Buying an AI system doesn't transfer responsibility for its effects. Procurement is a key point of control.
Before Buying
- Define the need and success criteria.
- Assess risk: will it affect people's rights, safety or opportunities?
Questions for Vendors
- How was the system trained and on what data?
- What testing for accuracy, bias and robustness has been done? Can we see results?
- How does it perform on populations like ours?
- What documentation — model cards, limitations — is available?
- How is our data used, stored and protected?
- How are updates tested and communicated?
- What support is there for explanations and appeals?
Contract Terms
- Data protection and use restrictions.
- Audit and testing rights.
- Notification of incidents and significant changes.
- Liability allocation.
- Exit and data return.
After Buying
Test on your own data before rollout, monitor performance, and review periodically.
Public Sector
Many governments publish AI procurement guidelines worth consulting.