Receipt verification costs: a model for processing, review, and rejected uploads

The cost of receipt verification is not one OCR price multiplied by uploads. A production program pays for capture, extraction, product matching, duplicate controls, human exceptions, integration, support, and the consequences of wrong decisions. A useful model makes those parts adjustable instead of hiding them inside one optimistic unit rate.

This article provides a planning model, not Steve pricing. All numbers below are synthetic assumptions. Use the current pricing page, a scoped proposal, and your own labor costs before making a purchase decision.

Define the unit first

Teams often compare prices that use different units: uploaded image, processed page, API call, submission, approved receipt, or awarded reward. Write the denominator next to every rate.

For a receipt program, track at least:

• uploads attempted;
• captures rejected before processing;
• submissions processed;
• submissions sent to human review;
• approved and rejected outcomes;
• delivery attempts and failures;
• members rewarded by the campaign.

A quality rejection is not the same as a processed rejection. A successfully verified receipt is not necessarily rewarded: Open Loyalty may find that the member is ineligible, the date is outside the campaign, or a cap has been reached.

Build the monthly model

Use these variables:

Variable	Meaning
U	Upload attempts
Q	Share stopped by capture quality before processing
P	Processed submissions: U × (1 − Q)
R	Share of processed submissions reviewed by a person
M	Average review minutes
L	Fully loaded reviewer cost per hour
V	Verification platform charge per billable unit
F	Fixed monthly platform or support fees
I	Monthly allocation of implementation and integration cost

The human-review cost is:

P × R × (M / 60) × L

An illustrative scenario uses 50,000 uploads, a 4% quality-stop rate, a 6% review rate, three minutes per review, and €30 per reviewer hour.

• Processed submissions: 50,000 × 0.96 = 48,000
• Reviewed submissions: 48,000 × 0.06 = 2,880
• Review hours: 2,880 × 3 / 60 = 144
• Review labor: 144 × €30 = €4,320

These figures are not performance expectations or prices. They show why review rate and handling time matter. At a 12% review rate, labor doubles even if the verification unit price stays unchanged.

Model three cases, not one forecast

Create low, expected, and high cases for volume, quality rejection, review rate, and handling time.

Assumption	Low	Expected	High
Uploads	30,000	50,000	80,000
Quality stop	2%	4%	8%
Human review	3%	6%	12%
Minutes per review	2	3	5

Label every figure as your assumption. Do not borrow a vendor’s best-case automation rate and combine it with your worst-case internal labor rate.

Include setup and change costs

The first campaign may require product-catalogue preparation, sample labelling, workflow configuration, integration, reviewer training, privacy review, and acceptance testing. Spread one-time costs over a sensible period or number of campaigns, but keep them visible.

Also estimate change work. Retailer abbreviations change, new pack sizes launch, campaign dates move, and product lists expand. Open Loyalty campaign changes and Steve evidence-workflow changes are separate. That separation can reduce unnecessary rework, but each still needs an owner.

If the workflow accepts invoices, include preprocessing for unsupported formats. Steve currently accepts images and single-page PDFs; multi-page PDFs need another step or a separately designed workflow.

Put error costs beside operating costs

False approvals can create excess rewards and fraud loss. False rejections can create support contacts and member churn. Review catches some uncertainty but costs time.

Do not assign dramatic values without evidence. Start with your historical data:

• average reward value;
• support cost per disputed rejection;
• confirmed duplicate or ineligible rate;
• appeal overturn rate;
• campaign margin or target behavior value.

Then run sensitivity ranges. If the average reward is €10 and 100 extra ineligible claims are approved, the direct reward exposure is €1,000. That is a scenario, not a prediction. The more useful exercise is identifying which controls reduce that exposure and what they cost.

Account for delivery and campaign operations

Verification is not finished when fields are extracted. Approved evidence must reach Open Loyalty or another system reliably. Budget for monitoring failed deliveries, fixing member identity, reconciling ambiguous outcomes, and testing retries.

Open Loyalty then evaluates campaign eligibility and earning. Keep the resulting reward liability separate from verification cost. A campaign that awards more points after a rule change did not make receipt processing more expensive; it changed the campaign economics.

Compare proposals with one worksheet

Ask every option—internal build, vendor, or combined approach—to populate the same rows:

1. billable unit and exclusions;
2. expected quality-stop and review assumptions;
3. review tooling and reviewer labor;
4. integration and maintenance;
5. product-catalogue and rule changes;
6. retry and support operations;
7. data storage and retention;
8. estimated error exposure;
9. campaign/reward platform cost;
10. fulfillment cost, if rewards need an external provider.

Do not compare a verification API quote with an end-to-end managed-service quote until the missing operations are priced on both sides.

Turn the model into a pilot budget

A pilot should use representative evidence and a fixed scope: selected retailers, product families, expected volume, review policy, and one Open Loyalty campaign. Record actual review causes and handling time. Replace assumptions with observed values before planning a national launch.

The goal is not the lowest theoretical price per receipt. It is a predictable cost per defensible decision, with a member experience and operating model the team can sustain.

Use the pricing page for current commercial terms and the receipt-verification page for workflow scope. To estimate a pilot, book a demo with expected volume, sample receipts, reward value, and your current review cost.