September 30, 2026
AI & systems
Assign an owner before automating a workflow
An AI workflow belongs to the business process it supports. Before building it, name the person responsible for the result, the information it can use, and the actions it is allowed to take.
Start with a bounded task. Summarizing an intake form for a team member is different from sending a customer an offer. The second action creates a commitment and needs an explicit review policy.
Write down what good output looks like. Test typical requests alongside incomplete records, conflicting instructions, duplicated submissions, and unavailable services. Define when the workflow should stop and ask a person.
Keep a record of important actions and give the team a way to recover. If the system is unsure whether a message was sent, it should not blindly send it again. A human needs enough context to resolve that uncertainty.
Measure the whole process, including review time and corrections. A faster first draft is useful only when the final work meets the standard and the team understands how to maintain it.
AN EXAMPLE FROM THE WORK
A rendered book still needs someone to release it.
The personalized-book analysis separates rendering, technical review, approval and production submission. The historical record reached proof review while production submissions remained at zero. A valid file was one stage of the process, not evidence that a book had been manufactured or shipped.
- Name the person who can authorize the external action.
- Attach approval to the exact version being released.
- If the provider response is lost, check whether submission happened before retrying.
Apply the same question to a customer message, a booking or an order: who can authorize the action, and how will that person resolve an uncertain result?
Read the release and recovery analysisDiscuss a similar workflow