The takeaway
A useful pilot is small enough to inspect and connected enough to answer a buyer’s question. Agree the sample and handling process before transferring records.
Choose the question before choosing the files
A sample evaluation should resolve a decision. Can the buyer reconstruct an issue? Do employee identities stay consistent across applications? Did de-identification break important joins? Name the question first.
For a first discussion, a high-level inventory is enough: systems, date ranges, approximate volumes, the workflow being considered, and any requirements the buyer has already supplied. Keep actual records in the business environment until the scope and transfer process are agreed.
Our quality-ranking overview can help you turn a broad “is this good data?” question into specific checks.
Use a shared period and connected work
A practical starting proposal is 3–5 connected projects or issues from a common 2–3 month period. It is a scoping suggestion, not a minimum required by every buyer or a promise that the sample will represent the whole dataset.
Choose workflows with accessible source records and an internal subject-matter expert who can explain them. Include the relevant email, calendar, chat, shared files, comments, and version history where available.
Document why each workflow was selected. A carefully selected demonstration sample can reveal reconstruction issues, but it cannot establish a population-wide completeness rate. A broader estimate requires a sampling design suited to that claim.
Assemble the context, not just the final outcome
| Item | Question to resolve |
|---|---|
| Workflow list | Which project or issue IDs define the sample? |
| Source inventory | Which systems and record types are included or unavailable? |
| Time window | Are all sources aligned to the same period and timezone rules? |
| Relationships | Which keys connect people, issues, messages, files, and versions? |
| Transformations | What redaction, de-ID, or normalization has already occurred? |
| Known gaps | What would an outside reviewer otherwise assume is complete? |
A connection manifest can organize these answers. Where needed, agree how to reference records just outside the window that explain the beginning or end of a selected workflow.
Agree handling and comparison access
Confirm the authorized recipients, intended review, transfer method, access arrangements, and retention or deletion terms before sharing. If a before-and-after de-ID test is needed, agree how the reference material can be inspected without including protected mappings in the buyer package.
NIST’s de-identification guidance treats the choice of sharing model and disclosure risks as part of planning. Our application here is to settle those questions before the sample leaves its current environment.
Read what a de-ID retention check actually requires. If no controlled baseline is available, label retention as unmeasured instead of estimating it from the delivered files.
Define what you want back from the test
Agree which metrics will be measured, how missing evidence is reported, and which deliverables are included. Useful questions include: Will the findings name the dataset version? Will the report show numerators and denominators? Will it distinguish verified findings from inferred relationships? Can the agreed report be shared with other potential buyers?
Also decide what the result will change. A pilot might justify a broader assessment, identify a limited repair, or show that the chosen workflow is a poor fit. Each is a useful answer if it resolves the original decision.
Start the partner conversation
Visit Clear Harness to discuss a sample quality evaluation. Clear Harness is Ohio Training Data’s partner for business-data quality ranking. Confirm pricing, scope, deliverables, and handling directly with the provider.
If you are still exploring whether licensing makes sense for your company, take the Troveo business assessment (referral link; we may be compensated). The quality review and commercial assessment can inform each other, but neither creates a guaranteed sale.
Sources & further reading
Sources checked October 9, 2026.