Hesper AI pricing and packaging: how investigation gets priced and how to get a quote

Hesper AI does not publish a fixed public price list, and neither does any other enterprise insurance fraud-tech vendor - FRISS, Shift, and Verisk are all quote-based. That is not opacity; it is what happens when price is a function of your claims volume, your lines of business, and how the platform integrates with your claims system rather than a per-seat sticker. So the useful question is not "what does Hesper cost" but "what determines my number."

This post answers that question for the people who own the budget line. It walks through the five pricing models live in the category, the six cost drivers that set a carrier's quote, how to build the business case before a vendor number exists, and the questions worth asking whoever you end up buying from. The through-line is coverage: what it costs you today to leave most of your flagged claims unworked, and what has to be true of any quote for that to change.

For the finance reviewer, the pricing page is the ROI page. This is the companion to the CFO ROI memo for AI claims investigation, which carries the full payback math; here the focus is what you actually pay for and what moves the number.

Enterprise insurance fraud-tech is quote-based across the entire category, and the absence of a public price list is the norm rather than a Hesper quirk. The reason is structural: a platform that investigates flagged claims is priced against the volume of flagged claims a carrier generates, and that volume swings by two orders of magnitude between a small regional carrier and a Top-25 national. A single published number would be wrong for almost everyone who read it.

The category proves the point. FRISS, one of the most established detection vendors, has no public pricing page at all - friss.com/pricing returns a 404. Shift Technology and Verisk are the same: enterprise quotes negotiated against a carrier's book, not a self-serve checkout. When every serious vendor in a category prices by quote, the right move for a buyer is to stop hunting for a sticker and start assembling the inputs that produce an accurate one.

There is a second structural reason specific to Hesper. Detection and investigation are different layers of the stack, and Hesper sits at the investigation layer that no detection vendor occupies. From fraud detection to fraud resolution is two jobs, not one: detection flags the suspicious claim, investigation resolves it end-to-end. That means Hesper is a separate budget line from your FRISS, Shift, or Verisk contract - additive spend, not a swap. The pricing conversation is not "replace what I have" but "fund the layer my current stack does not cover." The size of the problem that layer addresses is large: the Coalition Against Insurance Fraud estimates insurance fraud steals at least $308.6 billion every year.

Search for what claims investigation software costs and you will find numbers, which is more than this post has offered so far. They are worth reading. They are also, with one exception, counting something other than the end-to-end investigation of a flagged claim. Here is each figure that currently ranks for this query, checked against the page that publishes it.

Two patterns are worth naming. The first is that the category's own benchmark lumps detection and investigation into a single band, so the $400K to $2.8M range cannot tell you what either layer costs on its own. The second is that several dollar figures circulating in AI-generated answers about this category - a programme costing 1% to 2% of gross written premium, recovery-share fees of 10% to 30%, enterprise pricing at 0.1% to 0.5% of claims processed - were not on the pages they had been attributed to when we opened them. If a number is going to carry weight in your business case, open its source first.

The unit-mismatch trap

The most expensive mistake in this category is comparing two quotes on contract value without checking what each one meters. A per-screened-claim price bills every claim that enters the pipeline. A per-investigation price bills only the subset your detection layer flags. At the roughly 10% flag rate typical of P&C, those two denominators differ by an order of magnitude, so a screened-claim rate that looks ten times cheaper per unit is not cheaper at all - it is buying a different job on a much larger base. Normalise both quotes to the same two denominators before either enters a spreadsheet: cost per claim in the book, and cost per flagged claim worked through to a defensible finding. Any vendor should be able to state its number both ways. If it cannot, the quote has not been explained to you.

There are five live pricing models in the category, and each aligns the vendor's incentive on a different thing. Knowing which model you are being quoted under tells you what the vendor is optimizing for - and whether that matches your coverage goal.

Per-case or per-investigation

You pay for each flagged claim investigated. This aligns cost directly with usage and is the cleanest map to the value of an investigation platform, because the value is investigations completed, not seats occupied. It is also the model that makes the budget legible: your annual investigation spend is your flagged-claim volume times a single figure, so you can model a coverage scenario on the back of an envelope instead of inside a contract. The risk to watch is unbounded spend on a high-flag book, which tiered pricing exists to cap. For a TPA or shared-services operation, per-claim is also the one allocation model that survives regulatory review - the four ways to charge back AI investigation works through why the other three fail.

Per-seat

You license each investigator using the platform. This is the legacy software model, and it has a structural mismatch with autonomous investigation: the whole point is that one investigator now reviews far more cases, so a per-seat price penalizes exactly the leverage you are buying. Per-seat makes sense for tools where a human does the work; it makes less sense where the platform does the work and the human reviews and signs.

Platform subscription, tiered, and outcome-based

The remaining three round out the category. Platform subscription is a flat annual fee for access, simple to budget but loosely tied to value. Tiered pricing brackets cost by claims-volume bands, which gives a carrier a predictable number and caps per-case exposure. Outcome or value-based pricing ties a portion of the fee to loss avoided or recovered, aligning the vendor's incentive most tightly with the carrier's but requiring agreed measurement. For autonomous investigation specifically, per-case and tiered-volume map most naturally to the value delivered, because both track investigations completed rather than seats or flat access.

The right model is the one that aligns the vendor's incentive with your coverage goal - getting 100% of flagged claims investigated rather than the roughly 25% manual teams reach. Per-case and tiered-volume both do that. Per-seat works against it.

Six drivers set the number on a quote. None of them is seat count. They all trace back to one thing: how many flagged claims you generate, and how deep the investigation on each one has to go.

1. Annual claims volume. More claims means more flagged claims to investigate. This is the single largest driver, which is why a quote is organized around claims volume rather than carrier name.
2. Fraud-flag rate. About 10% of property-casualty claims involve fraud, but your actual flagged volume is set by your detection vendor's threshold. A conservative threshold flags fewer claims at higher confidence; an aggressive one flags more, raising both coverage need and per-period investigation count.
3. Lines of business. Workers compensation, auto bodily injury, and property carry different investigation depth. A staged-accident ring or a workers comp claim with medical and financial layers needs more of the 15+ investigation phases than a clean first-party property loss.
4. Claims-system integration. Whether you run Guidewire ClaimCenter, Duck Creek, or another system of record shapes the integration work. An investigation layer consumes a flagged claim and returns a report as a case attachment, so the surface is small, but the integration shape still factors into deployment scope.
5. Data-source breadth. The number and type of external sources an investigation draws on - public records, OSINT, document forensics, financial-pattern analysis - affects depth per case.
6. Deployment scope. A scoped pilot on one line costs far less than full production coverage across the book. This is the lever a carrier controls most directly: start narrow, prove the economics, then expand.

One driver deserves a CFO-specific note. The cost of fraud is not abstract overhead - it shows up in premium. The Insurance Information Institute, citing the Coalition Against Insurance Fraud, attributes as much as 14% of personal auto premium to premium leakage. When a line carries leakage that large, the investigation depth on that line is doing real loss-cost work, and the cost driver is also the value driver.

The sticker is also not the whole cost. Integration, data plumbing, and change management carry their own line items, which is the total-cost-of-ownership story we lay out in the hidden integration costs of legacy claims AI. An investigation layer with a small integration surface keeps that tail short, which is part of why deployment scope, not headcount, is the variable a buyer should be tuning.

You do not need a quote to know whether this is worth a conversation. The expensive half of the comparison already sits in your own numbers: a manually investigated SIU case carries a fully loaded cost your finance team already tracks and takes 14+ days, and you know how many claims your detection layer flags. Multiply that cost by that count and you have the honest price of full coverage under the model you run today - a number almost every carrier finds unpayable, which is precisely why coverage sits where it does.

Now look at what that money is actually buying, because throughput is the thing to interrogate. A manual SIU investigation runs 14+ days per case, and a single investigator completes roughly 10 investigations a month while 200+ sit open on the desk. That ceiling is why most US P&C carriers fully investigate only about 25% of flagged claims - the other 75% are paid, denied without full work, or queued. The constraint has never been detection recall; it is the human hours available downstream of the flag. AI investigation runs each case in minutes, lifting throughput toward 800+ cases per investigator per month and coverage toward 100%. Cost per case is what converts that throughput into a budget you can actually approve.

Run the math on a representative mid-market carrier and the shape becomes obvious. Take 50,000 annual claims. At the roughly 10% P&C fraud rate, that is on the order of 5,000 fraud-suspect claims a year (an illustrative derivation, not a guaranteed flag count). Working all 5,000 the manual way, at a fully loaded cost per case, runs to a budget no SIU is ever given - which is precisely why no manual team investigates all 5,000, and why coverage sits at 25%. So the only thing a quote has to settle is whether the marginal cost of one more investigation falls far enough that triage stops being a budget decision. To put your own volumes and leakage behind that, the claims leakage calculator runs it on your book in a couple of minutes.

The 5,000-claim line is an illustrative derivation from the roughly 10% fraud rate applied to a mid-market claim volume; your number depends on your detection threshold. But the shape holds at any volume. Under the manual model every additional investigation is expensive enough that somebody has to rank which flags deserve the hours, and about 75% of them lose that argument. An investigation layer earns its quote by making that ranking decision cheap enough to stop making.

Every page that ranks for this query prices software. None of them prices the thing software is being compared against, which is a strange omission, because the alternative to funding an investigation layer is not zero. It is investigator hours, and those have a published market rate.

The US Bureau of Labor Statistics puts median pay for claims adjusters, appraisers, examiners and investigators at $78,020 a year, or $37.51 an hour, as of May 2025, across 389,700 people in the occupation. That is a median wage, not a loaded cost. Your own figure has to carry benefits, payroll tax, supervision, systems, the external data and records subscriptions an investigation consumes, and whatever share of overhead your finance team allocates to the SIU. Your controller already has that multiplier. Apply it to the hours a 14-plus-day case actually consumes and you have the per-case number this post deliberately does not publish for you - computed on your cost base instead of someone else's assumptions about it.

The labour market is also moving in a direction that matters to any multi-year business case. BLS projects employment in the occupation to decline 6% between 2025 and 2035, and names the mechanism directly: "data processing speed and artificial intelligence (AI) capabilities will continue to increase, which will improve efficiency and make workers more productive." The planning implication is not that investigators disappear; the decision still belongs to a licensed human. It is that a budget built on hiring your way to full coverage is being built against an occupation that is shrinking and automating, while the 200+ open cases per investigator and the 14+ days per case stay exactly where they are.

There are two distinct ROI surfaces, and conflating them is the most common mistake in a pricing conversation. The first is cost per investigated case - a steep reduction on work you already do. The second, and larger, is the loss recovered from the flagged claims you previously could not afford to investigate. Detection-only ROI never reaches the second surface.

Start with the detection benchmark, accepted at face value. Shift Technology publicly cites 4x ROI in the first year on a detection deployment. That is a real return, and it is a detection return: it improves the quality and recall of the flags. What it does not touch is the roughly 75% of flags that never get fully investigated, because investigation is a different layer. Detection ROI makes the flags better; it does not get the flagged claims worked.

Hesper's ROI lever is that second surface - the coverage gap. Lifting investigation from roughly 25% to 100% of flagged claims means loss recovered from claims that were previously paid or denied without full work. The pool is large: US P&C fraud runs about $45 billion a year and workers compensation about $34 billion, per the Insurance Information Institute citing the Coalition Against Insurance Fraud. Most flagged claims at a given carrier are resolved without full investigation, so closing that gap is the dollar lever detection-only spend cannot reach. The two ROIs stack: detection improves flag quality upstream, investigation works the flags downstream, and a carrier budgets for both.

For the finance reviewer who needs the full payback model, two companion pieces carry it. The CFO ROI memo above lays out year-one net spend, payback, and IRR; three modeled carrier ROI scenarios run the math by carrier tier.

There is a set of costs in this category that no ranking page prices, because they are not software costs. They are the conditions a state insurance regulator puts on how you may staff, document and account for investigation work, and they land on the same budget line as the platform.

New York is the useful worked example, because its requirements are written down. Section 409 of the New York Insurance Law and Regulation 95 require covered insurers to maintain a special investigations unit and a fraud prevention plan, and the Department of Financial Services is explicit in its guidance on fraud prevention plans and SIUs that the SIU should be established as a separate unit with its own budget line, with investigative personnel kept separate from claims and underwriting and reporting into those departments only at executive level. There is no minimum investigator count - DFS says each company has broad latitude in deciding how much of its resources to dedicate to fraud prevention - but the company has to be able to justify that those resources are adequate.

Three things follow for a pricing conversation. First, the separate-budget-line requirement means investigation spend is not something you can quietly fold into a claims-system line item in a covered state, so the quote has to stand on its own merits in front of someone. Second, DFS permits insurers to use an outside contractor to perform the functions of an SIU, under a signed agreement and with continuing departmental oversight - that is the regulatory footing any vendor in this category stands on, and a vendor that has not thought about it is telling you something. Third, the required Fraud Detection and Procedures Manual and the adequacy standard both turn on documentation, which makes the audit trail behind each investigation part of what you are buying rather than a feature bullet.

One pricing model deserves a direct warning. Several pages in this category float a fee tied to a share of fraud recovered or loss avoided. Hesper does not price that way and will not. When the fee moves with the outcome of a claim, you have handed a market conduct examiner a reason to ask whether the investigation was scoped toward its finding, and that is not a question worth inviting into a claims file. The objection is not only regulatory: Monetizely, which recommends the per-screened-claim model this category is converging on, rejects a fraud-savings share on the grounds that it "moves the invoice away from the product's observable output and toward a debate over counterfactuals." A flat per-case rate has no counterfactual to argue about.

A faster, more accurate quote comes from arriving with five numbers. Because price is a function of your book, the conversation moves quickly once these are on the table, and you avoid the back-and-forth that comes from quoting against assumptions.

1. Annual claims volume across the lines you want covered. This is the single largest input to the number.
2. Your current flag rate and the detection vendor in place, if any. This sets your actual flagged volume and tells us whether Hesper runs downstream of detection or standalone with its built-in detection.
3. Your top two or three fraud-prone lines. Workers comp, auto bodily injury, and property drive different investigation depth.
4. Your claims system of record - Guidewire ClaimCenter, Duck Creek, or other. This shapes the integration scope.
5. Your target coverage. Whether you are aiming to lift one line to 100% or expand across the book changes deployment scope, which is the variable you control most directly.

With those five numbers, the quote is grounded in your economics rather than a sticker that would be wrong for your book. The framing to hold onto is that Hesper is an additive line to your detection spend, not a replacement for it - complementary to FRISS, Shift Technology, and Verisk - so the budget question is how to fund the investigation layer your current stack does not cover, and the per-case economics are what make funding it work.

What to ask any vendor in this category

These questions work on Hesper and on everyone we compete with. They are the ones that separate a quote you can budget against from a quote you renegotiate in month nine. If a vendor cannot answer them cleanly, that is information.

• What unit am I actually buying - claims scored, investigations completed, seats, or flat access? Detection and investigation vendors bill different units, and comparing two quotes on contract value without checking the unit is how carriers end up funding one layer twice.
• What happens when my flagged volume moves? Ask for the behaviour in both directions. A high-flag quarter should not produce an uncapped invoice, and a quiet one should not leave you paying for capacity you never used.
• Is implementation inside the number or beside it? Get claims-system integration, data plumbing, and change management quoted explicitly rather than discovering them in year one.
• What is the floor - minimum commitment, minimum term, and the cost of stopping? A pilot that cannot end cleanly is not a pilot.
• If any part of the fee is tied to outcomes, who measures the outcome and how? Value-based pricing aligns incentives only when both sides agree in advance on what counts as loss avoided and who audits the count.
• What does expansion cost? The second line of business and the second region are where a clean first quote usually stops being clean.

Phasing the spend is its own decision, and it usually starts narrow. A scoped pilot on a single high-fraud line proves the per-case economics before the budget commits to production scope. For the Claims VP funding and sequencing this, the Claims VP deployment playbook lays out the budget and 12-month timeline for moving from pilot to full coverage.

There is no rate card to compare against, and that is deliberate: the only number worth having is the one scoped to your book. Talk to us with the five inputs above and you will leave the first conversation with a scoped figure, the agents that fit your lines, and a pilot shape to prove the economics on.