Every team has the person everyone asks. They look at a late shipment, an odd request or a half-finished file and know in a minute what to do. Now picture the whole team making that call the way they do: the new coordinator in their third week, the colleague covering a holiday, the night shift. An AI assistant prepares each case the way your expert would, and a person makes the decision. That is how you 10X what your team can do with the people you have.
It starts with one conversation, and the questions decide how much of that judgment you capture. The conversation is where our method lands a capability on the role that does the work: start from the outcome, capture the judgment, prove it on real cases, then roll it out.
Ask about cases, and judgment becomes answerable.
Ask an expert “How do you do this?” and you get the tidy version: four steps and “then it depends”. Everything valuable is inside “it depends”. Ask what they did on Tuesday’s case, what they opened first and what would have changed their mind, and the same person gives you rules, thresholds and the names of the people they call.
Judgment has six parts you can ask about: what starts the work, what goes into it, how the call is made, what breaks the pattern, what gets handed on and who approves it. Those six answers are also what a capability needs written down before an AI assistant can prepare the work. Bring the work as your team does it today explains why real cases beat a blank template.
The interview guide.
Book an hour. Ask the expert to bring three recent cases: one routine, one awkward and one they would handle differently now. Put every question to a case, and write the answers in their words.
Interview guide: [the piece of work]
- Triggers. What tells you this work has started? Who or what brings it to you, and how often? What do you see that makes one urgent? When does something that looks like this job turn out to be a different job?
- Inputs. What did you open first on this case, and what next? Which records, messages and people do you consult? When two sources disagree, which one wins? What is usually missing, and where do you get it?
- Judgment. What did you decide here, and what made you decide it? What would have changed your mind? What do you look at that a newer colleague overlooks? Which rule of thumb do you use, and when does it fail?
- Exceptions. Take the awkward case: what made it awkward, and what did you do instead? When do you stop and ask, and whom? What must never happen, even once?
- Outputs. What do you hand over, to whom and in what form? Show me one you are proud of: what makes it good? What does the next person do with it, and what would make them send it back?
- Approval. Who signs off before this leaves the team, and what do they check? What goes out without review? Which decisions stay with a person, whatever the AI prepares?
- Close. What did I not ask? Who does this differently from you, and which way should become the standard?
For work that already has a written procedure, the shorter conversation in Your SOP describes the happy path. Where are the exceptions? adds what the document leaves out. This guide is for the judgment that was never written at all.
One hour, written down.
Here is what the guide produces for one judgment call at an invented company: what to do when a shipment is going to miss its delivery date.
Illustrative example with invented data. An invented 60-person freight forwarder. Notes from one hour with its senior shipment coordinator.
Interview notes: a shipment at risk of missing its delivery date
- Triggers
- A carrier delay notice, or no tracking scan for 24 hours on a shipment due within three days. Urgent when the customer has a booked receiving slot.
- Inputs
- The booking, the carrier’s tracking, the customer’s delivery terms and the account notes. When the carrier’s portal and the driver disagree, the driver wins.
- Judgment
- Tell the customer before they ask, with two recovery options and the cost of each. Offer expedited freight only when the customer’s terms carry a late-delivery penalty.
- Exceptions
- Temperature-controlled goods: phone the customer at once. A customs hold: stop and ask the compliance lead. Never promise a new date the carrier has not confirmed.
- Outputs
- A drafted customer update with the cause, the confirmed position and the options, plus a one-line note on the shipment record.
- Approval
- The account manager reviews and sends every update. Any recovery cost above $500 needs the operations manager.
From these notes the workflow prepares the update and the options. The account manager decides what the customer hears and what the company spends.
Six short answers, and a new coordinator can prepare the case the way the senior one would. From one person’s know-how to a method the whole team runs shows the next step: the notes become a procedure, a job aid, test cases and a named owner.
The expert shapes what gets built.
Capturing a colleague’s method increases their influence on how the work develops. Their rules become the team’s standard, their awkward cases become the tests, and nothing is built until they have corrected the written version. Tell them so before the interview, and they answer as the person setting the standard. Build your AI plan with the people who do the work shows how to keep their suggestions flowing after the first conversation.
Three checks keep the result sound. Where two experts answer differently, the owner of the result chooses the standard. The three cases from the interview become the first entries in the test set that proves the workflow is ready. And the approval answers hold: AI prepares, and the people named there decide.
The finished workflow runs in the platform your team already uses: ChatGPT, Claude, Copilot or Gemini. You set the pace, and your organization keeps control of how the work gets done.
Bring the judgment call your team waits on one person for, and the role of the person who makes it.