Scope & preparation
When this is useful
AI integration can be useful for tasks such as assisting with a draft or helping people navigate an information source. It should begin with a specific task and a way to judge output quality. Adding an AI interface does not remove the need for reliable records or responsible decisions.
How ADPR can help
ADPR can help scope an AI-enabled tool and assess how it fits an existing workflow. Data permissions, provider terms and the consequences of an incorrect response need review. Sensitive or consequential outputs should have an appropriate human check and a usable fallback.
A typical process
Define the task, allowed data and unacceptable failure cases.
Build a limited prototype and evaluate representative examples.
Integrate with monitoring, human review and fallback behaviour.
What to prepare
- A specific use case and sample non-sensitive inputs
- Data-access and confidentiality constraints
- Examples of acceptable and unacceptable outputs
Possible scope outputs
- Use-case and risk-boundary brief
- Prototype or agreed integration
- Evaluation and operating guidance
The proposal will confirm what is included, who is responsible and the applicable timeline. These are scope examples, not a fixed package.
Questions worth asking
Will AI outputs always be accurate?
No. The design should assume mistakes are possible and use evaluation, clear limits and appropriate review.
Can private company data be sent to any AI provider?
Data use depends on permission, contractual terms and the organisation’s policies. Assess these before integrating a provider.