Recycled Photos: The Oldest Fraud in Field Inspections

The pattern, and why the incentive builds it

Every field-service operation that pays per completed work order contains the same quiet arithmetic. The worker is paid for a unit of documentation, not for a unit of truth. Nobody at the office has seen the property. The only evidence that the visit happened is a small set of photographs that the person being paid for the visit selected and uploaded themselves. The scale matters here: inspection volumes, per-order fees, and the documented fraud cases are collected in our property preservation industry statistics.

Three behaviors follow from that arithmetic, in ascending order of how hard they are to catch.

Resubmission. Last month's photographs, or last week's, attached to this month's order. The property has not changed much, the reviewer has never seen it, and one exterior shot of a boarded window looks like every other exterior shot of a boarded window. This is the highest-volume version because it takes seconds and requires no craft.

Framing out. The visit genuinely happened. The photographs are genuinely new. But the standing water, the missing lock, the ceiling stain, the overgrowth that would trigger a costly work order sits eighteen inches to the left of the frame. Nothing about this submission is false. It is simply incomplete in exactly the way that keeps the job simple and the invoice clean.

Desk fabrication. The report is written without a visit at all, assembled from whatever the internet already knows about the address. County property appraiser sites carry photographs, square footage, structure type, sometimes recent imagery. A report built from that data will pass a checklist reviewer who is only confirming that the fields are filled in.

Notice that these three are not equally visible to the systems most QC teams have. A duplicate file is at least theoretically detectable. A well-framed photograph of the good half of a room is not detectable at all by anything that only examines file metadata.

What the record actually shows

This is not a hypothetical risk pattern, and the best-documented example is not a rumor from a vendor conference. It is a federal prosecution with a written plea agreement, and the details are worth reading closely because they describe the mechanics rather than just the outcome.

American Mortgage Field Services, a Florida company, held work from servicers including Bank of America to perform periodic preservation inspections on homes in foreclosure, most of them owned or insured by Fannie Mae, Freddie Mac, or the FHA. According to the United States Attorney's Office for the Middle District of Florida (justice.gov), the volume of inspection requests grew past what the company could actually deliver, and the company began fabricating inspection reports instead of shrinking the book.

The Justice Department describes the method plainly. The company employed individuals, many of them unskilled teenagers, to use previous months' photographs to fabricate subsequent inspection reports on properties. The same release says employees were also instructed to fabricate reports using publicly available websites, such as property appraiser sites, to obtain data about properties that were not inspected, and that employees who produced large numbers of false reports were often rewarded with cash bonuses.

The Federal Housing Finance Agency Office of Inspector General (fhfaoig.gov), which investigated alongside HUD OIG and the Secret Service, adds the detail that makes this a photo-provenance story specifically: employees were directed to create false inspection reports by inserting photographs from previous reports and changing the dates to create the illusion that new inspections had been conducted. The OIG also put a rate on it. Between at least 2009 and March 2012, it found, the company falsified up to 70 percent of the property inspections.

Two people were sentenced. Dean Counce, the company's president and founder, received 8 years and one month, which is 97 months, for conspiracy to commit wire fraud, alongside a money judgment of $12,774,102 and forfeiture of real estate and jewelry. Tammy Roaderick, his co-conspirator, received thirty-three months and a money judgment of $2,396,498.25.

The sentences are not the useful part. The useful part is the OIG's own conclusion about why it ran for years: because the employees were able to submit false inspection reports undetected over a period of several years, the processes for evaluating whether a contractor could actually perform the services appeared not to be as effective as necessary. The OIG called the problem one that may be systemic throughout the industry, and formally recommended that FHFA assess whether Fannie Mae and Freddie Mac were adequately overseeing property preservation inspections at all. That recommendation was made in November 2012 and the agency's response, published with the report, was to share the findings with examiners.

Read that as an operator rather than as a regulator. A vendor was submitting fabricated reports at scale, the client paid per report, and the control environment on the client side did not surface it. The fraud ended because investigators arrived, not because quality control worked.

Why provenance stamps are necessary and not sufficient

The industry's answer to this, and it was a good one, was capture software. Field apps built for property preservation stamp each photograph with location and time at the moment of capture rather than trusting a file that arrives later. Pruvan, now part of Verisk, is the best-known of them and describes exactly this on its own site: certified photos, videos, and mobile forms so you never doubt whether results have been tampered with or compromised. Customers quote the same value back in its testimonials, describing a business that relies on accurate photos with time, date, and location.

This genuinely closes a door. If a photograph can only enter the system through a camera that records where it stood and when it fired, the pure resubmission play gets much harder. Any QC program in this industry that has not moved to stamped capture should move.

But look again at the three behaviors. A provenance stamp is an assertion about a single file's origin. It says: this image was taken here, then. It is silent on three questions that matter at least as much.

Provenance is a chain-of-custody control. What went undetected for years at AMFS was a content problem wearing a custody costume. Custody controls are the floor. They are not the ceiling, and treating a stamped photo set as verified work is how a program convinces itself it has solved a problem it has only narrowed.

What content-level review catches

The gap is straightforward to name once you separate the two layers. Provenance asks where and when. Content review asks what.

The reason content review has historically been the weak layer is arithmetic again, this time on the QC side. Nobody can look at every photograph on a book of tens of thousands of monthly work orders, so programs sample. But sampling assumes the base rate is low. When the falsification rate in one documented case reached up to 70 percent, a 2 percent sample reviewed by a tired human is not a control, it is a ritual. And even a diligent reviewer looking at one submission in isolation has no way to know that this kitchen photograph is the same kitchen photograph they approved five weeks ago, because they do not remember the kitchen.

Three things become findable the moment review happens against a property's own timeline instead of against a checklist:

• Duplicate and near-duplicate submissions. A set that is materially the same as the set filed for that same address on the previous cycle is a question worth asking, whether or not the file metadata is fresh.
• Content that contradicts the property's own history. A back door that was documented as damaged on the last two visits and is now shown intact, without any work order that would explain the repair, is either a repair nobody billed for or a photograph of something else.
• Missing required shots. Not missing fields, which any form validates, but missing views: the room, the angle, or the exterior elevation that this property's scope calls for and that this submission quietly does not contain.

None of those are exotic. They are what a very patient reviewer with a perfect memory of every prior visit to that specific address would notice. That reviewer has never existed at scale, which is the actual reason the layer was missing.

The same pattern, one industry over

If this sounds like a mortgage servicing story, it is worth saying that the incentive structure is not specific to mortgage servicing, and neither is the behavior.

Short-term rental operators run the same shape: a distributed workforce, paid per job, documenting spaces that the office never sees, submitting photographs as the proof that the work happened. And they see exactly the same thing. Cleaners re-upload one good photo of a made bed or a clean kitchen for weeks at a time, because the photo requirement is satisfied by any photo and nobody is comparing this Tuesday's to last Tuesday's. It is common enough in that industry that it has its own folklore, and we have written separately about catching cleaners who reuse old turnover photos and about how faked cleaning photos actually work.

The dollar figures differ by three orders of magnitude. The mechanism is identical: piece-rate pay, self-reported evidence, and no comparison against the space's own history. Anywhere those three conditions hold, recycled photos are not a risk to be assessed. They are a default to be designed against.

Where RapidEye fits

RapidEye is an AI inspection layer that reads the photos and video field teams already capture. It does not replace your capture app or your work order system, and it does not ask anyone in the field to do anything differently.

What it adds is the content layer described above. Every submission is compared against that property's own baseline over time, so a set that repeats an earlier visit, a condition that contradicts what the property looked like on the last visit, or a required view that is absent gets surfaced for a human instead of waiting for a human to happen upon it. Damage, missing items, and standards not met come back the same way. Coverage is every submission rather than a sample, which is the part that matters when the base rate is unknown and possibly high.

It works with the platforms operations teams already run on across property management, hospitality, facilities, and field service, and it returns findings into whatever system your reviewers already live in. If you are running vendor QC on a book of field inspections and your current control is a stamped photo plus a sampled human read, the honest description of your coverage is that you can prove the visit and not the work.

Quick FAQ