Pick the first workflow: pain × low risk

The first workflow is a choice, not a brainstorm. A brainstorm produces a list. A choice produces a pilot with an edge.

In short. Score pain and the damage if you are wrong. Pick the high-pain, low-damage job. Say no to the next idea out loud.

Inventory the daily pain

For five working days, write down the tasks that stole the day. Use the shop’s words. Include the work that only happens when someone is out, because that work is usually undocumented and therefore tempting to “just automate” badly.

Group the notes into workflows, not tasks. “Copy the address into the text message” is a step. “Confirm tomorrow’s stops with the customer” is a workflow. You want workflows, because a pilot has to finish something a person recognizes.

Ask the people who do the work which item they would hand off tomorrow if they trusted the draft. Their answer outranks the owner’s theory. Owners feel strategy pain. Desks feel repetition pain. The first pilot should relieve repetition that the owner also cares about, or it will be cancelled the moment the week gets loud.

Score the risk if it fails

Give each candidate a plain risk label:

• Internal draft. A person always edits before anyone else sees it.
• Customer-visible. A wrong line changes what a customer believes.
• Money or record. A wrong line changes an invoice, a schedule commitment, or the system of record.
• Safety or trust. A wrong line could harm someone or break a promise you cannot walk back in a sentence.

The first pilot should sit in internal draft, or in customer-visible only when a named person still sends it. Money, record changes, and safety stay out until the checkpoint has a history.

If every candidate is high risk, you do not have a pilot. You have a wish to skip the awkward middle. Shrink the job until a bad draft is embarrassing at most, not expensive.

Choose one

Pick the workflow with the worst weekly pain that still has a low risk label. Write its name at the top of the AI Pilot Checklist. Fill pain, risk, and the system it lives in today.

Then close the list. The other workflows are not rejected forever. They are rejected for this window. Say that in the same note so the team does not hear “we are only doing one thing” as “your idea did not matter.”

Refuse “while we’re at it”

“While we’re at it” is how a two-week shadow becomes a platform project. A second workflow splits examples, splits the owner’s attention, and makes the decision meeting about feelings instead of outputs.

The test is physical: can you put both workflows’ outputs on one table at the end of the week and compare them to the old way? If not, you picked two. Put one back.

When the choice is made, the next page is the shadow-mode pilot. If you are still unsure which job is real, the pillar roadmap walks the whole sequence from assessment to the keep-or-stop decision.

If you want that choice pressure-tested with someone who builds this for a living, book a Free AI evaluation and bring the one-workflow note. Do not bring a catalog.