Founders
How to build an evaluation dataset for AI document processing
AI document processing: Build an evaluation dataset around the decisions the product must support.
Build an evaluation dataset around the decisions the product must support. Include ordinary cases, difficult cases, and known failure modes with documented provenance. Keep development examples separate from the examples used to judge a change.
Start with the decision
Define the unit being evaluated and the expected result before selecting examples. Record who labeled them, what disagreement means, and how the sample relates to actual use. Revisit coverage when the product or customer workflow changes.
The illustrative scenario
This is an educational workbook. The company or fund and all numerical inputs below are fictional. They are not ADIN results, customer claims, market benchmarks, or a forecast.
The fictional product extracts invoice data for an accounts payable team. Staff review exceptions before information enters the accounting system. Supplier formats and duplicate documents make the input stream uneven.
A misplaced decimal or duplicate record can matter more than an average extraction score. Define the fields and events that require review before comparing workflows.
An evaluation sample with a denominator
Imagine a permissioned evaluation set with 38 cases and 26 that meet a rubric defined before the test. The resulting percentage describes only this invented sample. Seventy percent is not a recommended target, a safety standard, or evidence about any real system. What matters is the rubric, sample coverage, and consequences of the remaining cases.
A misplaced decimal or duplicate record can matter more than an average extraction score. Define the fields and events that require review before comparing workflows. Break the 12 remaining cases into failure categories before interpreting the aggregate. A recoverable wording issue and an unauthorized action may require different product behavior even when both count as a failed case. Keep reviewer disagreement and missing evidence visible.
The assumed time comparison is 8 minutes before and 5 minutes after, including the stated human step. Validate the same start and end points in a real study, then include retries, repair, and handoffs. Compare correct, reconciled records with the current process. Include document intake, exception review, and corrections after export in the time and cost model. A short average time is not enough if the result cannot be used or trusted.
| Item | Illustration |
|---|---|
| Invented evaluation cases | 38 |
| Cases meeting the fictional rubric | 26 |
| Cases requiring investigation | 12 |
| Observed proportion in this exercise | 68.4% |
| Assumed human minutes before / after | 8 / 5 |
Use this worksheet
Answer the questions with a source, date, owner, and next action. Mark an unknown as unknown. Use public or permissioned information and involve a qualified specialist when a question requires expertise beyond the team’s competence.
- What is the evaluation unit?
- Where did the examples come from?
- Who labeled the expected result?
- Which failures are underrepresented?
- What data must remain held out?
Questions specific to this scenario
- Which fields require exact agreement?
- How are duplicate invoices handled?
- Who signs off on reconciliation?
Use permissioned documents, field-level corrections, duplicate flags, and reconciliation results. Keep document type and supplier format visible instead of relying on one blended score.
A concrete next test
Select permissioned invoices from different supplier formats and include a duplicate. Trace each field through correction and export to reconciliation. Ask the accounts payable owner which error would block use even if most fields were extracted correctly.
Interpret the result
A larger dataset is not automatically more representative. An evaluation should reveal consequential behavior in the intended workflow. If repeated tuning teaches the team the answers, add independent examples before claiming a general improvement.
Compare correct, reconciled records with the current process. Include document intake, exception review, and corrections after export in the time and cost model.
Sources and method
These primary resources provide background for the concepts. The fictional scenario, calculations, and worksheet are ADIN educational examples; the sources do not report or validate those invented results. Reviewed October 11, 2026.