Outsourced insurance investigation services vs autonomous AI: the demand thesis is right, the labor pool is 39,500 people

In July 2026, Command Investigations acquired CoventBridge's insurance division and put more than 900 investigators across all 50 states under one roof. It was the tenth business Command had absorbed, and the third time in ten years that a private equity sponsor bet on the same thesis: there is far more demand for claim investigation than there is organized capacity to supply it. Harwood Private Equity made that bet in 2016, Incline Equity Partners made it in March 2025, and Command made it again this summer. The thesis is correct.

The disagreement is not about demand. It is about the shape of the supply curve. Outsourced insurance investigation services are priced by the assignment and the investigator-hour, which makes capacity linear in licensed human headcount, inside an occupation the Bureau of Labor Statistics counts at 39,500 people nationally. Consolidation makes that workaround more reliable, better governed and easier to buy. It does not change the slope of the line.

What follows is the public evidence: the anchor transaction, the referral-to-investigation gap in the two states that publish it, the procurement documents that show how an assignment gets priced, the labor pool every carrier bids into, and the work that stays with the field firms no matter what software does. If you want the coverage-gap arithmetic on its own first, start with why 75% of flagged insurance claims are never fully investigated.

On 7 July 2026, Command Investigations announced it had acquired CoventBridge's insurance division. The combined business operates as CoventBridge Insurance Investigations, with more than 900 investigators across all 50 states, and the companies describe it in their own announcement as the largest insurance investigations organization in the United States. Command has now combined ten organizations. The acquired division served more than 200 insurance carriers, third-party administrators and self-insureds, and its SIU specialists average 15 years of experience. Terms were not disclosed. Insurance Journal covered the transaction on 9 July.

The lineage matters more than the headline. Incline Equity Partners bought Command in March 2025 and stated the plan in the announcement: Tom Ritchie, a partner at Incline, called Command a comprehensive provider in a fragmented industry and said the firm would pursue accelerated M&A. CoventBridge itself was built the same way. Harwood Private Equity formed it in March 2016 by merging ICS Merrill and GlobalOptions, and described the result at the time as the largest outsource provider of insurance claims investigative services in North America. Same claim, same method, ten years apart.

The pattern runs past investigation into claims services generally. In November 2025 Davies Group agreed to acquire SCM Insurance Services, Canada's largest claims processing and risk solutions business, in a deal that leaves the combined business with annual revenues of about $1.4bn, roughly 9,500 people across 22 countries and more than 1,500 SCM personnel joining. That is adjacent rather than directly comparable, claims adjusting and TPA work rather than investigation. It is the same bet: assemble people, sell their hours at scale.

Read the signal plainly. Three sponsors over ten years have concluded that carrier demand for claim investigation exceeds the industry's organized capacity to supply it, and they have committed capital to that conclusion. They are right, and the carriers buying from them are responding rationally to a queue they cannot staff.

What the deal buys is worth naming precisely, because the acquirers named it themselves. It buys coverage breadth, one panel across 50 states, a dedicated quality-assurance unit integrated with case-management technology, and one vendor relationship instead of twenty. Those are real operational wins. Seth Markham, Command's president and CEO, described the combination as the broadest investigator coverage in the country plus technology that makes every engagement faster and more precise. Note the unit of value in that sentence: the engagement. A carrier with 5,000 flagged claims and budget for 400 engagements has a problem that no improvement in per-engagement quality solves.

New York publishes the arithmetic. The Department of Financial Services Insurance Frauds Bureau received 52,105 reports of suspected fraud in 2024 and opened 316 cases for investigation. That is six-tenths of one percent of reports. The Bureau made 227 arrests, referred 148 cases to prosecutors and obtained 145 convictions. Roughly three-quarters of the 2024 reports were suspected no-fault. This is a state agency with a criminal mandate rather than a carrier SIU, so the ratio is not directly transferable. It is still the cleanest published measurement of the distance between suspicion and investigation anywhere in the industry.

The five-year series is the part to keep. Reports received went from 30,113 in 2020 to 52,105 in 2024, a 73% increase. Investigations opened went from 324 to 316 over the same period, down about 2%. The annual queue is now more than 20,000 reports larger than it was in 2020, and the number of investigations opened did not move. Capacity in that system is effectively a constant, and constants do not respond to demand.

California reports the same shape from a different mandate. The Department of Insurance Fraud Division received 12,559 suspected fraudulent automobile claims in fiscal year 2023-24 and assigned 602 new cases, roughly 4.8%, against $207,629,944 in potential loss. Its workers' compensation program identified 2,932 suspected fraudulent claims and assigned 291 new cases, against $157,201,942 in potential loss. Different state, different line, same band.

Carriers face the identical arithmetic one book at a time, and the internal version has no annual report. Manual SIU teams across US P&C carriers investigate roughly 25% of flagged claims; the rest are paid, denied without full work, or left open in a queue. An investigator carrying 200+ cases closes somewhere near ten investigations a month. Multiply that by panel budget divided by panel rate and you have the real capacity number, and it is set before anyone reads a file. The panel is the release valve on that arithmetic, and the reason the coverage ratio is 25% rather than something worse. For the full mechanics of how the other 75% forms, see the coverage-gap breakdown.

Outsourced insurance investigation services are third-party field and desk investigation work that carriers buy by the assignment rather than by the claim: surveillance, recorded statements, activity checks, neighborhood and medical canvasses, scene investigation, record retrieval, social media research and background checks. The service-line menu is the product, and every line is priced and dispatched separately.

Command's services page lists ten lines: Surveillance, Remote Surveillance, Field Investigations, Intelligence, SIU and Anti-Fraud, CEU/CLE Training, Vendor Management, Litigation Support Services, Simplified Screening Solutions and Speciality Services. Underneath them sit the activities a carrier buys one at a time: recorded statements, scene investigations, evidence gathering, interviews, medical canvassing, DMV searches, criminal and civil record reviews, asset and employment checks, background investigations, social media analysis, SSN validation and professional license verifications.

Frasco, founded in 1964, publishes 32 named lines on its insurance claims page: surveillance, activities check, ROC remote observational camera surveillance, interviews, document authentication, credibility analysis, alive and well check, clinic inspection, scene investigation, AOE/COE investigations, subrogation investigation, background work, SCORE investigations, social media monitoring and geo, medical and pharmacy canvass, vehicle sighting, work check, asset search, locate, record retrieval, process service, vendor management and SIU program management. Delta Group sells eRemote, Traditional Surveillance, Desktop Solutions, Incident Investigations, SIU Investigations, Fraud and Compliance, and Vendor Management.

Two things follow from reading those menus side by side. The first is that a large and growing share of the work is not physical. Desktop Solutions, Remote Surveillance, Intelligence, record retrieval, social media analysis and background checks are desk work sold by the assignment, and the industry's own product naming says so. The second is that incumbents are adding AI at the task level, inside an engagement that is still sold by the engagement. Frasco names Frasco IQ in its navigation, describes Document Authentication as using AI-driven analysis and Credibility Analysis as AI-powered. That is genuinely useful and it makes one line item on a 32-item menu faster. It does not change how many flagged claims get an assignment at all.

Volume gives the menu its scale. The New York State Insurance Fund's claims investigative services RFP reported 8,884 surveillance assignments and 21,474 non-surveillance assignments in 2017, 30,358 in total, across ten district offices. One state fund. Seven in ten of those assignments, 70.7% of them, were not surveillance. The same document describes distribution: each district office assigns investigations on a rotational basis to the companies on its regional panel, taking into account investigation company staffing, geographic coverage and prior work on the claim. That is an allocation mechanism, not a capacity-creation mechanism. Rotation decides which firm gets the assignment after someone has already decided this claim gets one.

Outsourced investigation is priced by the investigator-hour, which means the cost of investigating one more claim never falls. A county procurement document states that structure more plainly than any vendor page does. Miami-Dade County's scope of services for investigative and surveillance services requires the vendor to only utilize one (1) investigator on an investigative assignment or secure approval from the County's Project Manager in writing, prior to utilizing two (2) or more investigators. Parallel effort on a single case is a budget event that requires written approval. That is not bureaucratic excess. It is an accurate reflection of what parallelism costs when the unit of production is a person.

The rest of the invoice structure is consistent with it. The same scope requires itemized surveillance hours, itemized travel time and mileage, with travel charges commencing at the inception of the investigator's travel and mileage paid under Florida Statute 112.061. If two investigators render services simultaneously and their time overlaps, the vendor bills for only one. Every line reads back to the same unit: the investigator-hour. The buyer pays for the commute.

The NYSIF RFP shows the other half, which is how the price moves across a contract's life. Prices are all inclusive, defined to cover direct labor costs, overhead, fee or profit, clerical support, equipment, materials, supplies, administrative support, all documents, reports, forms and reproduction. Rates hold flat for years one through three, then may increase by not to exceed 4% or CPI, whichever is less. Read that as a finance line rather than a procurement detail: a panel rate is a labor cost indexed to labor inflation by contract. It has no structural path down.

Priced as a curve, the panel has three properties a finance reviewer should carry into the renewal. The marginal cost of investigated claim number 5,001 is approximately the marginal cost of number 1. The rate escalates on an index tied to labor, and NYSIF's 4%-or-CPI cap is the disciplined version of that mechanism, not the aggressive one. And capacity is inelastic inside a contract year, which means a volume spike gets absorbed by triage rather than by supply. The buyer polices the resulting quality risk financially: NYSIF charges $50 per incomplete report, reduces the fee by 15% for each week a report is late, and withholds full invoice value where false information is supplied. Those penalties exist because throughput risk sits with the vendor and the buyer has no other lever.

Which is what the outsourced field-investigation industry actually is on a carrier's balance sheet. It is not a competing investigation layer. It is the carrier's own deferred investment in investigation capacity, rented back by the hour with an escalator attached. That was a sound trade for as long as investigation could only be performed by a licensed person driving somewhere. It is a less obvious trade for the desk-work share of the menu.

The Bureau of Labor Statistics counts 39,500 private detectives and investigators employed in the United States as of 2025, with median pay of $51,220 a year or $24.62 an hour, and projects the occupation will grow 5% through 2035, adding 2,200 jobs. Typical entry-level education is a high school diploma. That is the national pool. Every carrier SIU, every TPA, every law firm, every government fraud bureau and every PI firm hires from the same 39,500 people. A roll-up reorganizes that pool. It does not add to it.

Licensing is why the pool grows slowly. BLS states it plainly: most states require private detectives and investigators to be licensed. The NYSIF contract makes the practical cost visible. Each investigator needs a New York State private investigator license, NYSDOS fingerprinting, and a criminal and financial background check, all at the vendor's expense, and any out-of-state investigator must hold a license in their own jurisdiction. Miami-Dade requires two tiers: a Florida Class A private investigation agency license for the firm and a Class C private investigator license for each individual, maintained as a continuing condition of award.

So the time to double field capacity is not a hiring cycle. It is recruit, license, fingerprint, background-check, train, add to the panel, and then repeat the licensing step per state. The market below the national firms reflects that friction. Maricopa County's private investigator services contract carries more than twenty separate vendors on one county's qualified list, most of them small local shops. Fragmentation at that grain is exactly why a roll-up thesis is attractive to a sponsor, and consolidating it genuinely reduces a carrier's vendor-management overhead.

None of this is an argument for cutting investigators, and the honest version of the staffing question does not go there. It is an argument about arithmetic. When referral volume grows 73% over five years and the national labor pool is projected to grow 5% over ten, buying capacity out of that pool means bidding against every other carrier making the same correct decision at the same time. Queues of this shape do not clear on headcount; the operational version of that math is in the claims investigator's guide to clearing a high-volume backlog without increasing headcount.

Physical surveillance carries evidentiary obligations that are physical by definition. NYSIF requires original recordings preserved with an unbroken chain of custody for the balance of the calendar year plus six additional years, time-and-date-stamped video secured at the beginning, at the end and at one-hour intervals throughout, a second device generating a non-manipulable date-time display captured at those same intervals, a new tape or disc for each day of surveillance, night vision scope where necessary, and a surveillance affidavit whenever a claimant is observed in activities inconsistent with the claimed disability. The contract states the reason in its own words: the laws of evidence require that original recordings be preserved under circumstances establishing both an unbroken chain of custody and that the original recording has not been tampered with.

Testimony is a person. NYSIF requires investigation companies to be ready and able to provide expert testimony of legally and ethically obtained information, and if the original investigator is not available, a principal of the company must be available to testify. The obligation to produce original recordings and authenticating witnesses survives the termination of the contract. Miami-Dade requires expert testimony at depositions, pre-trial meetings, and civil and administrative hearings. No software satisfies any part of that, and no vendor should suggest otherwise.

Then there is everything that requires being somewhere. Activity checks, neighborhood canvasses, medical canvassing, scene investigation, in-person recorded statements, alive-and-well checks, clinic inspections, work checks, vehicle sightings and process service. Record retrieval where the record exists only as paper at a courthouse or a clinic, which NYSIF contemplates directly by reimbursing hospital photocopying fees against a receipt.

And there is legal judgment exercised in the field, which is harder than it looks from a desk. The investigation company is responsible for knowing and complying with all federal and state laws regarding video and audio recording that apply to a surveillance assignment, specifically those relating to one-party versus two-party consent to audio recording, trespass for the purposes of surveillance, and any other restriction on the means that may be used. That judgment gets made in a parked car, in a jurisdiction the investigator has to know without looking it up.

This work also pays for itself, which is the part most often left out of technology arguments. Ohio's Bureau of Workers' Compensation Special Investigations Department reported $91,770,469 in identified savings in fiscal year 2025 at a return of $6.60 for every dollar spent, with $65,738 in average savings per closed case, and $2.4 billion in cumulative savings across 76,745 closed cases since 1994. Field investigation is not a low-return activity. The constraint on it is volume, not value.

An assignment and an investigation layer solve different constraints, and they are not substitutes. An assignment is discrete field or desk work on a claim someone has already decided to investigate. A layer is the thing that decides: it takes every flagged claim, runs the desk-work phases across all of them, and returns a file the assignment can then be written against.

Which matters because field investigation is the highest-value work in the SIU, and that is precisely the argument for not spending it on triage. A licensed investigator who can build a chain of custody and then authenticate it under oath two years later is the scarcest resource in the entire apparatus. Using that resource to establish whether a flagged claim deserves a closer look is the least productive available use of it. Aimed at assembled evidence instead of a partially informed guess made at the point of referral, the same assignment is worth more.

The reason coverage matters more than per-case throughput is in the same New York report. Insurers filing electronic annual SIU reports claimed $3.5 billion in savings from SIU investigations in 2023, against $40.3 million in recoveries. Savings run roughly 87 times recoveries. The value of an SIU investigation sits overwhelmingly in the claim that never gets paid, not in money clawed back afterwards, and a claim nobody investigated produces neither.

Cycle time has to be read against the standard the file must meet. Ohio BWC's Special Investigations Department closed 1,395 cases in fiscal year 2025 in an average of 206 investigative days, with 116 agents, supervisors and support personnel statewide and a 58.6% founded rate on closed cases. That is a designated criminal justice agency building files to a prosecutorial standard, not a carrier SIU desk file, so the two numbers are not comparable. Manual carrier desk investigation runs 14+ days per case. An autonomous investigation layer compresses the desk-work portion to hours, not weeks, which changes how many claims can be looked at. It changes nothing about what a prosecution-grade file requires.

Hesper AI is built at that layer, and it is the layer no other vendor occupies. FRISS, Shift Technology and Verisk are detection. Guidewire and Duck Creek are claims management. There is no incumbent at the investigation layer except manual SIU teams and the panels they overflow into. Hesper takes a flagged claim and runs 15+ investigation phases in parallel, including document forensics, OSINT, statement cross-reference, timeline reconstruction and financial pattern analysis, and returns an audit-ready file with every source, decision and timestamp logged for the SIU lead to review. Fraud detection is built in, so it runs standalone or downstream of an existing detection vendor, and it is complementary to FRISS, Shift Technology and Verisk rather than a replacement for them. The measurable change is coverage: from roughly 25% of flagged claims to 100%. Make every flagged claim investigable, and the panel gets aimed rather than rationed. From fraud detection to fraud resolution.

The investigator's role shifts from execution to decision-making, which is also the honest answer to the staffing question. Nobody leaves. The field assignments that get written are the ones a complete file justified. For how that interacts with the structure a carrier files with its state, see SIU team structure: the org chart you file with the state is not a capacity plan.

"My panel is surge capacity I do not have to carry on payroll"

Correct, and that benefit is worth keeping. The question is not whether to have a panel. It is what share of panel spend is going to desk work that never needed a licensed investigator to drive anywhere. Miami-Dade's structure bills travel time from the inception of the investigator's travel. Some assignments are worth paying for a commute. Triage is not one of them.

"AI cannot sit outside a house for eight hours, and it cannot testify"

Agreed on both counts, without qualification. NYSIF requires the original investigator, or a principal of the firm, to remain available to testify after the contract has ended, and to produce the original recordings and authenticating witnesses. No software satisfies that. The argument is not that field work disappears. It is that field work should be aimed by a complete file rather than by a budget-constrained guess made at referral.

"One national vendor means one panel, one QA standard, one invoice"

It does, and that is a genuine operational improvement over managing twenty regional firms, which is close to what a single Arizona county's qualified-vendor list looks like. Command's stated rationale of broad coverage plus a dedicated quality-assurance unit plus integrated case management is the right thing to buy if vendor-management overhead is the problem you have. It does not change what an additional 1,000 investigated claims cost.

"My panel already does desktop work, and they have AI too"

True, and the honest answer is not to argue around it. Frasco names AI-driven document authentication and AI-powered credibility analysis. Delta sells eRemote and Desktop Solutions. The distinction is the unit of sale. Task-level AI inside a per-assignment engagement makes one line item faster, which is real value for the carrier buying that line item. The number of flagged claims that receive any assignment at all is still budget divided by rate.

"Compliance will not accept an AI-produced file"

The bar is the documented-decision chain, not the identity of whoever assembled it. California's 10 CCR 2698.36 and the antifraud-plan requirements under NAIC Model Act 680, adopted in 48 states, require a reconstructable record. Human-produced vendor files are not automatically audit-clean either: NYSIF polices them with a $50 penalty per incomplete report and a 15% fee reduction for each week of lateness. The right comparison is audit trail to audit trail, not machine to person.

"If investigation is that valuable, fund the SIU properly"

Make that argument, and make it with the Ohio numbers. A $6.60 return per dollar spent in fiscal 2025 and $2.4 billion since 1994 is a stronger funding case than most SIUs put in front of their own finance committee. The constraint is that the money buys people out of a pool of 39,500 that BLS projects will add 2,200 over a decade, and every carrier making the same correct decision bids the same pool. Funding and capacity are not the same variable.

Which returns to where this started. Incline, Harwood and Command read the demand signal correctly, and ten years of transactions have made outsourced investigation capacity broader, better governed and easier to buy. The line still has the same slope. The question to answer before the next panel renewal is not whether field investigation is worth paying for, because the return data says it plainly is. It is which of those assignments needed to be assignments.