Fraud Detection
AI-powered fraud detection for claim evidence, including fraud scores, analysis details, and how to interpret results.
Fraud Detection
Umbrella’s AI-powered fraud detection automatically analyzes photo evidence submitted with warranty claims, giving you a risk score to help inform claim review decisions.
This feature is available on the Pro plan.
How It Works
- Customer uploads photo evidence with a claim.
- After submission, a background process analyzes each photo for manipulation.
- Results appear on the claim review page, typically within seconds.
No manual action is needed to trigger the analysis.
Fraud Score
Each photo receives a risk score from 0 to 10:
| Score | Classification | Meaning |
|---|---|---|
| 0-3 | Authentic (green) | No significant manipulation detected |
| 4-6 | Suspicious (amber) | Possible manipulation; warrants closer inspection |
| 7-10 | Fraudulent (red) | Strong indicators of digital alteration |
Additional states: Analyzing (blue) — analysis in progress. Error (gray) — analysis failed, can be retried.
Where You See Fraud Results
Evidence thumbnails — A colored dot in the corner of each photo provides at-a-glance risk assessment.
Full-size preview — Click a thumbnail to see the fraud analysis summary (score, status, confidence percentage) below the image.
Fraud Analysis Details modal — Click the shield icon on any thumbnail or the “Analysis Details” button in the preview to see:
- Overall assessment with color-coded banner
- Risk score and confidence percentage
- List of specific detected issues (e.g., image cropping, inconsistent EXIF metadata, clone stamp artifacts)
- File metadata and analysis timestamp
- A Re-Analyze button to re-run the analysis
Interpreting Scores
Low Risk (0-3)
No manipulation detected. Proceed with normal claim review — focus on claim details and policy coverage.
Medium Risk (4-6)
Some indicators found, but results are inconclusive. Common causes include resized photos, screenshots of photos, or images shared through messaging apps. Review the detected issues, compare photos for consistency, and consider requesting additional evidence. Do not deny based on a medium score alone.
High Risk (7-10)
Strong indicators of manipulation. Review all detected issues and consider:
- Requesting new photos with specific angles or a timestamp
- Asking for a video showing the damage
- Requesting photos with a reference object (newspaper with date, handwritten note)
- Denying the claim if evidence is clearly fabricated
Re-Running Analysis
To re-run analysis, open the Fraud Analysis Details modal and click Re-Analyze. This is useful when:
- The initial analysis failed due to a rate limit or server error
- You want to verify results with a fresh analysis
- The analysis method has been updated
If you see a rate limit error (HTTP 429), wait a few minutes before retrying.
Limitations
- Fraud analysis applies to photos only — videos are not analyzed.
- Analysis can fail due to rate limits, server errors, invalid images, or timeouts. Use the Re-Analyze button to retry.
Related
- Claim Evidence — Evidence management and upload requirements
- Claim Review — Where fraud analysis results are displayed
- Claim Resolutions — Using fraud information to guide resolution decisions