Run manual and assisted work on comparable packages. Avoid crediting the tool for a reviewer remembering the answer from an immediately preceding manual exercise; use separate but comparable cases, independent reviewers where available, or a documented order-balancing method.
Record whole-bid results as well as field totals. A useful result sheet includes expected fields, returned fields, wrong values, missing values, critical omissions, unsupported additions, source-link failures and review minutes. Report the number and type of bids so that a tiny clean sample cannot look like broad production evidence.
Calculate:
Net hours released = manual preparation + manual checking
− assisted preparation − assisted checking − correction
Then subtract recurring maintenance and administration at the programme level. Keep elapsed turnaround separate from human effort. A fast extraction job followed by a long queue for checking may not improve delivery time.
Exit evidence: results, failure analysis and a repeatable run configuration.
Sources cited on this page
No external sources are cited on this page. Every source for this section is listed on its Sources page.