01 / PLANNING NOTE
Pick a bounded task
Consider preparing a draft handoff, classifying a non-sensitive inquiry, assembling a status summary, or flagging missing information for review. Start with one workflow and a clear definition of an acceptable result.
02 / PLANNING NOTE
Write down the current process
Record who does the work, how often it occurs, where time is spent, and which errors create rework. This gives the team a baseline for judging whether the pilot helps.
03 / PLANNING NOTE
Keep authority clear
Decide which information the workflow can access and which outputs require human approval. Keep commitments, customer messages, and consequential changes within the agreed authority. Provide a simple path for exceptions.
04 / PLANNING NOTE
Test the difficult cases
Include incomplete inputs, conflicting information, and requests outside the intended scope. Review errors alongside useful results. A successful demonstration is a starting point for evaluation, not proof that the workflow is ready for every situation.
05 / PLANNING NOTE
Measure and improve
Compare quality, time, rework, and adoption with the baseline. Name an owner who reviews performance and can stop or change the workflow. NIST's AI Risk Management Framework is a useful reference for structuring trustworthiness considerations.
Reference material
NIST AI Risk Management Framework
Confirm current provider requirements during scoping.
