JwumaDue diligence checklist for AI data suppliers
· 3 min read · procurement, compliance
Ask how provenance is evidenced per item, how residency is enforced technically rather than contractually, how credentials are verified and how the supplier treats its own workforce. That last set is now a supply chain risk, not an ethical extra.
The short version
- Provenance should be demonstrable per item on rows you choose.
- Residency should be enforced technically, not only contractually.
- Credential verification differs materially between register and document checks.
- Worker treatment is a supply chain risk under emerging regulation.
- Published worker sentiment scores are an available external signal.
Scoping something like this?
We would rather talk about the problem than quote against a volume figure.
See capabilitiesThe data questions
How is provenance evidenced for a specific row I pick. What happens to capture metadata through your delivery pipeline. What consent scope was obtained and does it cover commercial training. Which rubric version produced this label and can you show me the rubric.
Ask these about actual delivered samples rather than in the abstract. The gap between a supplier's process documentation and their delivered artefacts is where the useful information is.
The compliance questions
Where is residency enforced in your system. What happens when an out-of-region account attempts access. Do backups and logs stay in region. Where do your support staff sit.
For expert work: are credentials checked against an issuing register or read as documents, is the check dated and what happens for professions with no public register.
The workforce questions most buyers skip
How are your contributors paid and on what schedule. Is there unpaid production work. Can an account be closed without a stated reason. Is there an appeal route. Are accept and reject decisions made by a person or automatically.
These used to read as ethical questions and now read as risk questions. Emerging regulation addresses algorithmic management of platform workers directly, and a supplier carrying that exposure passes it to you.
- Payment schedule and method by country
- Whether unpaid production work exists
- Whether accounts can close without reason
- Whether an appeal route exists
- Whether decisions are automated or human
What external signals exist?
Published worker sentiment indices now cover the larger platforms in this sector, and several of them score poorly and are trending down. That is public information and it is a reasonable input to a supplier decision.
A workforce that is leaving is a quality problem before it is anything else, because the people who understood your rubric are the ones going.
People also ask
What should I ask an AI data vendor about provenance?
Ask them to evidence provenance for specific rows you choose, including capture metadata, consent scope, contributor attribution and the rubric version applied. Process documentation is not evidence.
Why does a data vendor's treatment of workers matter to me?
Because emerging regulation addresses algorithmic management of platform workers directly, so a supplier's exposure becomes yours. Separately, a workforce that is leaving takes your accumulated rubric knowledge with it.
How do I check credential verification claims?
Ask whether credentials are checked against an issuing body's public register or only read as documents, whether the check is dated and what happens for professions with no published register.