Configure OCR confidence threshold
Set how sure the receipt-reading step must be before a receipt continues through your challenge.
Use the confidence threshold to control how sure the receipt-reading step must be before a receipt continues. It helps you keep blurry or hard-to-read receipts from slipping through, without rejecting normal shoppers.
When a participant uploads a receipt, OmniLab reads the text from the image automatically (the step we call OCR) to pull out the store name, amount, and date. The confidence score is its estimate of how reliably it read the image. Set the threshold too low and poor-quality scans pass too easily. Set it too high and normal store receipts get rejected too often.
What this setting affects
The confidence score does not replace the acceptance policy. It sets how much you trust the read data before those checks run.
A lower threshold usually means more borderline cases, more manual checking, and more shopper frustration when receipt quality varies.
When to set it stricter
Raise the threshold when:
- the reward value is high
- fraud is a real concern
- receipts often come from many similar-looking store formats
- your team is available to review borderline cases by hand
When to set it more permissive
Lower the threshold when:
- receipt quality varies a lot in the field
- the reward is low-risk
- you prefer fewer false rejections and can handle more manual review later
If you are unsure, start stricter and loosen it only if too many genuine receipts are rejected.
If the threshold causes problems
A few common reasons and how to fix them:
- The value is outside the accepted range — check that your entry is within the range OmniLab supports, then correct it.
- The threshold is set too low — the challenge may accept reads that are too uncertain for the campaign. Raise it gradually and re-test with a sample receipt.
Pair the threshold with photo guidance
The right threshold still depends on the photo the shopper sends. OmniLab already prompts shoppers to send a photo that is clear, sharp, centered on the receipt, and shows the store and the amount. If reading quality is a recurring problem, improve the upload guidance before you only tighten the threshold.