The short answer.

You build your first AI workflow by writing down how your best person does one piece of work from your plan, precisely enough that an AI assistant prepares it the same way every time. The workflow lists its inputs, steps, checks, output and the point where a person decides. You test it on last month’s real cases, then hand it to one owner, by name, who runs it with the team in the AI your company already pays for.

Building workflows is step two of the three-step method for planning AI: plan it, develop the workflows, deploy them.

Build the first workflow in five moves.

The first AI workflow takes five moves, in this order.

  1. Pick the piece of work from the plan.Take the first workflow your plan names for one team: work that repeats every week or month, starts from records someone can export, and already gets reviewed by one person.
  2. Capture how your best person does it.Sit with the person who does the work best and record what they start from, the order they work in, what they check, what they leave out and what good looks like.
  3. Write it as instructions.Inputs, steps, checks, output and the point where a person decides. Plain language, in the order the work happens.
  4. Test it on last month’s real cases.Run the workflow on cases where you already know the right answer, and record every miss.
  5. Hand it to its owner.One person keeps the workflow current, runs it first, then brings in the rest of the team.

A first workflow covers one piece of work, start to finish. Resist adding a second piece of work to it. A narrow workflow is easier to test, and the owner can tell quickly whether it helps.

How do we turn a procedure, or what our best person does, into a workflow?

You turn a procedure or a person’s know-how into a workflow by adding what the AI assistant needs and a written procedure leaves out: the exact inputs, the checks, the format of the output and the point where a person decides.

If the work has a written procedure, start there. A procedure tells a person what to do and trusts them to fill the gaps. An AI workflow fills the gaps in writing: which file it starts from, which fields matter, what to flag and what a finished output looks like. From SOP to AI workflow walks one procedure through that change.

If the work lives in one person’s head, interview them on real cases. Ask them to talk through three recent ones: what they looked at first, what made them stop, what they always check and what they never bother with. Those answers are the steps and checks of the workflow. From one person’s know-how to a team method gives the questions.

Are templates enough, or do we need workflows built for us?

Workflow templates are enough for common work, and work particular to your company is built from how your own people do it. A template gets you most of the way on work every company does in a similar way, such as preparing a meeting brief or reconciling a statement. You customize it with your own inputs, your own checks and your own approver.

Work that is particular to your company, such as how you price a job, qualify a client or set up an account, needs a workflow built from your people’s know-how, because no template knows your rules. The Workflows Directory holds free workflow templates by team and by industry. They are samples and a starting point, and we can build any workflow for your company.

How do we test a workflow before people rely on it?

You test an AI workflow by running it on real cases from last month, where you already know the right answer, and comparing its output with what your best person did. Pick ten to twenty cases, including the awkward ones: the incomplete record, the exception, the case that went wrong.

  • Keep a record of every miss.What the workflow produced, what was right and what changed in the instructions.
  • Run every case again after each change.A fix for one case can break another.
  • Decide what ready means before you start.The owner agrees which misses are acceptable for a person to catch and which are not.

What we test before we call an AI workflow ready shows the full record.

Build workflows before agents.

A small company builds workflows first and agents later. A workflow that a person starts and checks teaches you what the work needs: which inputs are reliable, which checks catch real problems and where a person’s judgment matters. That knowledge is what anything more automatic is made from. A workflow can also be designed to pull from the systems you already run, such as your accounting system, and push data back, with guardrails and a person approving. Start with the workflow, and make it more automatic once the test record shows where it is safe to.

A one-page workflow sheet to copy.

Copy these headings into a document and fill them in with the person who does the work best. One sheet per workflow.

  1. The work.One sentence: the piece of work this workflow prepares.
  2. Who starts it.The role that runs the workflow, and when.
  3. What it starts from.The files, exports or notes it needs.
  4. The steps.The order the work happens in, one line each.
  5. What it must check.The checks your best person always makes.
  6. What it produces.The output, and what a good one looks like.
  7. Who decides.The person who approves the output before anything is sent, paid or changed.
  8. How it is tested.Which past cases it runs on, and what counts as ready.
  9. The owner.One name: the person who keeps it current.

Illustrative example.

Take a 60-person distributor. Its plan names the customer service team’s first workflow: preparing a reply to a customer asking where an order is. The customer service lead is the best person at it, so she talks through three recent requests: she checks the order record, then the shipping export, then any note from the warehouse, and she always flags a partial shipment. Those become the steps and checks. The workflow starts from the customer’s message and the two exports, and produces a draft reply with any problem flagged at the top. The customer service lead decides on every reply before it goes out. The team tests it on last month’s requests, including the partial shipments and a wrong address. She owns the workflow, runs it for two weeks, then brings in the rest of the team.

What is different in a company of under 200 people.

In a company of under 200 people, the person who does the work best is down the hall, and you can build the first workflow with them in days. Most published advice on AI workflows assumes a platform team, a data project and a long pilot. You need none of them to begin.

  • The expert and the owner are often the same person.The one who does the work best usually owns it too, so capture and handover happen together.
  • The records are already exportable.Your accounting system, your order system and your shared folders export what a first workflow needs.
  • The assistant is already paid for.Workflows run in your own ChatGPT, Claude, Copilot or Gemini, on your account, run by your people.

Where we come in.

FullyPowered AI develops workflows with you. FullyPowered Studio captures existing know-how as workflows, from a written procedure or from the person who does the work best. A 3-Day Workflow Sprint or a 2-Week Workflow Sprint takes one piece of work from how your team does it today to a tested workflow your people run. You will find every price on the pricing page.

The first workflow comes from your plan. How organizations plan their use of AI shows the whole method, and FullyPowered Planner is where the plan lives: your teams, their roles, the work between them, and what to build first.