Hesper AI
BlogResearch
ResearchApril 15, 2026·10 min read·Hesper AI Threat Research

Benchmarking SIU performance in 2026: why 10 investigations per month is obsolete

The industry benchmark of 10 investigations per investigator per month was built around manual evidence gathering and report writing. Autonomous AI investigation shifts the ceiling to 800+. New benchmarks for carrier size, investigation coverage, and cost per case.

~10
Industry standard investigations per investigator per month
Manual workflow ceiling, 2020-2025
800+
New ceiling with autonomous AI investigation
Review-oriented investigator throughput, 2026+
25%
Current investigation coverage of flagged claims
Remaining 75% closed without full investigation
100%
Coverage achievable at current SIU headcount
With autonomous investigation agents

Every SIU leader in the US P&C industry uses the same benchmark: an investigator handles 200 active cases and closes about 10 investigations per month. The number is worn smooth by repetition. It is the ceiling around which SIU headcount plans, loss ratio projections, and fraud-recovery budgets are built.

The benchmark is obsolete. It was built around a manual workflow - 14+ days of evidence gathering, 4-8 hours of report writing - that autonomous AI investigation has fundamentally changed. The new ceiling is 800+ investigations per investigator per month, 80x the legacy standard.

This piece lays out the core SIU performance metrics, the benchmarks that defined the last five years, and the 2026 benchmarks that account for AI-augmented workflows. The intent is practical: SIU leaders rebuilding their capacity plans need new numbers, and the industry needs a shared baseline.

Why benchmark SIU performance

SIU performance benchmarking serves three purposes: board-level accountability (how is fraud cost changing?), internal capacity planning (can we investigate everything we flag?), and vendor evaluation (is this new tool actually moving the numbers?). Without shared benchmarks, every conversation about SIU effectiveness defaults to anecdote.

For most of the last decade, the shared benchmarks came from Coalition Against Insurance Fraud surveys and NAIC annual reports. The numbers described a system that was investigating ~25% of what it flagged at a cost of approximately $2,500 per investigation. Those numbers held steady because the workflow held steady. They are now moving.

The core metrics

Seven metrics define SIU performance. Track them quarterly at minimum:

  1. Flagged claim volume - total claims referred to SIU per period. Driven by detection quality and claims volume.
  2. Investigation coverage rate - percentage of flagged claims that receive full investigation. The headline throughput metric.
  3. Cases per investigator per month - investigations completed, a per-FTE productivity measure.
  4. Time-to-close - average days from referral to final report. Measures workflow efficiency.
  5. Confirmed fraud rate - percentage of investigated claims where fraud is substantiated.
  6. Recovery rate - dollars recovered or denied relative to investigated claim value.
  7. Cost per investigation - fully loaded per-case cost (investigator time + vendor fees + technology).

These seven metrics interact. Investigation coverage depends on cases per investigator. Confirmed fraud rate depends on referral quality and investigation depth. Cost per investigation depends on workflow efficiency. Moving any one meaningfully requires addressing the others - which is why benchmark changes tend to arrive together rather than incrementally.

Current benchmarks (2020-2025)

Metric2020-2025 benchmarkDriver
Investigator caseload200 active casesCarrier staffing norms
Cases closed / month~1014+ day manual workflow
Investigation coverage25% of flaggedCapacity constraint
Time per investigation14+ daysEvidence gathering + report writing
Confirmed fraud rate30-50% of investigatedReferral quality + depth
Cost per investigation~$2,500Investigator time + vendors
Recovery rate3-8x costDeny + claw-back

These numbers are stable across the industry. Variance exists - top-20 carriers are marginally more efficient than mid-size regionals; line of business matters - but the central tendency holds. They describe what manual investigation capacity looks like.

2026 benchmarks with AI investigation

Metric2020-20252026 with AI investigationChange
Investigator caseload200 active200+ active (review-oriented)Unchanged
Cases closed / month~10800+80x
Investigation coverage25% of flagged100% of flagged4x
Time per investigation14+ days2-4 hours~95% faster
Confirmed fraud rate30-50%40-60% (broader base)Modest lift
Cost per investigation~$2,500~$150~94% lower
Recovery rate3-8x cost5-12x costModest lift

Throughput and cost changes are the headline. Confirmed fraud rate and recovery rate changes are more modest, because the underlying fraud density in the flagged population is roughly constant. What changes is how much of the flagged population actually gets investigated. Cases that were previously closed without investigation are now being investigated, which widens the base.

Cases per investigator per month - legacy vs AI-augmented

Manual SIU (2020-2025 benchmark)~10
Manual SIU with better tooling (2024-2025 best-in-class)~40
Autonomous AI investigation + human sign-off (2026+)800+

Benchmarks by carrier size

Throughput expectations scale with carrier size, but ratios are stable. Smaller carriers face a steeper economic hurdle to add investigation capacity, which is precisely why AI investigation has a disproportionately large impact on mid-size and specialty lines:

Carrier sizeSIU team sizeAnnual flagged claimsManual investigations / yearAI-augmented / year
Top-20 carrier30-100 investigators50K-200K+5K-15K50K-200K+ (full coverage)
Mid-size regional5-15 investigators5K-20K500-2K5K-20K (full coverage)
Small / specialty1-3 investigators500-5K50-300500-5K (full coverage)

Closing the 75% gap

The most important consequence of the new benchmarks is the collapse of the coverage gap. For a decade, the SIU industry has operated under the assumption that ~75% of flagged claims will not receive full investigation because there is no capacity to investigate them. That assumption is no longer accurate.

When cases-per-investigator shifts from 10 to 800+, the constraint moves from investigator capacity to detection quality. If your detection platform flags 10,000 claims per year and you have 5 investigators who can each handle 800+ cases per month, you can investigate everything that is flagged. The remaining question is whether the flagging is right.

The new bottleneck is detection precision

Investigating 100% of flagged claims matters only if the flags are meaningful. Rules-based detection runs at 60-85% false positive rates. With autonomous investigation, the low cost of investigating a false positive (a few minutes of human review) makes the false positive rate manageable - but the throughput wave will also put pressure on detection platforms to improve signal quality.

For the downstream economic impact of closing the coverage gap, see insurance claims leakage: how uninvestigated claims drain profitability. For the architectural comparison of detection vs investigation, see legacy rules-based systems vs. autonomous AI.

Key takeaways

  • The 2020-2025 SIU benchmark of 10 investigations per investigator per month was a manual-workflow ceiling, not a fundamental limit.
  • 2026 benchmark with autonomous AI investigation: 800+ investigations per investigator per month (80x), 100% flagged-claim coverage (4x), 2-4 hour time-to-close (95% faster), ~$150 per investigation (94% lower cost).
  • Track seven core metrics quarterly: flagged volume, investigation coverage, cases per investigator, time-to-close, confirmed fraud rate, recovery rate, cost per investigation.
  • Investigation coverage gap closes from 25% to 100% at current SIU headcount. The new bottleneck becomes detection quality, not investigation capacity.
  • Benchmarks scale with carrier size but ratios are stable. Mid-size and specialty carriers see disproportionate benefit because they face the steepest economic hurdle to manual SIU expansion.

Frequently asked questions

The long-standing benchmark is approximately 10 investigations completed per investigator per month, with each investigator carrying a caseload of 200 active cases. This benchmark was stable across the US P&C industry from approximately 2015 through 2025, driven by the 14+ day manual investigation workflow and 4-8 hour report writing per case. With autonomous AI investigation agents (like Hesper AI) handling evidence gathering, document forensics, and report generation, the ceiling shifts to 800+ investigations per investigator per month - an 80x change. The investigator's role shifts from execution to decision-making and sign-off.

Approximately 25% of claims flagged by fraud detection systems receive full SIU investigation. The remaining 75% are closed with abbreviated review or without investigation entirely, driven by investigator capacity constraints (~10 cases per investigator per month under manual workflows). With autonomous AI investigation agents that compress investigation time from 14+ days to 2-4 hours, investigation coverage can rise to 100% of flagged claims at current SIU headcount. The constraint shifts from capacity to detection quality.

A manual SIU investigation costs approximately $2,500 per case on a fully loaded basis, including investigator time (40-80 hours), vendor fees (surveillance, medical peer review, forensic accounting), and database access fees. Autonomous AI investigation reduces per-case cost to approximately $150 while completing the investigation in 2-4 hours. The cost reduction comes primarily from automating evidence gathering, document forensics, database queries, and report generation - the tasks that consume ~88% of investigator time in the manual workflow.

Seven core metrics: (1) flagged claim volume, (2) investigation coverage rate (% of flagged receiving full investigation), (3) cases closed per investigator per month, (4) time-to-close (days from referral to final report), (5) confirmed fraud rate (% of investigated claims where fraud is substantiated), (6) recovery rate (dollars denied or recovered relative to investigated value), (7) cost per investigation (fully loaded). Track quarterly at minimum. Current (2020-2025) manual benchmarks: 25% coverage, 10 cases/month per investigator, 14+ days, 30-50% confirmation, 3-8x recovery, $2,500 per case. 2026 AI-augmented benchmarks: 100% coverage, 800+ cases/month, 2-4 hours, 40-60% confirmation, 5-12x recovery, $150 per case.

Yes, significantly. The 2020-2025 benchmark of ~10 investigations per investigator per month was driven by the manual workflow ceiling: 14+ day investigations with 4-8 hour report writing per case. Autonomous AI investigation agents (Hesper AI and similar platforms) run 15+ investigation phases in parallel and produce audit-ready reports automatically. The investigator role shifts from executing evidence gathering to reviewing AI-generated findings and signing off. Throughput rises from ~10 to 800+ cases per month, cost drops from ~$2,500 to ~$150 per case, and investigation coverage expands from 25% of flagged to 100%.

← More articles on the Hesper AI blog

See Hesper AI on your documents

Request a demo and we'll run an analysis on your real document samples.