Insight
How to Write an RFP for Expert AI Data
A practical RFP structure for buying managed expert capacity: profile, artifacts, qualification, review mix, security, and pilot criteria.
Published 2026-08-26 · Written by the Stratum editorial team. No individual author or outside clinical or legal reviewer is named for this article.
An RFP for expert AI data should specify the cohort and the quality system, not ask vendors to “provide annotators.” If the document could be sent unchanged to a crowd platform, it is not yet an expert-data RFP.
What the RFP must define
Write the Expert Capacity Specification first: domain, credential, experience, geography, task, volume, availability, and quality threshold. Then add data-access, security, and acceptance rules.
Before sourcing experts, define the specification the project will be staffed and measured against.
- Domain
- Credential
- Experience
- Geography
- Task
- Volume
- Availability
- Quality threshold
Suggested RFP sections
- Purpose and work types (annotation, evaluation, benchmark, post-training).
- Expert profile, including what is out of scope.
- Example artifacts or a synthetic sample.
- Qualification and calibration process the buyer will accept.
- Review mix: single, dual, adjudication, gold-set size.
- Environment: client tools versus export, and who provisions access.
- Reporting: vacancy, throughput, agreement, escalations.
- Replacement and scaling rules.
- Commercial breakdown: production, review, management.
- Pilot design and go/no-go criteria.
What not to ask
Do not require a public logo wall, an invented headcount, or a certification the work does not need. Do not ask for named individuals in the first response if confidentiality or availability makes that meaningless.
Pair this template with the due diligence checklist and the outsourcing page.