The short answer.

Using AI is a person asking an assistant such as ChatGPT, Claude, Copilot or Gemini for help with whatever is in front of them. Deploying AI is a team running the same written workflow for one piece of work every week, with a named owner, a named person who checks the output, and one current version kept in one shared place. Use belongs to the person and leaves when they move; deployment belongs to the role and stays.

Most companies use AI and have deployed none. That is a fine place to have started. The test of which you have is plain: if the person who is best at a piece of work left for two weeks, would that work come out the same?

Using AI belongs to the person.

Using AI means one person opens an assistant, asks for help and decides what to do with the answer. A sales lead drafts a proposal. A controller asks for a formula. An office manager rewrites a difficult email. Each gets real help, and each works it out alone.

Personal AI use has three traits.

  • The method is in one head.The prompts, the files attached and the judgment about what to keep live with the person who found them.
  • The result varies by person.Two people asking for the same report get two different reports, in two formats, checked two ways.
  • The method leaves with the person.When the person changes role, goes on leave or moves on, the method goes too.

Personal use is worth keeping. Ad hoc work, the kind that never repeats the same way, is best served one question at a time. The trouble starts when the work that does repeat is left to personal use as well.

Deploying AI belongs to the role.

Deploying AI means a piece of repeat work is written down as a workflow, and everyone in the role runs that same workflow every week. A workflow is your best way of doing one piece of work, written precisely enough that the assistant prepares it the same way every time.

A deployed workflow does not depend on who is at the desk. The person who joins the role next month runs the version the person who left was running. The method belongs to the role, its owner keeps it current, and a person still decides on every output. The workflow prepares; your people check and decide.

We all use ChatGPT. Have we deployed AI?

Everyone using ChatGPT, Claude, Copilot or Gemini every day is use, not deployment. Deployment begins with the first piece of work a whole role runs the same way.

Count your deployed workflows, not your seats. Name each piece of repeat work in the company where AI prepares it the same way for everyone in the role. Most companies, after years of experimenting, count zero. The stages of AI at work places that count on a six-stage scale, from switched on to managed.

The five marks of a deployed workflow.

A piece of work has been deployed on AI when it carries all five of these marks. Four out of five is still use.

  1. It is written down.The inputs, the steps, the checks, the output and the point where a person decides are written as instructions, and tested on real cases.
  2. One current version lives in one shared place.The instructions sit in a shared project in your company’s own assistant, on your company’s account. Nobody runs a private copy.
  3. The whole role runs it, every week.Everyone who does the work uses the workflow for it, without being reminded, because it is the way the team does that work.
  4. It has an owner, by name.One person keeps the workflow current, logs what goes wrong and decides when it changes.
  5. It has a person who checks, by name.One person reviews what the workflow prepares before it leaves the team. Anything that sends, pays or changes a record waits for that person’s approval.

These marks define deployment. How to get a workflow there, role by role, is set out in How to deploy AI workflows to a whole team, and keep them working.

The two-week test.

The quickest test of deployment is a question about your best person. If the person who is best at a piece of work left for two weeks, would that work come out the same: same format, same checks, same standard, on time?

If the answer is no, the work runs on one person’s know-how, with or without AI. Their prompts may be excellent, and they are still theirs. If the answer is yes because a workflow carries the method, a colleague runs it and a named person checks it, the work is deployed.

Ask the question for each team’s most important piece of repeat work. The answers give you a first list of what to deploy. When the know-how sits with one person shows how to write the method down with the expert who holds it.

A checklist to copy for one piece of work.

Copy these five lines for one piece of repeat work. Write a name, a place or a date on every line. A blank line means the work is used, not deployed.

  • Written down.Where the instructions are, and the date they were last tested on real cases.
  • One version, one place.The shared project, the assistant it runs in, and the current version number.
  • Run by the role.The role, the people in it, and the date all of them started running it every week.
  • Owner.One name.
  • Checker.One name, and what waits for that person’s approval.

Then ask the two-week test about the same piece of work. Once the lines are filled in, measure the workflow month by month with Installed, used or useful?

One piece of work, from used to deployed.

Illustrative example.

Take a 30-person civil engineering firm. Every Friday each project manager sends each client a progress report. All six project managers use the company’s Claude, each in their own way. One senior project manager writes the reports clients praise: she starts from the site diary and the schedule, flags every change to cost or programme first, and lists the decisions the client owes. Her method lives in her own saved prompts.

The firm asks the two-week test about the Friday report and answers no. So it deploys the work. The senior project manager’s method is written down as a workflow and tested on the last month of reports. The workflow goes in a shared project on the firm’s account, as version one. She becomes its owner, the only person who edits it. All six project managers run it every Friday. The principal on each project checks every report before it goes to the client. When she is on leave, the reports go out in her format, with her checks, run by her colleagues.

What deployed means in a company of under 200 people.

In a company of under 200 people, a deployed workflow means three to eight colleagues in one role running one shared version, owned by someone who does the work. Most published advice on deploying AI is written for large enterprises, with platform teams and release managers. A company your size needs far less.

  • The owner does the work.The person who owns a workflow sits in the team that runs it. There is no separate team to hand changes to.
  • A role is a handful of people.Getting the whole role onto a workflow is one meeting and a first week of reviewed outputs.
  • The assistant is already paid for.The workflow runs in the ChatGPT, Claude, Copilot or Gemini seats you have, on your own account, run by your own people.
  • One gap shows fast.When the best person in a five-person role is away, the role’s best method is away with them. Deployment closes that gap first.

Deploying is the third of three steps in How organizations plan their use of AI: plan it, develop the workflows, deploy them. The plan decides which pieces of work to deploy first, and in which team and role. Set out your teams, their roles and the workflows each one runs in FullyPowered Planner, and choose what to build first.