BUSINESS AUTOMATION
Start with a workflow, not a model.
The most useful first AI conversation is often about an ordinary piece of work.
Joseph W. Anady · September 28, 2026 · 4 minute read
Name the work.
“We need AI” does not yet describe a project. “Every Monday, someone spends the morning turning five exports into one report” does. The second statement identifies a process, an owner, an input, a deliverable, and a recurring cost that can be measured.
Record the baseline.
Before changing the process, measure how often it runs, how long it takes, and how much rework it creates. Keep those measurements separate. A faster first draft is not an improvement if reviewing and correcting it takes longer than the original task.
Keep judgment visible.
Separate predictable transformations from decisions that need context or accountability. A workflow can prepare a draft, gather evidence, or route an exception without automatically making the final decision. Design that boundary explicitly.
Test the failure path.
Use duplicate submissions, missing fields, unavailable services, and unexpected inputs in the acceptance test. A workflow that succeeds only on its best example is not ready for everyday work.
Make the decision from the evidence.
Run the scoped process against its baseline. Include operating costs and review time. Expand only after the improvement is visible and someone owns the system. Sometimes the useful result is a simpler non-AI process. That is still a result.
THE NEXT CHAPTER
What would you like to make possible?
Start with a conversation about the work, the idea, or the problem you want to solve.
Talk through your project