Hesper AI
BlogPillar
PillarApril 26, 2026·14 min read·Nitish Badu, COO·Updated 2026-05-01

SIU Operations: How Special Investigation Units Run in 2026

How Special Investigation Units (SIUs) run in 2026: structure, workflow, performance benchmarks, the capacity constraint, and how AI is rewriting operations. The reference for SIU directors and claims VPs.

1:200
Investigator-to-case ratio (typical)
Mid-size carriers; large carriers run 1:300-500
14+ days
Manual investigation cycle time
From referral to signed-off report
25%
Of flagged claims fully investigated
Remaining 75% closed without full review
$2,500
Average cost per manual investigation
Investigator time, vendor fees, database pulls

How SIU teams are structured

SIU structure varies significantly by carrier size and the lines of business covered. Three patterns are common.

Small carriers (under 500K policies)

Typically run a 2-5 person SIU under the claims VP. Investigators are generalists covering all lines. External vendors fill specialized capabilities (surveillance, medical peer review, forensic accounting). The team handles everything from referral to SAR filing without specialization.

Mid-size carriers (500K - 5M policies)

Run a 10-30 person SIU with line-of-business specialists. Auto investigators distinct from workers compensation fraud investigators distinct from property. A field investigation team handles surveillance and witness interviews. A document forensics specialist or contracted vendor handles document authentication. An SIU manager coordinates and reports up to claims leadership.

Large carriers (over 5M policies)

Run a 50-200+ person SIU with specialized teams: auto fraud, workers comp, medical/provider fraud, property, network/ring detection, intelligence analysts. Definitions of each role in the insurance fraud glossary. Full in-house capabilities including legal liaison and dedicated SAR filing teams. Often regional substructures matching state DOI requirements.

The investigation workflow

Across all sizes, the workflow follows a consistent six-stage structure. For the deep-dive on each stage with examples, see how insurance companies investigate fraud.

  1. Referral and triage - SIU receives the referral, manager assesses severity, assigns to investigator based on complexity and load.
  2. Case planning - assigned investigator builds the investigation plan: required evidence, databases to query, vendors to engage, anticipated timeline.
  3. Evidence gathering - the longest stage. Documents, NICB and ISO ClaimSearch queries, statement analysis, medical records, financial analysis, OSINT, field investigation as warranted.
  4. Analysis and findings - synthesize evidence into a finding (fraud confirmed, suspicious-but-insufficient, no fraud), build timeline, evaluate evidence strength.
  5. Report generation - audit-ready report with summary, evidence inventory, timeline, findings, and recommendation.
  6. Resolution and reporting - apply the recommendation (deny, pay, recover), file SAR with the state fraud bureau if fraud is confirmed.

For the canonical deep-dive on each stage, see how insurance companies investigate fraud.

Performance benchmarks for 2026

The 10 investigations per investigator per month benchmark is now obsolete. With autonomous AI investigation as a force multiplier, top performers run 50-100+ investigations per month per investigator while maintaining quality. For the full benchmark framework see SIU performance benchmarking 2026.

MetricManual baselineTop quartile (with AI)
Investigations per investigator per month8-1250-100+
Cycle time per investigation14+ days2-4 hours
Coverage rate (flagged claims investigated)20-30%85-100%
Confirmed fraud rate (of investigated)30-45%55-70%
Cost per investigation$2,500$150-400
Investigator overtime hours per month20-400-5

The capacity constraint

SIU operations are constrained by investigator capacity in nearly every carrier. The math is unavoidable: a typical investigator can complete 8-12 investigations per month working full-time. A mid-size carrier with 2,000 monthly flagged claims would need 200+ investigators to investigate them all manually. No carrier staffs to that ratio - the cost is prohibitive.

The standard response has been triage - investigate the highest-value claims, close the rest. This works for the very top of the queue but leaves substantial leakage in the middle and bottom. For the leakage analysis, see insurance claims leakage. For the underlying operational analysis of why this happens at scale, see why 75% of flagged insurance claims are never investigated.

How leading teams clear the backlog

Five interventions, applied in priority order, transform SIU throughput. Each compounds on the prior ones.

  1. Documented triage criteria - explicit rules for which flagged claims investigators must work on first, removing per-investigator variance.
  2. Documentation automation - report templates and one-click drafting reclaim 1-2 cases per investigator per month.
  3. Database query automation - parallel NICB, ISO ClaimSearch, LexisNexis lookups reclaim 2-3 cases per investigator per month.
  4. Autonomous AI investigation - the largest-impact intervention; investigator throughput rises from ~10 to 50-100+ cases per month.
  5. Detection precision tuning - reduces false positive volume so investigators work higher-confidence cases.

For the operator-level guide to clearing the backlog, see a claims investigator's guide to clearing the high-volume backlog.

Where AI changes operations

AI changes SIU operations at three layers: investigation execution, report generation, and triage decisions.

  • Investigation execution - autonomous agents run the 15+ investigation phases in parallel, compressing 14+ days into 2-4 hours. See parallel processing in SIU.
  • Report generation - audit-ready reports written by the agent during investigation; investigator review is 30-60 minutes vs 4-8 hours of manual writing. See audit-ready fraud reports.
  • Triage decisions - AI-assisted triage uses signal density (red flag combinations, claim characteristics) to prioritize the queue beyond what rules alone can do. See three weeks to four hours.

Key takeaways

  • SIU structure scales with carrier size: 2-5 generalists at small carriers, 10-30 specialists at mid-size, 50-200+ at large.
  • The six-stage workflow (referral, planning, evidence, analysis, report, resolution) is consistent across sizes.
  • Performance benchmarks have shifted: 50-100+ investigations per investigator per month is the new top quartile, not 10.
  • The capacity constraint is structural: typical carriers can investigate ~25% of flagged claims manually.
  • Five compounding interventions clear the backlog; autonomous AI investigation is by far the largest lever.
  • AI changes operations at investigation execution, report generation, and triage decisions.

Frequently asked questions

Small carriers (under 500K policies) typically run 2-5 person SIUs. Mid-size carriers (500K-5M policies) run 10-30 person SIUs with line-of-business specialists. Large carriers (over 5M policies) run 50-200+ person SIUs with specialized teams for auto, workers comp, medical/provider, property, and network analysis. The investigator-to-case ratio ranges from 1:200 (mid-size) to 1:300-500 (large).

Manual SIU investigations average $2,500 per case in 2026, accounting for investigator time (60% of cost), external vendor fees for surveillance and forensic services (25%), and database query fees (15%). Complex cases with multi-day surveillance can run $5,000-15,000. Autonomous AI investigations cost $150-400 per case at typical pricing.

It is a capacity math problem, not a process failure. With one investigator per 200+ cases and a 14+ day manual investigation cycle, the throughput is approximately 8-12 investigations per investigator per month. Detection systems generate flags at much higher volumes - typically 5-10x what investigators can process. Triage closes the gap: high-value and high-signal claims get investigated, the rest are closed without full review.

Manual SIU investigations average 14+ days from referral to signed-off report, with significant variance by complexity. Simple inflated estimate cases close in 5-7 days; complex fraud rings or contested workers comp claims take 30-45 days. Autonomous AI investigation completes in 2-4 hours with the same evidentiary standard, because the 15+ investigation phases run in parallel.

The traditional 10 investigations per investigator per month benchmark is obsolete. The 2026 top-quartile benchmark with autonomous AI is 50-100+ investigations per investigator per month while maintaining or improving quality (confirmed fraud rate of 55-70% vs 30-45% historical). Coverage rate (flagged claims investigated) shifts from 20-30% to 85-100%. The change is not human productivity improving; it is autonomy doing the heavy execution while the investigator focuses on judgment.

Most carriers redirect rather than reduce. AI raises throughput per investigator dramatically, but the better strategic move is to investigate the previously-uninvestigated 75% rather than reduce a team that could now achieve 100% coverage. The savings come from reduced fraud paid out, not from reduced staff. Headcount neutral with 4-5x throughput is the typical pattern.

← 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.