How to use this calculator
- Enter the concurrent worker count.
- Use measured end-to-end seconds per processing attempt.
- Add average pages, utilization, and operating hours.
- Enter the share requiring one retry and calculate.
#898 · AI & Technology Tool
AI Document Processing Throughput Calculator
Estimate the sustainable output of an AI document processing system from observed end-to-end service time. Enter the number of concurrent workers, average seconds required for one document, pages per document, effective utilization, daily operating window, and retry rate. The result reports successful documents per hour as the primary throughput measure, plus daily document and page capacity. Because retries consume worker time, the calculator reduces successful output rather than treating every processing attempt as completed work.
Calculator
Processing capacity assumptionsFormula
Raw attempts/hour = Workers × 3,600 × Utilization ÷ Seconds per attempt
Successful documents/hour = Raw attempts/hour ÷ (1 + Retry rate)
What the result means
Successful hourly throughput discounts the raw attempt capacity for one additional attempt on the stated share of documents. Daily and page capacity use that adjusted rate.
If failed documents require multiple retries or different processing times, use a measured average attempts-per-success factor instead of treating the retry percentage as exact.
Example calculation
Twelve workers at 75% utilization and 30 seconds per attempt provide 1,080 attempts/hour. With 4% of documents requiring one retry, throughput is 1,080 ÷ 1.04 = 1,038.5 documents/hour, or about 20,769 documents in 20 hours.
Tips for better results
- Measure attempt duration over a representative document mix.
- Track successful completions separately from API calls.
- Use lower utilization when arrivals are bursty.
- Calculate large-document queues separately.
- Revisit retry rate after prompt or OCR changes.
Frequently asked questions
Why divide by one plus the retry rate?
If a fraction of documents needs exactly one extra attempt, total attempts per successful document average 1 plus that fraction.
Is utilization the same as worker uptime?
Not exactly. It represents the effective fraction of theoretical worker capacity available after idle time, coordination, and routine interruptions.
Should processing time include OCR and validation?
Yes. Use end-to-end time for every stage occupying the constrained worker, or model separate bottlenecks independently.
Can page throughput be accurate when document sizes vary widely?
It is an average-based estimate. Segment documents by size when the distribution is broad or processing time is nonlinear.
How do multiple retries change the formula?
Replace one plus retry rate with the observed average number of total attempts required for each successful document.
Related calculators
Throughput outputs
| Metric | Definition | Unit |
|---|---|---|
| Raw attempts | Worker processing starts | attempts/hour |
| Successful throughput | Completed unique documents | documents/hour |
| Daily capacity | Hourly success × operating hours | documents/day |
| Page capacity | Daily documents × average pages | pages/day |