What you are buying
Choose one intended use and identify the groups that could be affected directly or indirectly. Check what evidence supports your assessment and which changes to deployment, input data or user behaviour would make the original conclusions unreliable.