The claims process end-to-end: from FNOL to settlement

The insurance claims process runs through eight stages, but the way it is usually taught - as a linear checklist from First Notice of Loss to settlement - hides the two things that actually decide the outcome. From the moment a claim is reported, two meters start running at once: cycle time, which the customer watches and the reserve pays for, and leakage, which never shows up as its own line item but quietly erodes the loss ratio. This post maps all eight stages and shows where those two meters run fastest.

The structural argument is simple. Carriers have spent a decade automating the front of the process. Digital FNOL intake, automated triage, systematic coverage verification, and computer-vision estimating have collapsed those stages from days to minutes. But automation has not reached one stage - investigation - which is still measured in 14+ days per case. That is where the process stalls, and it is where the money leaks. You cannot compress the claims process end-to-end while one stage stays manual.

This is the operational reality behind the loss-ratio and cycle-time numbers a Claims VP reports up. For how automation maturity varies stage by stage, see our companion piece on what is automated and what is still manual in 2026. This post is part of our claims fraud leakage cluster, which tracks where uninvestigated claims drain profitability.

The claims process is best read as two meters running in parallel from FNOL: a clock measuring cycle time and a ledger measuring leakage. The clock is what the customer feels and what a Claims VP is measured on; every day a reserve sits open costs the carrier. The ledger is overpayment, missed exclusions, and un-investigated fraud that never surfaces as its own number. Both are set at intake and paid at settlement.

The clock is not abstract. Per the J.D. Power 2024 U.S. Auto Claims Satisfaction Study, the average auto repair cycle time was 22.3 days, down 1.0 day from 2023, and "time to settle claim" is one of the eight dimensions the study uses to score satisfaction. Cycle time is a customer-satisfaction driver, not just an internal metric. Every stage in the process is a handoff, and every handoff adds days to that 22.3-day figure.

The ledger is the quieter meter. Per the Coalition Against Insurance Fraud, insurance fraud steals at least $308.6 billion a year from American consumers, and fraud occurs in about 10% of property-casualty losses. Beyond outright fraud, the Insurance Information Institute notes that as much as 14 percent of all personal auto premium can be attributed to premium leakage. The clock and the ledger collide at exactly one stage - investigation - and that collision is the spine of this post.

First Notice of Loss (FNOL) is the moment the policyholder reports the claim, and triage is where that claim gets routed to the right handler and priority. Together they start the settlement clock and produce the first fraud signals. Every decision made here - severity, complexity, routing - sets the speed and the exposure for everything downstream, which is why intake quality compounds through the whole process.

Both stages are heavily automated now. Digital FNOL intake through web, mobile, or telematics captures the loss in minutes, and the J.D. Power study tracks "ease of starting claim" as one of its satisfaction dimensions precisely because carriers have invested there. Automated triage scores incoming claims for severity and routing, and for a full breakdown of how that routing works, see our claims triage automation guide.

The first fraud signals appear at intake, not investigation

Triage is also where suspicious claims get flagged. This is where the leakage ledger opens: a claim scored as low-risk and fast-tracked can be exactly the one that needed a second look. Rules-based flagging carries a real cost here - 60-85% of what a rules layer flags is not fraud unless the rules are heavily tuned - so triage generates noise that flows straight into the investigation queue. Hesper has built-in fraud detection, so it can surface signals at intake rather than depend entirely on a bolt-on scoring layer, though the deeper point is what happens to a flag once it is raised.

Coverage verification confirms the policy actually responds to the loss, and evaluation and reserving set the estimated exposure and open a reserve. These two stages fix the financial size of the claim. Coverage checks catch exclusions and policy limits; reserving is the carrier's first formal estimate of what the claim will cost, and it drives capital allocation across the book.

Automation has compressed this middle of the process hard. Systematic coverage verification runs against policy data in seconds, and computer-vision estimating tools assess damage in minutes rather than days. That is genuine progress - but it has a side effect. When the front of the process gets fast, the bottleneck does not disappear; it moves downstream. Automating coverage and estimating just relocates the drag to the one stage that stayed manual. For where that leaked money accumulates across stages, see our analysis of how uninvestigated claims drain profitability.

The exposure set here is where leakage first becomes measurable. The III attributes as much as 14 percent of personal auto premium to leakage, and the Insurance Information Institute puts annual property and casualty fraud at $45 billion, workers compensation fraud at $34 billion, and auto theft fraud at $7.4 billion. Every dollar of that either gets caught at investigation or gets paid at settlement. The stage that decides which is the one still running on human time.

Investigation is where a flagged or complex claim gets examined for fraud and validity - document forensics, statement cross-reference, timeline reconstruction, financial pattern analysis - and it is the one stage in the process still measured in weeks. A manual Special Investigations Unit review takes 14+ days per case, while every stage around it has been compressed to minutes. This is where the clock stalls and the ledger bleeds.

The bottleneck is structural, not a staffing accident. A human investigator works one case at a time because attention is the constraint, and a typical investigator already carries 200+ open cases. Because capacity is finite, only about 25% of flagged claims get fully investigated; the rest are paid, denied without full work, or queued indefinitely. With roughly 10% of P&C losses involving fraud and $308.6 billion lost annually, the un-worked 75% is the largest single leak in the process.

Automating the investigation stage is what closes the loop. AI investigation agents run 15+ investigation phases in parallel on every flagged claim and return an audit-ready report in 2-4 hours instead of 14+ days. That is the difference between a partially automated process and an end-to-end one. It also changes the coverage math structurally: from about 25% of flagged claims to 100%.

The economics follow the coverage. Manual investigation runs about $2,500 per case; an AI-run investigation runs about $150. Throughput moves from roughly 10 investigations per investigator per month to 800+. The investigator's role shifts from execution to decision-making - reviewing an audit-ready report and handling exceptions instead of assembling every case by hand. This is the move from fraud detection to fraud resolution: Hesper has built-in fraud detection, not just a downstream position of other tools, so it takes a flag and resolves it end-to-end.

Negotiation settles the amount with the claimant, settlement pays the claim, and closure documents the outcome and closes the file. These final three stages are where the clock and the ledger get paid. Faster, fuller investigation changes all three - not just the SIU spend line - because it changes the information available before money goes out the door.

When investigation takes 14+ days, reserves sit open longer, settlement is delayed, and recovery windows narrow. Compressing investigation to 2-4 hours lets a carrier set more accurate reserves earlier, settle valid claims faster, and deny or pursue fraudulent ones before payment. Settlement speed and fairness are customer-facing: the J.D. Power study ranks "time to settle claim" and "fairness of settlement" among its core dimensions, and 48% of respondents reported a premium increase in the prior 12 months - a reminder that leaked loss cost eventually reaches the customer as price.

Closure quality also improves. An audit-ready investigation report flows back into the claims management system as a case attachment, so the documented-decision trail is complete when the file closes. Investigating 100% of flagged claims instead of 25% means fewer fraudulent claims settle by default simply because the SIU ran out of capacity to look. The compression at one stage propagates through every stage after it.

Laid out side by side, the eight stages make the bottleneck visible. Most stages have credible automated durations measured in minutes to hours; only investigation carries a hard, sourced gap on both sides - 14+ days manual versus 2-4 hours augmented - and the 22.3-day total cycle time anchors the whole table. The leakage column shows where money escapes at each handoff.

Read the investigation row against the rest of the table. Every other stage has an automation-augmented duration in minutes to hours; investigation is the only row where the manual side is measured in weeks. That single row is why the 22.3-day average cycle time does not fall further even as intake and estimating get faster - the drag has concentrated in the one stage automation has not reached.

Mapping the vendor landscape onto the stage model shows that the investigation stage is the layer no product occupies. Claims management systems own the workflow rails, estimating vendors own evaluation, and detection vendors own the flagging at and after FNOL. Investigation - taking a flag and resolving it end-to-end - has no incumbent except manual SIU teams.

Claims management systems like Guidewire ClaimCenter and Duck Creek Claims run the FNOL-to-settlement workflow and increasingly automate intake and triage. Hesper integrates into those systems as the investigation stage: a flagged claim flows out, and an audit-ready report flows back as a case attachment. Hesper never replaces the claims management system.

Estimating vendors like Tractable and CCC compress evaluation to minutes for damage assessment; they sit upstream of investigation and are complementary, not competitive. Detection vendors like FRISS, Shift Technology, and Verisk flag suspicious claims at and after FNOL. Detection is upstream; investigation is downstream. Hesper takes the flag those tools raise and resolves it - complementary to FRISS, Shift Technology, and Verisk, not a replacement. Because Hesper also has built-in fraud detection, it can operate standalone if a carrier has no detection vendor on a given line.