Imagine submitting a genuine receipt after a business trip. You’ve paid for your meal and the receipt you’ve uploaded online for reimbursement is not only rejected but it’s also marked as suspicious putting your account on a lockdown.
Not only is your reimbursement delayed, but now you have to make your case to support, HR, or the managers. For Fintech platforms, this is an important issue which requires extensive testing and training to ensure that the platform doesn’t penalize honest customers.
But why does this occur and the false-positive definition:
A false positive occurs when a legitimate submission is incorrectly classified as fraudulent or suspicious. This can occur due to poor image quality, detection errors and overly aggressive workflow rules which can contribute to the false flag. Regular testing with varied documents helps teams to identify these false positives or negatives and address them.
So in our case…
Submissions which trigger a duplicate signal require attention as it may be due to low resolution or because the image was cropped. These results need to be evaluated and depending on the platform.
The signal explains what needs attention. Your workflow determines what happens next.
Consider a customer who submits a screenshot of a receipt received via email. Veryfi can identify digitally originated submissions, but as for what kind of flagging occurs depends on your platform’s policies.
Here’s an example of a platform policy in action:
If your platform requires the original document, then you can adjust the API to mark a screenshot as ineligible under your policy. Digitally submitted images aren’t always a cause for concern and can be easily resolved without user suspension.
Duplicates require similar attention. A customer may have uploaded the same receipt twice because they were unsure if the first submission went through. Veryfi’s documentation signal identifies what needs attention from the document alongside fraudulent activity while your workflow determines the next action.
It also notes that poor image quality can cause false positives or false negatives in duplicate and similarity checks. Duplicate detection documentation. You still need to prevent a second payout. However, holding a duplicate submission while checking the original claim is different from declaring the customer dishonest.
What about handwriting? This one accounting teams tend to deal with the most. For instance, a restaurant receipt with a handwritten tip may warrant checking the final total and supporting records. The handwriting on a receipt shouldn’t be the sole cause for rejection or flagging.
These examples are hypothetical, but they highlight a practical distinction: document characteristics, claim eligibility and fraudulent intent are separate issues needing to be addressed inside the workflow.
Veryfi Fraud Detection helps teams assess potentially suspicious documents through document and device-related signals with Veryfi Lens. The assessment response consists of meta.fraud.color, meta.fraud.score, meta.fraud.attribution, and meta.fraud.types. The meta.fraud.color values are the color strings themselves; green = automates, yellow = review, red = hold.
For a product team, a useful starting point is a workflow like this:
| Submission context | Suggested response |
| Low document risk, required information present and business rules satisfied | Continue through the normal approval process. |
| Ambiguous signals, missing information or conflicting evidence | Request better evidence or route for review. |
| High risk or a confirmed violation of submission rules | Hold the claim or block progression according to your policy. |
This table showcases suggested workflow to minimize false signals.Your application still remains responsible for deciding which claims qualify and when money moves. For best results, the platform should respond to Veryfi’s findings based on submission policies. Final verdict should remain within your application. With this configuration Veryfi’s API would accurately signal which documents users are allowed to submit, including screenshots where your policy permits them.
To make that workflow useful in production, focus on these five things:
- Find out why legitimate submissions are getting stopped.
Separate incorrectly detected fraud from real document characteristics that your rules interpret too aggressively. Investigate image quality, submission format, duplicate history and policy exceptions. - Test with the documents your customers actually submit.
Include genuine digital receipts, photographed paper receipts, different merchants, currencies and capture conditions. Add labeled fraudulent samples so that improving the legitimate-customer experience does not conceal a drop in fraud detection. - Tune supported signals deliberately.
Veryfi’s Fraud Config lets administrators adjust supported signal settings. Higher thresholds require stronger evidence; lower thresholds make signals more aggressive. Test changes against both legitimate and fraudulent samples. Device signals require Veryfi Lens and are not self-editable. Fraud configuration guide - Keep document risk and business policy connected.
A genuine receipt can still fall outside an expense policy or promotion’s eligibility rules. Use the extracted data and fraud findings together. Veryfi supports applying business logic through Business Rules or your own downstream validation. Configuration and business-rule guidance - Give customers a specific next step.
“Upload the original receipt” or “This receipt is linked to an existing claim” is more useful than an unexplained rejection. Offer a review path where appropriate, without exposing sensitive detection details.
Tracking how often legitimate submissions are wrongfully rejected and fraudulent submissions go through will allow you to fine tune Veryfi’s API to your specific application and its unique policies. With these configurations your team makes better decisions depending on the configuration, business context and how exceptions are handled.
For teams building expense, AP or rewards workflows, the aim is practical: identify suspicious documents, prevent inappropriate payouts and resolve legitimate submissions without unnecessary friction.
Explore Veryfi’s Fraud Detection capabilities and build a workflow that gives each signal an appropriate response.
