NEWAutomated document-fraud detection, built natively in Salesforce.
For insurance operations, underwriting & risk leaders

Catch forged documents
in seconds, not hours

FraudCheck 360 returns a clear verdict, a trust score, and full evidence in seconds, without a single document leaving Salesforce.

Automated document-fraud detection for insurance, built natively in Salesforce under the Einstein Trust Layer. It turns hours of manual review into seconds, and never auto-rejects a borderline case.

A clear answer in seconds, with the evidence attached.

Document in
Repair invoice · claim #CL-88142
Header & branding
Line items & totals
Dates & references
Logos & fonts
FraudCheck
360
Verdict in Salesforce
FraudulentTrust score 38
Totals don't reconcileFigures
Pasted logo detectedVision
Font change mid-tableVision

Built on Salesforce + MuleSoft, inside your own org

SalesforceMuleSoft IDPEinstein Trust LayerAI VisionData Cloud
Why forged documents cost insurers

Claims, underwriting, and onboarding all run on documents

Loss estimates, invoices, receipts, proof-of-loss forms, financial statements, and IDs all come from the customer. Altered figures, pasted logos, and re-typed text are hard to catch by eye and getting better. Accept a forgery and you approve a bad claim or policy; reject a genuine one and you lose a good customer. Both mistakes are expensive.

Manual review is slow

Minutes per document, and reviewers tire and miss details under volume.

Forgeries are improving

Altered amounts and pasted logos increasingly slip past the human eye.

A missed forgery is costly

A bad claim, policy, or account approved on false proof is expensive.

Wrongful rejection is costly too

Reject a genuine document and you lose a good customer and damage trust.

The problem, by the numbers

The scale of document fraud

$308.6B

Estimated annual cost of insurance fraud in the U.S.

Coalition Against Insurance Fraud

48%

Of carriers saw a rise in suspicious claims linked to AI-generated documents

Gallagher Bassett 2026 Carrier Perspective, via Claims Journal

What makes FraudCheck 360 different

Built for accuracy, speed, and trust

FraudCheck 360 turns hours of manual review into seconds, and never auto-rejects a borderline case.

01

Hours of review, done in seconds

Upload a document and the system reads it, runs a tailored set of fraud rules plus an AI visual check, and returns a clear answer in seconds. Your team handles far more volume with the same headcount.

02

Three verdicts, no false rejections

Clear cases are decided automatically; only genuinely uncertain ones become Review and go to a person, with the reasons attached. Good customers are never auto-rejected on a borderline case.

03

Nothing leaves Salesforce

Every check, including the AI, runs inside your Salesforce org under the Einstein Trust Layer. No external API calls, no data sharing, and your existing record and field security is honored.

What your team gets

Everything in one native workspace

Speed at scale

Documents checked in seconds, not minutes, so your team clears far more volume.

Fraud rules + AI vision

A tailored rule set plus AI vision catches figures that don't add up, altered text, pasted logos, and re-typed fonts.

It shows its work

Every verdict carries a trust score, the exact reasons, and a map to the problem on the page.

Audit-ready record

Each check is saved to the customer record with full evidence and a shareable PDF report.

Secure by design

All AI runs in the Einstein Trust Layer. No external services, no data ever leaves Salesforce.

You stay in control

Rules, point weights, and thresholds are changed by your team in Setup, with no developer needed.

Two ways to build your fraud rules

Fits how you already fight fraud

Stand up coverage fast with AI-generated rules, encode the rules your team already trusts, or combine both. The rule set is dynamic, and you adjust either path in Salesforce Setup with no developer.

Option 1 · AI-generated

Dynamically through AI

AI derives the fraud rules from what a genuine document should look like, including branding, figures, expected wording, and visual integrity. You get working coverage quickly, without authoring rules by hand.

Option 2 · Your existing rules

Using rules you already have

Already have fraud criteria, watchlists, or underwriting checks? Encode them directly as configurable rules with your own weights and thresholds, so the system enforces the policies your team already trusts.

How it reaches a verdict

Four steps, one clear answer

Step 1

Read the document

MuleSoft Document AI extracts the header, key fields, line items, and totals from the document.

Step 2

Run the fraud rules

Tests a tailored rule set: branding, figures, wording, dates, and internal consistency.

Step 3

AI visual check

Inspects the image for pasted logos, font changes between cells, and uneven spacing.

Step 4

Score & explain

Sets the trust score, applies the verdict, and writes a plain-language reason.

Authentic

Trust score 85–100

Looks genuine; all checks passed. Accept and continue your normal process.

Review

Trust score 55–84

Borderline, with no clear tampering. A person checks it, with the reasons attached.

Fraudulent

Trust score below 55

Clear signs of tampering. Reject or escalate; the evidence is shown.

Why three outcomes matter: a simple accept/reject system either lets fraud through or turns away good customers. The middle Review outcome protects you from both, so only the genuinely uncertain documents ever reach a human.
See it in Salesforce

The real product, right inside your CRM

No new tab, no separate portal. Reviewers get the verdict, evidence, and every rule they can tune — all on the record they already work in.

A clear verdict, with the evidence

Every check returns a verdict, a trust score, layered IDP + AI Vision confidence, and a plain-English AI summary — so a reviewer knows exactly why in seconds.

Rules and thresholds you control

Tune the Authentic, Review, and confidence thresholds, and weight each fraud rule to match your risk appetite — fully transparent, no black box.

ROI by vertical

Where it pays for itself, line by line

The four ROI levers are the same in every insurance line — only the documents, fraud patterns, and numbers change. Pick a vertical to see the case.

Hard-dollar loss savings

Fraudulent or inflated payouts avoided, and leakage recovered, by catching altered documents before money goes out.

Labor & cycle time

Manual review — minutes per document — is replaced by a verdict in seconds, so the same team clears far more volume, faster.

Detection lift

Higher catch rates and fewer misses than the human eye, with a Review tier so good customers are never auto-rejected on a borderline case.

Loss / combined ratio

Because fraud and leakage sit on the loss side, reductions flow directly to loss ratio, combined ratio, or medical loss ratio.

Property & Casualty

The broadest exposure — every claim, policy, and onboarding runs on customer-supplied documents.

For P&C claims, underwriting, SIU, and risk leaders
$308.6B

Estimated annual cost of insurance fraud in the U.S.

Coalition Against Insurance Fraud, 2022

99%

Of insurers report encountering manipulated or AI-altered documents

Verisk State of Insurance Fraud, Mar 2026

~$45B

About 10% of P&C losses and loss-adjustment expense is attributable to fraud

Insurance Information Institute

Documents most at risk

Repair estimates and contractor invoices, proof-of-loss statements, medical bills and records, police and incident reports, damage photographs, receipts for lost or stolen property, valuations and appraisals, and government IDs.

How the forgeries show up

Doctored invoices pairing real company names with fabricated line items, padded repair estimates, digitally and AI-altered damage photos, altered amounts on legitimate documents, pasted logos and cloned letterheads, and fictitious or prior-loss claims.

The ROI case — four levers

Hard-dollar loss savings

Fraud is about 10% of losses plus LAE. Catching an altered estimate or fabricated invoice before payout is direct loss avoidance; claims leakage typically runs 5–10% of total claims cost.

Labor & cycle time

Manual review runs 2–4 minutes per document and reviewers tire. An automated read and verdict in seconds lets the same team clear far more volume — IDP benchmarks cite 70–90% review-time savings.

Detection lift

99% of insurers already see AI-altered docs and 76% say they are getting more sophisticated. AI visual forensics catches pixel-level tampering the eye misses.

Loss / combined ratio

The 2024 P&C combined ratio was ~97.2% and tightening toward 99% by 2026. Fraud sits on the loss side, so even a partial reduction flows straight to the ratio.

Under the hood, the technology stack

A native Salesforce application

FraudCheck 360 is added on top of your existing org without changing its other apps or automation. All AI reasoning runs inside your org under the Einstein Trust Layer; MuleSoft handles only secure document extraction. No external LLM keys, no data leaves the platform.

1 · Data
Stores documents, checks, findings, and configuration.
Custom Objects · Custom Metadata Types · Platform Events
2 · Document Reading
Extracts structured fields, line items, and totals from the uploaded file.
MuleSoft Intelligent Document Processing (CloudHub 2.0) · DataWeave
3 · Intelligence
Runs the configurable rule engine, AI Vision forensics, and the AI Summary.
Apex · Einstein (Trust Layer)
4 · Orchestration
Ties the steps together, scores, decides, and persists the result.
Apex services · async Queueable
5 · Experience
The workspace screens, Command Center dashboard, and PDF report.
Lightning Web Components · Visualforce
AI that stays inside your trust boundary

One place for AI logic, one for integration

All LLM calls route through the Einstein Trust Layer via the Models API and Prompt Builder; MuleSoft IDP extraction is likewise Trust-Layer governed. Rules, weights, and thresholds live in Custom Metadata, tuned in Setup with no deploy.

Document extraction

MuleSoft IDP

A CloudHub 2.0 app reads the file via Intelligent Document Processing and returns normalized JSON. Salesforce reaches it through a Named Credential with OAuth, with no secrets in code.

Governed AI

Einstein Trust Layer

AI Vision (prompt template) and the AI Summary (Models API) run governed, with PII masking, zero retention, toxicity scanning, and audit. The org holds no LLM key and makes no direct external callout.

Resilient by design

Safe degradation

If a live AI call can't run, the layer falls back to a deterministic forensic check that still catches tampering. The check never fails outright, and your record and field security is always honored.

See FraudCheck 360 on your documents

Let's talk through your claims, underwriting, and onboarding workflows, and map how native, Salesforce-governed fraud detection fits your operation.

contactus@psadvisory.com · www.psadvisory.com