Practical checklist
Prepare the workflow before you evaluate the technology.
Use this checklist with the people who own and perform one recurring workflow. The goal is not to prove AI belongs everywhere; it is to make a responsible next decision possible.
Five checks
Get the relevant facts onto the table.
If an answer is unknown, write it down as a gap. An assessment can clarify gaps; a tool purchase cannot make them disappear.
Current work
- Name the trigger that starts the work.
- Map the systems, handoffs, decisions, exceptions, and rework.
- Ask the people who do the work where judgment is essential.
Value and baseline
- Count the frequency, cycle time, waiting time, and rework.
- Record the quality, service, or risk consequence of delay.
- Choose one measure the owner will revisit after a change.
Data and access
- Identify the source systems, representative examples, and data owner.
- Classify sensitive, personal, regulated, or confidential information.
- Confirm what access and retention constraints apply before using real data.
Controls and adoption
- Define which steps must remain human-reviewed.
- List harmful failure modes, escalation paths, and correction mechanisms.
- Identify the users who need training, context, or a way to challenge the output.
Ownership and decision
- Name the sponsor, process owner, technical contact, and frontline participants.
- State the decision to make: implement, test, buy, defer, improve the process, or stop.
- Write the smallest responsible next step and its owner.
A responsible boundary
Do not move sensitive information through this checklist.
Use placeholders and ranges in a public or shared document. Before representative data enters an AI tool, the team should establish the data flow, access, retention, human review, and contractual requirements.