Describe who is doing what today, where time or quality is lost, what a better outcome looks like, and how improvement can be observed. Avoid broad goals such as “use AI” without a defined user, decision, or workflow.
Useful questions:
Who experiences the problem?
What decision or task should improve?
What should remain human-led?
What evidence would justify continuing?
02 · Foundation
Assess information, systems, and ownership
Review whether the required information is accurate, accessible, current, permissioned, and owned. Identify systems the solution must read from or write to, and whether safe integration paths exist.
Readiness also includes subject-matter reviewers, product ownership, security, privacy, legal context, user support, and the capacity to improve the solution after launch.
03 · Risk
Match controls to the consequence of error
Consider harm from an incorrect answer, missing information, biased output, inappropriate disclosure, or unintended action. Higher-consequence work needs stronger source controls, human approval, auditability, testing, fallback behaviour, and operational limits.
04 · Pilot
Design a pilot that creates evidence
Choose a bounded audience and workflow, define a baseline, prepare representative evaluation examples, and agree on quality, safety, latency, cost, and adoption measures. A pilot should answer a decision—not simply demonstrate that a model can generate output.
Have a project in mind?
Turn the guidance into a practical next step
We can help assess the use case, prototype the experience, evaluate behaviour, and plan responsible delivery.
