The investigation gap nobody talks about
The insurance industry has a fraud detection problem. But it is not the problem most people think it is.
Every major insurer has invested in fraud detection. Rule engines. Predictive scoring models. Analytics platforms from vendors like FRISS, Shift Technology, and Verisk. These systems flag suspicious claims. They score risk. They generate alerts. And by that metric, they work. According to the Coalition Against Insurance Fraud, the industry spends over $5.3 billion annually on anti-fraud technology and operations.
But here is the uncomfortable truth: flagging is not investigating. And 75% of the claims these systems flag are never fully investigated.
The gap is not in detection. Detection is largely solved. The gap is in what happens after the flag. Most flagged claims enter an SIU queue and sit there. Investigators are overloaded. Cases expire. Claims get paid. Fraud walks out the door with a check.
“We flag plenty of claims. The problem is we can only investigate a fraction of them. The rest just age out of the queue.”
- SIU Director, Top-20 US P&C Insurer (2025 industry survey)
This investigation gap is the single largest source of preventable fraud loss in the insurance industry. It is not a technology gap - it is a capacity gap. And it has been hiding in plain sight. For broader context on the scale of insurance fraud, see our insurance fraud statistics for 2026.
The SIU capacity crisis
Special Investigation Units are the frontline of insurance fraud investigation. Every major carrier has one. They are staffed with experienced investigators - former law enforcement, claims adjusters with decades of experience, certified fraud examiners. These are skilled professionals doing essential work.
The problem is math. There are approximately 15,000 SIU investigators working across the US insurance industry. Each one handles 200 or more cases per year. Each case takes 3 to 10 days of manual investigation - pulling records, interviewing claimants, cross-referencing documents, writing reports. The throughput ceiling is hard and fixed.
Meanwhile, the volume of flagged claims keeps growing. Better detection models flag more suspicious claims. AI-generated deepfake insurance claims are increasing the overall fraud rate. The result is a widening gap between what is flagged and what gets investigated.
SIU investigation capacity vs flagged claim volume
Look at those numbers. For every 100 claims flagged as suspicious, only 25 receive a full investigation. The other 75 are either deprioritized, partially reviewed, or simply paid out because the queue moved on.
The triage trap
SIU teams triage by dollar amount. High-value claims get investigated first. But fraud rings operate at scale with mid-value claims precisely because they know SIU teams will never get to them. The triage strategy creates a predictable blind spot that organized fraud exploits systematically.
What happens to uninvestigated claims
When a flagged claim is not investigated, it does not disappear. It follows one of three paths - and none of them are good.
- Paid as filed. The claim ages in the queue, regulatory pressure or claimant complaints escalate, and the adjuster pays it to clear the backlog. The flag is noted in the file but never acted on.
- Partially reviewed. An investigator spends 30 minutes scanning the file, finds nothing immediately disqualifying, and closes the case as inconclusive. The fraud - if it exists - is never uncovered because no deep investigation occurred.
- Denied without investigation. In rare cases, a claim is denied based solely on the flag score without supporting investigation. This creates E&O exposure, regulatory risk, and potential bad faith litigation.
The National Insurance Crime Bureau (NICB) estimates that approximately 10% of all property and casualty claims contain some element of fraud. On a $900 billion annual P&C premium base, that translates to roughly $45 billion in fraudulent claims every year in the US alone.
The industry spends $5.3 billion fighting fraud and still loses $45 billion. That is a 9:1 loss ratio on anti-fraud spend. The bottleneck is not detection. It is investigation throughput.
The compounding effect
Uninvestigated fraud does not just cost money on the individual claim. Fraud rings learn which claim types and dollar thresholds avoid investigation. They optimize their operation around your SIU capacity constraints. Every uninvestigated claim is a signal to fraudsters that the playbook works.
According to the FBI IC3 Annual Report, insurance fraud is one of the fastest-growing categories of financial crime. The bureau notes that organized fraud rings - not individual opportunists - account for the majority of losses. These rings are sophisticated enough to study SIU patterns and adjust their submissions accordingly.
The economics of manual investigation
To understand why the investigation gap persists, follow the money. Manual fraud investigation is expensive, slow, and does not scale.
The direct cost of SIU investigator labor is approximately $1.5 billion annually. That is roughly 15,000 investigators at an average fully loaded cost of $100,000 per year. On top of that, insurers spend an estimated $3.0 billion on third-party investigation firms - private investigators, forensic accountants, document examiners, and surveillance contractors.
Annual anti-fraud spend breakdown ($B USD)
That is $5.3 billion spent fighting fraud - and the industry still loses $45 billion. For every dollar spent on anti-fraud operations, $8.50 in fraud still gets through. The return is not the problem. SIU investigations that are completed have a strong ROI. The problem is that 75% of flagged claims never receive that investigation.
The unit economics are clear. A single SIU investigation takes 3 to 10 days and costs $1,500 to $5,000 when you factor in investigator time, document retrieval, database access fees, and report generation. At that cost, insurers cannot afford to investigate every flagged claim. So they triage. And triage means most claims never get looked at.
Compare this to the detection side. Modern fraud scoring platforms process a claim in seconds for pennies. The Deloitte 2025 insurance predictions report highlighted that AI-driven detection has reached near-maturity. The bottleneck has shifted entirely downstream - from detection to investigation.
“We have more flagged claims than we know what to do with. The detection models work. What we need is a way to actually investigate at scale.”
- VP of Claims, Regional P&C Carrier
From flagging to investigating: closing the gap with AI
The investigation gap exists because investigation has been a purely human activity. Detection can be automated. Flagging can be automated. Scoring can be automated. But the actual investigation - pulling records, cross-referencing documents, verifying claims, building a case narrative - has required an experienced human sitting at a desk for days.
That is changing. AI investigation agents can now perform the core activities of a claims investigation end-to-end. Not scoring. Not flagging. Actual investigation - document verification, cross-referencing, timeline reconstruction, anomaly detection, and report generation.
This is the critical distinction. The insurance industry does not need another fraud score. It does not need another alert. It needs investigation throughput. For context on how AI-generated documents are complicating claims further, see our analysis of ChatGPT and deepfake documents in insurance fraud.
The AI investigation pipeline works in five stages:
- Ingest - the AI agent ingests the full claim file: documents, images, adjuster notes, policy data, claimant history, and third-party records.
- Detect - the agent runs fraud detection across all submitted documents, identifying manipulated images, forged documents, and inconsistent data points.
- Verify - every claim element is verified against external sources. Medical records are cross-referenced. Repair estimates are benchmarked. Identities are validated.
- Investigate - the agent builds a complete investigation narrative. Timelines are reconstructed. Anomalies are documented. Patterns are matched against known fraud typologies.
- Resolve - a structured investigation report is generated with findings, evidence references, confidence scores, and recommended actions. Ready for SIU review and decision.
Every claim gets its own AI investigator. Not a score. Not a flag. A full investigation. What takes a human investigator 14 days is completed in 60 minutes.
Verify, not just flag. Investigate, not just score. Explain, not just alert.
Existing fraud detection vendors - FRISS, Shift Technology, Verisk - stop at the flag. They tell you a claim looks suspicious. They do not tell you why, verify the underlying documents, or build an investigation case. Hesper AI starts where they stop. For a deeper comparison of detection approaches, see our document fraud detection software guide.
Investigation time per claim: manual vs AI
What SIU teams should do next
The investigation gap is not going to close itself. Hiring more investigators is slow, expensive, and does not scale with the growing volume of flagged claims. Third-party firms are a cost multiplier, not a solution. The only way to investigate every flagged claim is to change the unit economics of investigation itself.
Here is what SIU teams should evaluate:
- Audit your investigation completion rate. How many flagged claims actually receive a full investigation? If the number is below 50%, you have a capacity problem that technology can solve.
- Quantify your investigation gap in dollars. Take your flagged-but-uninvestigated claim volume, multiply by your average fraud amount, and multiply by your estimated fraud prevalence rate. That number is your preventable loss.
- Evaluate AI investigation agents - not more scoring tools. The market is saturated with fraud scoring platforms. What you need is investigation throughput. Ask vendors whether they verify, investigate, and explain - or just flag and score.
- Run a parallel pilot. Deploy an AI investigation agent alongside your existing SIU workflow on a subset of flagged claims. Measure investigation quality, time-to-resolution, and fraud recovery rates side by side.
- Rethink your triage strategy. If AI investigation can handle every flagged claim, you no longer need to triage by dollar amount. Your SIU team can focus on complex cases, fraud ring identification, and law enforcement coordination.
The data is clear. Detection is solved. Investigation is the bottleneck. The insurers that close the investigation gap first will recover billions in fraud losses that are currently walking out the door. For a broader view of the fraud landscape driving these losses, see our document fraud statistics for 2026.
Every flagged claim investigated. Not just scored.
Hesper AI deploys AI investigation agents that don't just flag fraud - they investigate it end-to-end. Every claim gets its own AI investigator. 14 days of manual work compressed to 60 minutes. Learn more at gethesperai.com.
Key takeaways
- 75% of flagged insurance claims are never fully investigated due to SIU capacity constraints - not detection failures.
- The insurance industry spends $5.3B on anti-fraud operations but still loses $45B annually to P&C fraud alone.
- Approximately 15,000 SIU investigators handle 200+ cases each per year, with each case requiring 3-10 days of manual work.
- Fraud scoring tools from FRISS, Shift Technology, and Verisk flag suspicious claims but do not investigate them - the gap is downstream of detection.
- AI investigation agents can complete a full claims investigation in 60 minutes - compressing 14 days of manual work and reducing cost from $1,500-$5,000 to under $50 per case.
- Organized fraud rings exploit the investigation gap by operating at claim volumes and dollar thresholds they know SIU teams will never reach.