Map the real workflow
Write down the steps as they actually happen, including the messy parts. Note the users, the inputs, the handoffs, and the exceptions.
Find the data and the risk
Identify what data each step touches, where it lives, and what cannot leave your boundary. Risk and data sensitivity shape what AI is allowed to do.
Decide where AI helps
Look for steps that are repetitive, slow, or knowledge heavy. Mark where human review must stay. Not every step should be automated.
Set success criteria and cost
Decide what good looks like and what it should cost. A workflow without success criteria cannot be judged later.
When this matters
- You are about to add AI to a process.
- A pilot is running with no clear measure of success.
- Different teams disagree on what to automate.
What to avoid
- Adding AI to a broken workflow and expecting it to fix the process.
- Automating steps that need human judgment.
- Skipping the data and risk review.
- Launching without a way to measure the result.