The short answer.
To write down how a piece of work gets done before you give it to AI, capture six things: the trigger that starts it, the inputs it starts from, the steps in order, the judgment calls and the reason behind each one, what a good finished result looks like, and the one person who approves it. Capture them by walking through last month’s real cases with the people who do the work, because a process described from memory leaves out the exceptions. The judgment calls are what turn a written process into a workflow that prepares the work the way your best person would.
How to hand a piece of work to AI shows one person writing down their own work. This piece is for a leader capturing a team’s work, where the method sits across several people and nobody has written it down before.
The six things to capture.
Every piece of work that can become a workflow has the same six parts, and a written process missing any one of them produces a workflow that guesses.
- The trigger.The event that starts the work: a form arrives, an order lands, the month closes, a client asks. Without one, the work runs when someone remembers.
- The inputs.Every record the work starts from, and where each one lives: an export from your order system, an email, a spreadsheet, a file on the shared drive. Name the system and the field.
- The steps.What the person does, in the order they do it, in the team’s own words.
- The judgment calls.Each point where the person stops and decides, what they look at, and why they decide one way or the other.
- What good looks like.One finished example the team is proud of, and what makes it good. A workflow aims at an example far better than at a description.
- Who approves.One person, by name or by role, who reads the output before it goes anywhere. The workflow prepares; that person decides.
Write it from real cases, not from memory.
A process written from real cases is accurate; a process written from memory describes how the work should go, not how it goes. Open ten finished cases from last month with the person who did them, and you get the missing form, the customer who always sends the wrong file and the step they skip when the order is small.
Take the cases in three groups: a few ordinary ones, a few that went wrong and a few that needed a decision. The ordinary cases give you the steps. The ones that went wrong give you the checks. The ones that needed a decision give you the judgment calls. Keep the cases; they become the test set when the workflow is built.
Ask everyone who touches the work, not only the lead: each knows a part nobody else does.
Judgment calls are rules with a reason.
A judgment call written down well is a rule with its reason attached: “If the delivery date is inside five working days, flag it to the lead, because rush work needs a person to agree the date.” The reason matters as much as the rule. With the reason, a workflow handles the next case that does not quite match; without it, the workflow follows the letter and misses the point.
Ask three questions at every pause in a real case: what did you look at, what would have made you decide the other way, and who would you ask if you were unsure. The third answer tells you where the workflow stops and hands the case to a person. Some judgment calls stay with a person for good, and the written process says so plainly. What work should you not give to AI? covers where that line sits.
A capture sheet to copy.
Copy these headings onto one page for each piece of work. Fill the page in with the people who do the work, with real cases open in front of you. Write “Unknown” where nobody knows, with the name of who will find out.
- The piece of work.Its name, the team that does it and how often it runs.
- Trigger.The event that starts it, and where that event shows up.
- Inputs.Each record, the system or file it lives in, and what is often missing.
- Steps.Numbered, in order, in the team’s words.
- Judgment calls.For each: what the person looks at, the rule, the reason, and when it goes to a person.
- What good looks like.One finished example, attached, and three things that make it good.
- Who approves.One name or role, and what they check before it goes out.
- Cases used.The real cases you walked through, kept for testing.
- Method owner.The person who keeps this page current when the work changes.
A dental laboratory writes down case intake.
Take a 30-person dental laboratory that makes crowns, bridges and dentures for dental practices. The owner chose case intake as the first piece of work to write down: checking each new case before production starts, so technicians never begin a crown on an unclear prescription.
Illustrative example.
- Trigger.A new case arrives: a digital scan lands in the lab’s case management system, or a box with an impression comes to the front desk.
- Inputs.The dentist’s prescription, the scan or impression, the shade notes and photos, and the practice’s preferences kept on file.
- Steps.Match the case to the practice, read the prescription, check the scan, check the shade information, set the due date, write the case ticket for production.
- Judgment calls.The intake lead and two senior technicians walked through last month’s cases, including the remakes. Three calls came out. If the margin on the scan is unclear, hold the case and call the practice, because a crown made to a guessed margin comes back. If the shade note gives a shade with no photo, check the practice’s file before asking, because some dentists always mean the same thing. If the requested material does not suit the tooth, a senior technician decides, never intake.
- What good looks like.A case ticket a technician starts on without walking over to ask a question.
- Who approves.The intake lead reads every ticket and every hold before it goes to production or back to a practice.
The page took the senior technicians’ knowledge of each practice out of their heads and onto paper, and the remakes they walked through became the cases the workflow is tested on before it handles a live one. FullyPowered Studio is how we capture existing know-how like this as workflows: we interview the people who do the work, walk their real cases with them and build the workflow they check line by line. When the know-how sits with one person covers the case where the whole method belongs to one expert.
What is different in a company of under 200 people.
In a company of under 200 people, the people who do the work are the people you capture it from, and they are down the hall. You sit with two or three of them, open last month’s cases and fill in the capture sheet together.
- The expert is also the approver.The person with the judgment calls often reads the output too. Write both roles down, and name who covers when that person is away.
- Nothing is written down yet.Most small companies have no process documents. The capture sheet is the first one, and it is enough to build from.
- The inputs are already yours.Your order system, your accounting system and your shared drive hold the records. A workflow can be designed to read from them, with a person approving anything written back.
Once the page is written, the work is ready to become a workflow in the ChatGPT, Claude, Copilot or Gemini you already pay for. We have a plan. How do we build the first AI workflow? takes it from there, and we build it with you in a 3-Day Workflow Sprint. FullyPowered Planner is where the plan lives: you set out your teams, their roles and the work between them, and choose which piece of work to write down first. How organizations plan their use of AI covers the whole method.