Use caseHomeowners insurance
An invoice $500 above its own lines. A photo already paid on a claim from 2025. In a queue of forty claims a day, such things get caught by luck.
The AI flags what does not line up, and shows whereThe referral travels with every document and noteThe adjuster and the investigator decide, never the AI
The invoice lines add up to $6,350, and the total says $6,850. The plumber's visit is dated two days before the leak was discovered.
source · the invoice and the loss noticeSomething does not line up · the AI shows where, an adjuster reviews
$500. That is how far the total on Gary Chandler's plumbing invoice sits above its own three lines, and it looks like any other invoice. One of his four photos was already filed on a claim paid in March 2025, which only surfaces if the same adjuster happens to remember it. When a doubt does come up, the referral is a pile of screenshots for the investigator to rebuild.
On every claim, the documents are checked the day they arrive. The AI flags what does not line up and points to the page, and a nightly pass over the repository finds files that already sit on another claim. The adjuster clears or refers each signal with a reason, and the investigator decides on the evidence.
Adjusters and investigators already use every one of these screens. What protects the honest customer is the order: the signal is raised on arrival with its evidence, the adjuster looks for a plain explanation first, and only what stays unexplained goes to the investigators. The acts below stay on CL-26-27731, one Phoenix condo and three signals.
The claim looks routine. The checks have run before Kyle Foster, the adjuster, opens it.
Gary Chandler reports water damage in his condo in Phoenix and uploads an invoice from Desert Valley Plumbing for $6,850. The AI extracts the lines and the total, and checks that they agree. The three lines add up to $6,350.
The loss notice gives Sep 16 as the day the leak was discovered. The invoice says the plumber came on Sep 14. The AI reads the dates printed in each document and places them in order on the claim, so Kyle sees the sequence at a glance.
Every night, the operations engine runs a check across the claims repository and lists files that appear on more than one claim. This morning it lists one of the four photos: Bathroom floor.jpg is the same image file as one filed in March 2025 on claim CL-25-08214, under another policy.
The other claim belongs to another team. Kyle sees that the match exists and its reference. Opening that claim stays with the special investigations unit.
Three signals sit on the claim. Kyle looks for a plain explanation before anything else, and one call settles a third of it.
Kyle calls Mr. Chandler with a neutral question about the dates. The leak started on Sep 14, the plumber came the same day, and he only reported it once the neighbor downstairs wrote about the stain. That matches the neighbor's letter. Kyle clears the date signal and writes the reason.
The invoice gap and the reused photo have no explanation yet. At the end of his step, Kyle launches an investigation from the processes he is allowed to start. The child case receives the claim's data and documents, the highlights and his call note. He holds the invoice payment, as your procedure asks, and releases the other lines of the claim.
Diane Rollins, an investigator, picks up the referral the next morning and checks each signal at its source. Four steps, and a monthly tune-up.
Diane opens REF-26-0193 and asks the assistant for a summary. Her role lets her open claim CL-25-08214 too: it was paid in March 2025, at another address.
Diane asks Desert Valley Plumbing for their own copy of invoice DVP-2026-0418. It comes back the same day and is filed on the referral. She opens both copies side by side in the viewer: everything matches except the total.
Diane records her finding for each signal. The claim is settled on the supplier's figures for the work that is proven, and the decision on the policy goes to the fraud committee as an approval task. Kyle gets the claim back with the finding.
Once a month, Sarah Bennett reviews the document checks. Her report counts, per check, the signals raised, cleared by adjusters and confirmed by investigators. The date check is often cleared, so she rewrites its instruction in the console: a gap under three days is now shown as information. The change needs no release.
A signal is a question about a document. Each claim ends up in one of three places, and the investigators only see one of them.
Amounts, dates and files agree. The checks leave no trace on the adjuster's path.
What follows: the claim follows its usual process.A late report, a typo on a date. The adjuster writes the reason on the claim.
What follows: the claim goes on, and the reason feeds the monthly review.The signal and its evidence go to the special investigations unit as a case of its own.
What follows: the investigator checks at the source; the committee decides on the policy.It accuses nobody, refuses nobody and blocks no payment. Across the nine steps on this page, every decision is signed by an adjuster, an investigator or the committee, with their own rights and their own trace.
The AI extracts, compares and summarizes, and the operations engine lists duplicate files. The control is in how the procedure is written, so honest customers are never treated as suspects.
Counted on Gary Chandler's claim and its referral. A batch of auto claims or a stricter procedure would count differently.
Where these numbers come from. They are counted on the scenario described on this page, with example data. They are not a customer result.
An adjuster, an investigator and an architect all have to be comfortable with a fraud check before it touches real claims.
Forty claims a day, and somewhere in them a doctored invoice you are expected to spot.
Referrals land as screenshots, and the first hour goes into rebuilding them.
Every fraud tool you have seen wanted another platform and another copy of your claims.
With one check. Take the invoice your adjusters add up by hand, and move the check to the door. Run it first on claims your investigators already closed, so the results can be judged against known answers.
Your claims system, your repository and your investigation procedure stay in place. The checks run inside them.
Next step
Bring a batch of anonymized claims your investigators have already closed. We will show which signals the checks raise and how your team would review them.