Fraud Detection Overview

In rewards promotions and expense management, proof-of-purchase fraud is a real cost: the same receipt submitted twice, a total edited before upload, or a receipt that was never printed at all. Taggun's fraud detection suite runs the checks enabled for the account or configured for the validation campaign and returns machine-readable results alongside the extracted data, so your workflow can approve, reject, or route to review automatically.

Availability

📘

Availability

The fraud detection suite is included on the Advanced plan and above. Enable the Fraud Suite controls in your account Feature Settings when they are available on your plan. See Plans & Feature Access.

The Checks

How It Works

1. Submit — send a file to a verbose extraction endpoint or a validation endpoint.
2. Analyse — the fraud detection engine runs the enabled checks on the file.
3. Score — tamper, digital, and handwriting checks return flags plus scores; duplicate and similarity checks return matched receipts with similarity scores.
4. Respond — the results come back in the same synchronous response as the extracted data.

Sample Response

This is a verbose extraction response, so the checks sit under entities. On Receipt Validation endpoints, tamperDetection, digitalDetection, and similarReceipts come back at the root level instead, and handwriting detection is not run.

{
  "entities": {
    "similarReceipts": [
      {
        "referenceId": "REC-20240105-ABC123",
        "userId": "USER-9876543",
        "trackingId": "T-20241005-9276439",
        "similarityScore": 0.9779201745986938
      }
    ],
    "tamperDetection": {
      "data": {
        "isTampered": false,
        "tamperedScore": 0
      }
    },
    "handwritingDetection": {
      "data": {
        "isHandwritten": false,
        "handwrittenScore": 0.05660377358490566
      }
    },
    "digitalDetection": {
      "data": {
        "isDigital": true,
        "digitalScore": 0.92
      }
    }
  }
}

In this example, one prior receipt match was returned, isTampered and isHandwritten are false, and isDigital is true with a score of 0.92. Each check's page documents its public response fields.

Accuracy & Thresholds

Fraud performance depends on your campaign rules, document mix, and confirmed-fraud test set. Measure it on historical receipts with known outcomes, record false positives and false negatives separately, and tune your review workflow to your risk tolerance. See Test With Your Own Receipts.

Tamper, digital, and handwriting checks return flags plus scores; duplicate and similarity checks return matched receipts with similarity scores. The flags (isTampered, isDigital, isHandwritten) are Taggun's tuned recommendation — build accept/reject logic on the flag. Use the associated scores only to apply thresholds that suit your risk tolerance, and route borderline cases to human review.

Best Practices

1. Use the checks together. Enable the relevant checks together to broaden signal coverage rather than relying on one result.
2. Add your own signals. Complement receipt analysis with user behaviour analysis, IP monitoring, user ID, and device fingerprinting where fraud pressure is high.
3. Keep a human review lane. Route flagged receipts to reviewers with clear escalation rules and an appeal path for rejected users.
4. Don't tell submitters fraud was detected. A specific rejection reason teaches fraudsters what to work around — use neutral wording ("validation in progress") and never expose fraud flags in end-user messaging. Do tell users when an image is too blurry to read; that's a quality problem a genuine user can fix.
5. Educate submitters. Clear capture guidance reduces unintentional flags — see End-User Capture Guidance.

Additional notes

• PDFs. Not all image-level tamper techniques run on PDF content. Do not assume standalone-image checks apply identically to PDFs.
• You can report bypasses. If your team confirms a fraudulent receipt that passed the checks, follow Contacting Support. Include the email attached to the account, the affected receipt's trackingId when returned, the sample when safe to share, and a clear description of the issue.

Need something not covered here? Contact [email protected] — see Contacting Support.