it precisely. The claims platform page describes "Transparent Claim History: Track a complete, time-stamped history of every action on a claim for full visibility and compliance." A complete, time-stamped history of every action is what "permit reconstruction of the insurer's activities" asks for, almost word for word. It is one of the more underrated things on that page, and it is the kind of feature that only gets appreciated during a market conduct exam. The same page also ships "Document Management: Preview, compare, and adjust photos and documents using a side-by-side comparison or a single view."

The second record is a different animal. 10 CCR 2698.36(a) requires the SIU to "establish, maintain, distribute, and adhere to written procedures for the investigation of possible suspected insurance fraud," and lists what an investigation shall include: a thorough analysis of the claim file, application or insurance transaction; identification and interviews of potential witnesses; "Utilizing one or more industry-recognized databases identified by the SIU as appropriate for use in fraud investigations involving the particular line of insurance in question."; "Preservation of documents and other evidence obtained during an investigation."; and then the fifth item, which is the one no workflow platform produces.

2698.36(a)(5), in full: "Writing a concise and complete summary of the entire investigation, which is specific to the investigation at hand, is separate from any other document prepared in connection with the investigation, and includes the investigators' findings regarding the suspected insurance fraud and the basis for their findings." The summary has to answer a set list of questions, of which two are enough to make the point here: "What facts caused the reporting party to believe insurance fraud occurred or may have occurred?" and "a statement as to whether or not the investigation is complete."

Three phrases in that provision are doing structural work. Separate from any other document prepared in connection with the investigation means a findings record cannot be assembled by exporting the activity log. Findings and the basis for their findings means the record has to carry reasoning, not just events. A statement as to whether or not the investigation is complete means somebody has to be willing to say the work is done, on the record, in writing.

The distinction is not pedantry, and it is the most common stack-planning error in this category. An activity log answers what did we do and when. A findings record answers what did we conclude and why. A regulator reading the first can verify the file moved. A regulator reading the second can verify the conclusion was earned. Only the second satisfies the prompt-payment carve-out, because only the second contains specific information available for review.

Hesper is built against the second column. Every phase of an investigation is logged with its sources, its reasoning and its timestamps, and the output is a written record carrying findings and the basis for them rather than a score with a confidence interval attached. 15+ investigation phases run in parallel on every flagged claim, which is what makes the record complete instead of partial: document forensics, prior-loss history, open-source research on the parties, statement cross-reference, timeline reconstruction and financial-pattern analysis all land in the same file, on the same claim, at the same time.

Two procurement facts belong here because they get asked in the same meeting. Hesper holds SOC 2 Type I. And Hesper prices a flat per-case rate with no contingent, recovery-share or share-of-savings component, ever, which matters for a compliance reason as much as a commercial one: 10 CCR 2698.33(a) provides that any contract entered into by an insurer "shall not relieve the insurer of any obligation under these regulations or the IFPA," and 2698.33(c)(2) requires such contracts to "Not include provisions that could provide disincentives to the referral and/or investigation of suspected insurance fraud." Ask that question of every contract in the claims stack. For what makes an AI-produced investigation record hold up under examination, see how to generate an audit-ready fraud investigation report.

The modal deployment runs both. Snapsheet handles intake, workflow, appraisal and payments, either as the system of record or as a layer over one. Detection reaches the platform through its partner surface. The investigation layer consumes the flagged claim and returns a written finding as an artifact on the claim. Nothing gets ripped out and no budget is redirected.

Picture the stack a digital-first carrier or MGA actually runs in 2026. Snapsheet is the claims platform. Shift Force, Charlee.ai, CLARA Analytics or ISO ClaimSearch supplies the fraud signal through the partner directory. Payments applies OFAC and ISO screening at disbursement. Talem's crash-severity inputs arrive on the auto file. Intake, workflow, appraisal, payments and detection are all covered by a competent vendor, and covered well. Then a claim gets flagged, and the next step is a person carrying 200+ open cases. That is the seam, and it is open in almost every carrier of this shape.

The integration shape is the one Snapsheet already describes for itself. It ships an Integrations & APIs module, its Snapsheet AI page states that "Our platform provides the open APIs and development tools you need to bring your AI solutions to life," and its CEO has publicly characterized the platform as API-driven. The flagged claim and its attachments flow out for investigation; the finished findings record flows back onto the claim as an artifact the workflow can route on. Hesper never replaces the claims system of record and is not a claims adjudication tool: the human SIU lead and adjuster make the call. The detection-layer version of this same boundary is in Hesper AI vs. Verisk, and the agentic-handling version is in Shift Claims vs. Hesper.

One planning detail is worth flagging for a CIO. Guidewire and Duck Creek do not appear on Snapsheet's published partner directory as of August 21, 2026, although Snapsheet does carry a "Core Systems & Policy Admin" partner category. That is not evidence that Snapsheet cannot integrate with them and should not be read that way. What it reflects is that Snapsheet is frequently the system of record itself, particularly for digital-first carriers and MGAs, which changes where the investigation layer plugs in. If Snapsheet is your system of record, the investigation layer connects to Snapsheet directly rather than to ClaimCenter, and the integration conversation involves one vendor instead of two.

The cycle-time row is the one most likely to be misread, so read it carefully. Snapsheet's 3-5 days and Hesper's minutes are not comparable numbers. Snapsheet's figure is a reduction in the elapsed time of an appraisal that was always going to be completed. Hesper's is the elapsed time of an investigation that, in most carriers today, does not happen at all on three out of four flagged files. One is a delta on work already being done. The other is work moving from not-done to done, which is why coverage rather than speed is the number to hold an investigation vendor to.

For the virtual claims and workflow RFP

1. What is the measured cycle-time baseline before implementation, and how is the delta measured afterward? The published figure to test against is the vendor's own, in Snapsheet's case a 3-5 day average reduction on auto physical damage appraisals.
2. How is appraisal accuracy defined and measured? A 20% improvement is a ratio, so ask what sits in the denominator and who computed it.
3. What is the implementation window for a book this size, and what did comparable go-lives take? Snapsheet publishes implementation as fast as 12 weeks, and Aspire General Insurance reported 90 days from kickoff to first claims processed in November 2025.
4. When a red flag fires mid-workflow, what happens to the file? Which queue does it enter, who is notified, and does the payment step hold automatically or by exception?
5. How is an SIU referral configured inside the no-code workflow, and what evidence exists afterward that the referral fired and was received?
6. Which detection partners are live in our configuration, what do they cost, and what is the expected referral volume once they are on? That number is the input to the staffing-adequacy question under 10 CCR 2698.32(a).

For the investigation RFP

1. What percentage of flagged claims will actually be investigated end to end after deployment? Cycle time per case is the wrong headline number if coverage stays where it was.
2. Does the output satisfy 10 CCR 2698.36(a)(5): a concise and complete summary specific to the investigation at hand, separate from any other document, carrying findings and the basis for them, and including a statement as to whether the investigation is complete?
3. Can an investigator see every step the system took, override any of them, and reproduce the reasoning months later in a deposition or an examination under oath?
4. What is the security posture? Hesper holds SOC 2 Type I, and the answer should be a specific attestation rather than an aspiration.
5. How is it priced? Hesper charges a flat per-case rate and takes no contingent or recovery-share compensation, ever. Any structure tied to savings on paid losses should be read against the disincentive language in 10 CCR 2698.33(c)(2).
6. What is the integration shape with whatever is the system of record, and does the finished record land back on the claim rather than living in a separate tool nobody opens?

Three seats read those lists differently, which is useful, not a problem. The claims VP is scoring basis points of loss ratio against loss adjustment expense and wants to know which list moves which number. The compliance officer is scoring the antifraud plan she files with the state, and cares most about questions five and six on the first list and questions two and three on the second. The CIO is scoring integration cost and security posture, and will veto on question four of the second list before anyone gets to price.

So the decision is not Snapsheet or Hesper. It is which of the two duties is currently under-resourced. If claims settle slowly, if appraisal is manual, if payment takes days, that is the virtual claims and workflow category and Snapsheet is a credible answer to it. If flagged claims close without investigation, if each case takes 14+ days, if the gap between referrals received and investigations opened is not closing, that is the investigation layer, and no amount of workflow speed addresses it. Making every claim faster and making every flagged claim investigable are two different projects with two different owners.