Shift Claims vs. Hesper: agentic AI for claims handling vs. autonomous investigation

Shift Claims and Hesper are both agentic AI, but they do different jobs: Shift Claims accelerates the claim handler's workflow, and Hesper runs the SIU investigation on a flagged claim end-to-end. That distinction is the whole comparison. If you are an SIU director or a claims VP looking at Shift's September 2025 launch and wondering whether it touches your investigation backlog, the short answer is that it sits upstream of it.

This post lays out the honest dividing line. Shift genuinely wins on handler workflow, detection maturity, and its cross-carrier data network - we will say so plainly, because strawman comparisons help no one make a procurement decision. What Shift Claims does not do is perform the deep, per-claim investigation a flag requires: evidence gathering, statement cross-referencing, timeline reconstruction, financial-pattern analysis, and an audit-ready report. That is the layer Hesper occupies.

The frame for the whole piece is a layered model: prevention, then detection, then investigation. Shift Claims sits at detection and handling; Hesper sits at investigation. They are complementary, not mutually exclusive. For the wider map of who sits where, this is a cluster post under our AI fraud platforms compared buyer's guide.

Shift Technology launched Shift Claims on September 16, 2025. Per the official launch release, the agentic AI assesses and prioritizes cases, guides and assists claim handlers, and automates tasks across the claims lifecycle. The product is described as human-in-the-loop, with human intervention when required. In plain terms: it advises and acts alongside a human claim handler, end to end through the handling process.

Shift reports four numbers for early adopters of Shift Claims: 3% lower claims losses, 30% faster claims handling, a 60% overall automation rate, and greater than 99% accuracy in claims assessment. Read those as Shift's stated early-adopter results, not guaranteed product specifications and not independently verified facts. They are real handling-tier gains on the path a claim takes through processing, and they are worth taking seriously on those terms.

The launch release carries a named early-adopter quote from Markus Keller, COO and Head of Customer Operations at AXA Switzerland. We flag that specifically because there is a separate, older AXA Switzerland fraud-detection case study on Shift's site - a different deployment. For Shift Claims specifically, the Sept 2025 release is the source. If you are mapping Shift against the rest of the field, our AI fraud platforms compared buyer's guide places Shift among the detection vendors.

Both products use the word agentic, so it pays to be precise about what each agent is aimed at. Shift's own product page is explicit: per the Shift Claims solution page, the agents are "human-in-the-loop by design" and "work alongside human experts." They assess and unify claim characteristics, develop claim-strategy recommendations, and prepare steps for human action or automation. That is handler-assist: the agent does the staging work and a human handler stays in the driver's seat.

Hesper's agent is aimed at a different job. It takes a claim that has already been flagged and runs the full SIU playbook autonomously: document forensics, OSINT, statement cross-referencing, timeline reconstruction, and financial-pattern analysis - 15+ phases running in parallel, not sequentially. The output is an audit-ready investigation report a human SIU lead reviews, produced in 2-4 hours versus the 14+ days a manual investigation takes. That is autonomous investigation, not assistance.

The cleanest way to hold the difference: from fraud detection to fraud resolution. Detection is upstream; investigation is downstream. Shift Claims assists the handler as a claim moves through processing. Hesper resolves the flag once a claim has been singled out as suspicious. One compresses the handling step; the other compresses the investigation step. They are not the same step.

Use the layered model to place each product. Prevention blocks bad claims before they enter the system. Detection flags suspicious claims after first notice of loss. Investigation takes a flagged claim and resolves it with an evidence chain and a defensible report. Shift Claims operates at detection and handling - assessing, scoring, prioritizing, routing, and automating tasks while a human handler stays in the loop. Hesper operates at investigation.

This is the part competitors cannot copy onto their own byline. Shift's positioning is detection and handler workflow; ours is autonomous investigation downstream of the flag. Hesper AI sits downstream of detection, which is why it does not score or flag a single claim - it accepts flagged claims as input. There is no incumbent at the investigation layer except manual SIU teams. That is the layer Hesper is purpose-built for, and it is the one neither Shift nor any other detection vendor occupies.

The consequence is concrete. A faster handler workflow changes how quickly a claim moves through processing. It does not change which flagged claims get a full investigation, because investigation depth still depends on human SIU capacity sitting downstream. Across US P&C carriers, manual SIU teams investigate roughly 25% of flagged claims because each case takes 14+ days and an investigator carries 200+ cases. Hesper lifts that coverage from about 25% to 100% by investigating each flag in 2-4 hours.

This coverage gap is the real loss-cost lever, and it is the subject of a dedicated piece on why most flagged claims are never fully investigated. Shift Claims is complementary to that problem, not a solution to it - the handling-speed gain lands upstream of the SIU bottleneck.

The scale of the underlying problem is not handling speed. Insurance fraud steals at least $308 billion a year from American consumers, and fraud occurs in about 10% of property-casualty insurance losses, per the Coalition Against Insurance Fraud. The Insurance Information Institute puts P&C-specific fraud at roughly $45 billion and auto premium leakage at least $29 billion a year, with as much as 14% of all personal auto premium attributable to the cost of covering that leakage.

Here is a logical derivation, labeled as such, not a measured statistic: if fraud occurs in about 10% of P&C losses and a carrier's SIU fully investigates only about 25% of the claims it flags, then roughly 75% of flagged claims are paid or denied without a full investigation, regardless of how fast the handler workflow runs. Compressing handling by 30% speeds those claims through processing. It does not add investigation capacity to the 75% that never gets the depth a flag deserves.

That is the difference between two cost levers. Handling speed is an operating-efficiency lever on the adjuster side - real, but bounded by handler productivity. Flagged-claim investigation coverage is a loss-cost lever, because under-investigated flags are where fraud leakage concentrates. For a claims VP watching loss ratio on workers comp or auto bodily injury, the basis-point question lives downstream, at the investigation layer, not in handling throughput.

The unit economics follow from the cycle time. A manual SIU investigation runs about $2,500 per case at 14+ days of investigator attention; Hesper runs about $150 per case in 2-4 hours. The reason the cost collapses is structural: a human investigator's attention is the bottleneck and works one case at a time, while Hesper's 15+ phases run in parallel and its per-case attention is not a constraint. That is what makes investigating 100% of flags affordable, rather than rationing investigation down to the 25% a human team can reach.

Side by side, with the honest wins on both sides. Shift Claims is the stronger product for handler workflow and detection; Hesper is the stronger product for autonomous investigation of a flagged claim. They are not substitutes.

The detection and cross-carrier data strengths are real and worth weighting in any evaluation. For a fuller side-by-side of the data-utility layer, see Hesper vs. Verisk, which covers the same complementary framing for the industry contributory-data incumbent.

The modal stack runs both. Hesper is complementary to FRISS, Shift Technology, and Verisk - not a replacement. Shift Claims handles detection and accelerates the claim handler's workflow; the claims it flags as suspicious then flow downstream to Hesper for autonomous investigation. Hesper returns an audit-ready report - with sources, reasoning, and a documented decision trail - back into the claims system for the SIU lead and adjuster to act on.

Spelling out the handoff: a claim comes in, Shift Claims assesses and prioritizes it and the handler works it faster, and if it scores as suspicious it is referred for investigation. In a manual shop, that referral hits the SIU backlog and waits, because the team can only reach about 25% of what gets flagged. With Hesper in the stack, every referred claim gets a full 2-4 hour investigation and comes back as a report. The handler workflow and the investigation workflow each get their own purpose-built agent.

This is not human-replacement on either side. Handlers keep owning handling; the investigator's role shifts from execution to decision-making. SIU headcount does not shrink in this model - it gets re-aimed at the higher-judgment review of finished investigations rather than spent grinding through the 14+ day manual workflow one case at a time. Before you scope a deployment, the 12-point vendor evaluation checklist gives you the framework to separate the two layers in your own RFP.

The fastest way to separate handler-assist from autonomous investigation is to ask what the system outputs and who does the work after it runs. Handler-assist outputs recommendations, routing, and automated tasks a human handler still acts on. Autonomous investigation outputs a finished report an SIU lead reviews. Concrete questions that draw the line:

• Does the system investigate 100% of flagged claims, or accelerate the ones a human still works? Coverage is the loss-cost lever; handling speed is the efficiency lever.
• What is the literal output artifact - a recommendation and routing action, or an evidence chain, reconstructed timeline, statement cross-references, and a documented decision trail?
• Who acts after it runs - a claim handler who still owns the task, or an SIU lead reviewing a completed investigation?
• Does the output satisfy your documented-decision requirements, for example California 10 CCR 2698.36, and how does it appear in the antifraud plan you file under NAIC Model Act 680?
• Where does it sit in the layered model - detection and handling, or investigation downstream of the flag?

For a compliance officer, the documented-decision question is the one that stalls or clears a deal. Handler-assist actions are logged as handling steps; an autonomous investigation has to produce a fully reconstructable trail a state DOI auditor can pull and read. Hesper is audit-trail-native by design - every decision the agent makes is logged with sources, reasoning, and timestamps - precisely so a Hesper-investigated case survives that audit. The distinction in the RFP determines whether the tool moves your coverage gap or only your handling speed.