Mapping the carrier buying center for AI claims investigation: who decides, who blocks

A carrier AI claims-investigation purchase does not die on price. It dies in the seams between the six-to-eleven people who have to say yes. Buying claims-automation technology is a group decision now, and the vendor is in the room for a fraction of it - so the deal is won by the internal champion who carries the room, not by the pitch that opens it. This post maps every seat at that table: what each one is measured on, which two seats champion, which three quietly block, and the one message that moves each.

The scale sets the stakes every seat shares. Per the Insurance Information Institute, insurance fraud costs the US about $308.6 billion a year, and fraud is present in roughly 10% of P&C losses. That is the number the Claims VP is measured against in board materials, and it is the reason AI and fraud-identification automation now sit at the top of anti-fraud leaders’ priority lists. The trigger is warm. The blocker is internal alignment.

One framing note runs through the whole map, because it is the single lever that keeps any seat from perceiving a rip-and-replace. Hesper AI sits downstream of detection. It is the investigation layer, complementary to the FRISS, Shift Technology, and Verisk stack a carrier already funded - not a swap for it. Get that clear early and the buying center stops arguing about displacement and starts arguing about coverage. For the role that holds the budget chair, the deeper deployment view lives in the Claims VP AI investigation deployment playbook.

Enterprise claims-technology decisions are group decisions, and the group is large. Gartner finds a typical B2B technology purchase involves six to ten stakeholders, rising to about eleven for enterprise software above $100k in annual contract value - the band most carrier AI-investigation deals sit in. Gartner also finds 77% of B2B tech buyers describe their most recent purchase as very complex or difficult. The complexity is not in the product evaluation. It is in getting a room of people with different jobs to agree.

The second Gartner number reorders how you sell. Buyers spend only about 17% of the entire purchase journey with any single vendor’s sales team. The other 83% happens in internal meetings the vendor is never in - the hallway conversation where Finance raises a concern, the Slack thread where the CIO asks about SOC 2, the review where Compliance decides whether the output is defensible. A deck cannot attend those rooms. Only the champion can. That reframes the vendor’s job from persuading the buyer to arming the champion who persuades the buyer.

This is why mapping the seats has to come before mapping the pitch. Every seat has a metric, an objection, and an unlock. If you know all three for all eight seats, you can hand your champion a room-by-room script instead of a brochure. If you do not, the deal stalls in a seam you never see - most often Finance, IT, or Compliance - and you learn about it as a slipped close date, not a stated objection.

Name every seat and the one thing it optimizes for, because the same product fact gets framed differently for each. A carrier AI-investigation buying center has eight seats: the Chief Claims Officer or VP of Claims, the SIU Director, the Head of Claims Operations, the CFO or Finance reviewer, the CIO or CTO, Compliance, Legal, and Procurement. Each owns a different gate, and no single one can say yes alone - several can say no.

The seats also carry hard numbers that shape the room. On the money side, carriers already spend about 0.12% of premium on SIU and just above $1,200 per investigation on average, up roughly 11% since 2022, per the Coalition Against Insurance Fraud’s 2024 Insurer SIU Benchmarking Study. That $1,200 is the average budget cost allocated per investigation across all carriers, including cases that get only shallow review - it is the CFO’s baseline for how big this line can get. It is not the same as the fully loaded cost of a complete manual investigation, which runs closer to ~$2,500 (Hesper benchmark). The AI investigation layer runs about ~$150 per case.

On the capacity side, the same study puts investigator load at about 174 accepted referrals per investigator per year and SIU staffing at 0.67 full-time equivalents per $100M of gross premium, down from 0.9 - so the SIU is being asked to do more with proportionally fewer people. That is the Head of Claims Operations’ pain in one line. And on the detection side, automated tools now drive 29% of accepted SIU referrals (up from 21% in 2022) and automated sources make up 45% of all referrals. Read those three numbers together and the buying center’s real problem comes into focus: flags are rising faster than the capacity to work them.

The economic buyer is the Claims VP, and the credibility champion is the SIU Director - and the reason it is almost never IT is structural. Claims technology is co-owned, not IT-led. Deals that win only the CIO stall, because the CIO does not own the loss-ratio outcome the purchase is justified against.

Deloitte’s research on claims transformation is explicit: claims professionals must be "closely involved with developing and testing automated systems," and claims managers now own "a whole ecosystem" with an outcome orientation rather than a task orientation. Deloitte also documents one personal-lines insurer whose digital claims handling rose from single digits to as high as 55%. The lesson for this map is that the business side of claims co-owns the automation decision. A vendor who only convinces IT has convinced the wrong owner.

The Claims VP: arm her with loss-ratio math, not technology depth

The Claims VP sits in the budget chair for fraud tech and is measured on loss ratio. Her first questions are the procurement cycle, what her CIO needs to integrate, and the loss-ratio impact in basis points on the line where fraud concentrates - workers comp, auto BI, property. She shuts down pitches that lead with technology before loss ratio. The message that moves her is the coverage shift: manual SIU teams investigate roughly 25% of flagged claims and the rest are paid, denied without full work, or queued indefinitely. Closing that gap to 100% is the single biggest loss-cost lever in the fraud stack, and it is a board-ready narrative, not a headcount story.

The SIU Director: arm him with the audit trail

The SIU Director champions or vetoes internally, and his comfort with the audit trail decides the Claims VP’s confidence. Often ex-law-enforcement or ex-adjuster, his top fear is an AI conclusion he cannot defend in a deposition, an examination under oath, or a Suspicious Activity Report filing. His first question is whether his investigators can see what the agent did, override it, and produce a documented trail for the state DOI. The message that moves him is that the investigator’s role shifts from execution to decision-making: the agent runs 15+ investigation phases in parallel and returns an audit-ready report a human reviews, so the SIU gains throughput without giving up defensibility. Win these two seats and you have the room’s champions.

Three seats quietly kill AI-investigation deals: Finance, the CIO, and Compliance-plus-Legal. None of them announces a veto. They raise a concern in a room the vendor is not in, and the deal slips. Each block has a specific, nameable unlock.

Finance: the opex-that-outgrows-recovery fear

The CFO reviews fraud-tech procurement as one of dozens of vendor decisions and cares about cost per investigated case, payback period, and loss-ratio basis points - not technology. His top fear is an opex line item that grows faster than the loss-recovery story justifies. Against a baseline where carriers already spend about 0.12% of premium on SIU, a new line has to read as unit economics, not as a headcount ask. The unlock is a cost-per-investigated-case model anchored on coverage: ~$150 per AI investigation versus ~$2,500 fully loaded manual, applied across 100% of flags instead of 25%. Lead with FTE displacement and you lose him - it is politically untenable. Lead with the coverage-times-unit-cost curve and you have a case. The full memo lives in the CFO ROI memo for AI claims investigation.

The CIO: SOC 2, data flows, and build-versus-buy

The CIO reviews after the SIU Director has championed and Finance has nodded, and she reviews quickly - in days, not weeks - but vetoes if anything is off. Her top fear is a SOC 2 gap, an ambiguous data flow, or a vendor that requires training on her customer data. Her first questions are where the data sits, the SOC 2 or ISO posture, and the integration shape with her Guidewire or Duck Creek instance. Carriers on those platforms sometimes float a build-versus-buy alternative: our data-science team could build this. The honest answer is that no incumbent occupies the investigation layer, and building 15+ parallel investigation phases plus an audit-trail-native output that satisfies 10 CCR 2698.36 is a multi-year effort, not a sprint. The unlock is a clean data-flow diagram, a current SOC 2 posture, and a defined integration shape. The full gate is in the CIO checklist for AI investigation rollout.

Compliance and Legal: the undocumented AI decision

Compliance owns the antifraud-plan filings with state DOIs, and her top fear is a DOI audit that pulls an AI-investigated case and finds the documented-decision chain unreadable. Her first question is whether the output satisfies California 10 CCR 2698.36 and how it appears in the antifraud plan filed under NAIC Model Act 680, adopted in 48 states. Legal shares the concern from the evidence side and shuts down black-box conclusions presented without supporting evidence. The unlock for both is the same: an audit-trail-native output where every decision the agent makes is logged with sources, reasoning, and timestamps. This is also why detection alone does not close the loop - rules-based flagging carries a 60-85% false-positive rate, so a flag is not a determination, and the documented determination is exactly what these two seats need.

One table maps each seat to its top objection and the single message that moves it. This is the consensus matrix - the script the champion carries into the 83% of the journey the vendor never attends. Each row leans on the differentiator that unlocks that specific seat: coverage for Claims and Finance, the audit trail for Compliance and Legal, downstream-of-detection for the CIO.

The most common blocker in this buying center - we already bought FRISS, Shift, or Verisk, why do we need this - is the reason to buy the investigation layer, not the reason to pass. Detection and investigation are different layers of the stack. Detection scores and flags suspicious claims; investigation takes a flag and resolves it end-to-end with a defensible report. FRISS, Shift, and Verisk are the detection layer. Hesper AI sits downstream of them.

The numbers make the reframe concrete. The 2024 SIU Benchmarking Study shows automated tools now drive 29% of accepted referrals, up from 21% in 2022, and automated sources make up 45% of all referrals - a six-point jump. Detection volume is rising. But those flags meet the same investigator ceiling of about 174 cases per person per year. More flags, same capacity: that is the coverage gap, and it is created by the detection stack the carrier already funded. The sunk cost is not the reason to skip investigation - it is the reason investigation is now the constraint.

Where a technical buyer confuses the layers - "isn’t this what Shift’s new agentic AI does?" - the distinction is workflow position. Shift Claims is handler-assist agentic AI for adjusters, upstream in the workflow; Hesper is autonomous end-to-end investigation, downstream. Carriers can run both. For the same reason, publicly disclosed detection deployments like AXA Switzerland on Shift Technology (1M+ claims, €12M+ stopped) show detection working exactly as intended - and still leave the per-flag investigation to a human team downstream. Complementary to FRISS, Shift Technology, and Verisk - not a replacement. For the full layer breakdown, see prevention vs detection vs investigation.

Run the seats in order, because the vendor is in the room only about 17% of the time and sequencing is how the champion runs the other 83%. The order is not about hierarchy - it is about clearing the gates that can veto before you spend momentum on the gates that only reason.

1. Warm the champions first. Align the Claims VP on loss-ratio impact and the SIU Director on the audit trail. Nothing else moves until these two are convinced, because they carry the room when the vendor is absent.
2. Pre-clear Compliance and Legal on the documented-decision output before Finance sees numbers. Getting the regulatory risk off the table early removes the objection most likely to surface late and kill a deal that already had budget.
3. Hand Finance a unit-economics memo. Build it on cost per investigated case and coverage - ~$150 per AI investigation against ~$2,500 manual, applied to 100% of flags - not on headcount reduction. Give the CFO a curve, not a story.
4. Let the CIO sign off late. The technical gate is fast when the architecture is clean: SOC 2 posture, data-flow diagram, and integration shape with Guidewire or Duck Creek. Reviewed in days, not weeks, but it vetoes if anything is off - so bring it the clean version.
5. Supply Procurement early. Get the standard MSA and the security package into the pipeline before the champion has finished the room, so contract paper never becomes the bottleneck that stalls a decided deal.

Cycle length tracks the size of the buying center, not the calendar. Small regional carriers move fastest because there are fewer seats. Mid-market multi-state carriers are the strongest fit and move at a moderate pace. Top-25 national carriers run the slowest cycle - the most seats and the most gates - but sign the largest contracts. Across all three, a scoped pilot before an enterprise commitment lets the buying center see the audit trail and the coverage math on their own claims, which compresses the consensus phase more than any deck can. Manual SIU investigation takes 14+ days per case; showing a room a 2-4 hour audit-ready result on their own flagged claim does more consensus work than a quarter of meetings.