The short answer.

You start with one decision: a one-page AI plan, written in a single meeting with your team leads, that names your teams, the work each one repeats, the one piece of work each moves onto AI first and the person who owns it. Then the team that goes first writes down how its best person does that piece of work, builds it as a workflow in the assistant the company already pays for, tests it on last month’s real cases and puts it into daily use with one named person checking every output. Meanwhile everyone writes better single requests, and the whole company hears what is decided, what each team is asked to do and what stays the same. When the first workflow runs without anyone asking, you choose the second from the plan.

The articles behind this guide each settle one of those decisions. Read together, they show something none of them says alone: every choice in starting turns on the same four questions. Does the work repeat? Are its records where someone can open them? Does one person check it before it leaves the team? Will one person own it? The five ways AI helps, the stage a piece of work has reached, the team that goes first and the order of the work are those four questions asked at different sizes. Learn them once, and each next decision is quick.

Starting is the first stretch of the three steps in How organizations plan their use of AI: plan it, develop the workflows, deploy them.

Feeling behind ends with one decision, not a hundred.

An owner who feels behind on AI stops being behind on the day the company has one page that says which teams start, what each builds first and who owns it. The feeling is heavy because it has no edges; a page gives it edges, and from then on you are on a step of your own plan.

Experimenting was the right way to begin: your people learned what an assistant does well and where it slips. After years of experimenting, the next step is a decision about direction, which only the person who runs the company can make. Without it, AI use stays undirected: people are unsure what is allowed, team leads wait, and good methods stay with whoever found them. I run the company and I feel behind on AI. Where do I start? sets out the first week, month and quarter, and Why companies get so little from the AI they already pay for explains why the gap does not close by itself.

Starting well is a short sequence of decisions, each made once, in this order.

  1. Which work, and which team?The repeating work worth a workflow, and the team that goes first.
  2. What goes on the page?The one-page plan, with an owner on every first workflow.
  3. What do we tell people?One announcement, repeated by each team lead.
  4. How is the work done today?Written down from real cases.
  5. Is the workflow ready?Tested on finished cases, then handed to its owner.
  6. What comes next?Chosen from the plan.

Every piece of work climbs four stages: Ask, Save, Share, Sustain.

Any piece of work done with AI goes through four stages: Ask, where someone types the request from memory each time; Save, where the method that worked is written down so it runs the same way every time; Share, where the whole team runs that one version; and Sustain, where one named person owns it and checks on a set date that it still does the job. Starting with AI means moving one named piece of work up one stage at a time. Ask, Save, Share, Sustain follows one piece of work up the whole ladder.

The stages describe a piece of work, not a company: the Monday report can sit at Share while the reply to a complaint sits at Ask, in the same team. What the ladder adds is a map of which decision moves work up each rung.

  • Ask needs better requests.Nothing is built; each person writes a fuller request.
  • Save needs the work written down.Inputs, steps, checks, the finished output and the decisions a person keeps.
  • Share needs a plan.Which work gets a shared workflow, in which order, and which version is current.
  • Sustain needs an owner.One person, by name, who keeps the workflow right when the work changes.

The first rung asks nothing of the plan, so start it on day one. A single request gets a better result when it carries four asks: time back on the whole routine part, fewer misses through a check against a standard you name, a better decision through options and objections set out, and work you could not staff through the view of a role you do not employ. The four asks has a four-line template. The larger gain still starts at Save, because a request retyped every week is a method nobody keeps. Better requests and the plan run side by side.

Which work should AI help with first?

AI should help first with work that happens every week, that it can help with in at least three of five ways, and whose output one person already checks. The five ways are Gather, Check, Draft, Think it through and Repeat. Walk one team’s week with its lead, mark each repeating piece of work with the ways that apply, and the weekly work that touches three or more is where a first workflow comes from. The five ways AI helps with a piece of work has the worksheet for that walk.

The lens finds candidates; a second pass removes the wrong ones. Three kinds of work stay off a first plan: work that happens a few times a year, work whose whole value is one person’s judgment in the moment, and work where a mistake would leave the building before anyone saw it. That work still gets help from the assistant one request at a time. Three things stay with a person in every workflow: decisions about a person, commitments on the company’s behalf, and the call when the facts disagree. What work should you not give to AI? has the four-question test that sorts each piece of work.

The two passes are one judgment seen from two sides. Repeat, the fifth way, is the same question as the test’s first, and Check only helps when someone owns the standard. The lens asks what AI can prepare; the test asks what a person must still decide.

Which team goes first, and in what order does the work come?

The team that goes first is the one whose week repeats most, whose work starts from records someone can export, where one person already reviews the output, and whose lead says yes to owning the first workflow by name. The keenest team often does work that changes shape every time, and the team with the biggest fire has a lead with no room to test; both go on the plan with a date. Which team should go first with AI? has a comparison sheet, one column per team.

Inside each team, the order of the work comes from a score: how often it happens, how much friction it carries, whether its inputs are ready, what a miss would cost and how much there is to set up. Write “Unknown” wherever nobody knows; an unknown becomes a question with a name and a date beside it, never a zero. Choose your first AI wins has the scorecard. Start one workflow per team, and no more than two or three across the company at once, because each needs its owner’s attention while it is tested.

Set the team comparison, the four-question test, the five ways and the scorecard side by side, and the same four questions sit under every one. That is why one sheet can carry the whole choice.

The starting sheet to copy. One row per piece of work, filled in with each team lead. Write “Unknown” where nobody knows, with who will find out.

  1. Team and lead.______
  2. The piece of work.Its plain name, as the team says it: ______
  3. Does it repeat?Weekly, monthly, or less often: ______
  4. The ways AI helps.Circle each: Gather / Check / Draft / Think it through / Repeat
  5. Where its records are.Exportable today, scattered, or in people’s heads: ______
  6. Who checks it now.One name, or “nobody yet”: ______
  7. Who would own it.One name, and whether that person has said yes: ______
  8. What stays with a person.The decisions the workflow prepares and never makes: ______
  9. The call.First, next, later, or a person’s own use, with one line of reason: ______

A piece of work comes first only with weekly or monthly in row 3, a name in row 6 and a yes in row 7. A team goes first only when one of its rows does.

Write the one-page plan in one meeting, then tell everyone what it says.

A company’s first AI plan is written in one meeting, chaired by the person who will own it, with the lead of each starting team and a note-taker who writes the page as the room answers. Each lead brings five pieces of work the team repeats, and the room answers six questions in order: where are we going, which teams go first, what repeats in each, which piece comes first and why, who owns each, and when do we review. How to write your company’s AI plan in one sitting has the agenda to paste into the invitation.

The plan needs a home every lead can open. FullyPowered AI Planner is where the plan lives: you set out your teams first, then their roles and the work between them, choose what each team builds first and put an owner on every workflow, and your colleagues add their own teams.

A one-page plan to copy. Fill in each line before the room moves on.

  1. Where we are going.One sentence every lead can repeat: ______
  2. Owner of the plan.One person, by name: ______
  3. The assistant we use.Which one, on the company’s account: ______
  4. Teams that start, and teams that wait.Each with its lead; each waiting team with its reason: ______
  5. Work that repeats.Five pieces per starting team: ______
  6. First workflow per team.What it starts from, what it produces, and why it comes first: ______
  7. Owner and checker.Who keeps each first workflow current, and who approves every output: ______
  8. Review.The fixed day each month the page is looked at: ______

Announce the plan on the day it names a first team, a first workflow and its owner, because an announcement with nothing concrete in it gets filled with guesses. The person who runs the company says four things to everyone: what is decided, what each team is asked to do, what stays the same, and where to bring ideas. What stays the same carries the most weight: a person decides and a workflow prepares, and whoever signs off on a piece of work today still signs off on it. Each team lead repeats it to their own team within the week, and anyone using a personal AI account for company work moves onto the company’s account. What do we tell employees about AI? has a script to fill in.

Write down how the work gets done before AI prepares it.

Before a piece of work becomes a workflow, someone writes down what starts it, what it starts from, the steps in order, the checks and judgment calls with the reason behind each, what a good result looks like, and who decides. An assistant given a short request fills every gap with a guess, and next week it guesses differently; writing the work down removes the guesses. How to hand a piece of work to AI shows one person writing down their own weekly work, and How to write down how work gets done before you give it to AI shows a leader capturing work that sits across a team.

The two articles agree on the method and differ on who holds the pen. For a team’s work, sit with the people who do it and open ten finished cases from last month: ordinary ones for the steps, ones that went wrong for the checks, and ones that needed a decision for the judgment calls. Memory describes how the work should go; real cases show how it goes. Keep the cases, because they become the test set.

Write each judgment call as a rule with its reason: “If the client’s deadline falls inside a week, flag it to the lead, because short deadlines need a person to agree the date.” With the reason, a workflow handles the next case that does not quite match. When the method lives in one expert’s head, FullyPowered AI Studio is how we capture that existing know-how as a workflow: we walk the expert’s real cases with them and build the workflow they check line by line.

A capture sheet to copy. One page per piece of work, with real cases open in front of you.

  1. The piece of work.Its name, the role that does it, how often: ______
  2. What starts it.The event, and where it shows up: ______
  3. What it starts from.Each record by name, where it lives, how fresh it must be, and what to ignore: ______
  4. The steps.In order, one line each, starting with a verb: ______
  5. Checks and judgment calls.For each: what is looked at, the rule, the reason, and when it goes to a person. A missing figure is flagged, never filled: ______
  6. What good looks like.One finished example attached: ______
  7. What a person decides.Every decision the workflow prepares and never makes: ______
  8. Who approves, and the cases used.One name, and the real cases kept for testing: ______

Build the first workflow and test it on last month’s cases.

The first workflow is built in five moves: pick the piece of work the plan names, capture how the best person does it, write it as instructions, test it on finished cases where the right answer is known, and hand it to its owner. The capture sheet covers the second and third; what remains is the test and the handover. We have a plan. How do we build the first AI workflow? walks through each move.

Testing is what lets a team rely on a written method. Run the workflow on ten to twenty real cases from last month, including the incomplete record and the case that went wrong, record every miss, and rerun every case after each change, because a fix for one case can break another. The owner runs it first and brings in the team once it holds.

Common work can start from a free workflow template in the Workflows Directory; templates are samples and a starting point, and work particular to your company is built from your own people’s know-how. The workflow runs in the ChatGPT, Claude, Copilot or Gemini your company already pays for, on your account, and a 3-Day Workflow Sprint builds one with you from how your team does the work today. The first 90 days of an AI adoption plan lays the same path across a quarter, from baseline to the decision to expand, adjust or stop.

What comes after the first workflow?

After the first workflow comes the workflow that receives its output, or the same piece of work in the next team that repeats it, chosen from the plan once the first is in daily use, meaning the people who own the work run it without being asked. Before anything new starts, the team writes what it learned back into the procedure, the first owner sits beside the second through the next build, and each team with a workflow in daily use starts one more, with two or three in development across the company at most. What comes after the first AI workflow? has a now, next, later sheet. A team that hears “you are next” waits happily; a team that hears nothing starts its own experiment.

Good methods spread the same way. A shared folder of prompts decays, because no prompt in it has an owner, says what it starts from or names who checks it. Collect what people actually ran, keep one version for each piece of work, build it as a workflow and attach it to the role that does the work, so the next hire finds it by asking what the role runs. How do we share what works with AI across the company? has the collection form, and How to roll out AI across a company, and tell whether it is working takes over once several teams are under way.

Here the ladder closes its loop: attaching a workflow to a role is Share, and naming its owner and a review date is Sustain.

One company’s start, from the first page to the second workflow.

Illustrative example.

Take a 50-person civil engineering firm with four teams: design, project management, project accounts and business development. Engineers use an assistant for emails and meeting notes, each in their own way, and the managing director decides to write one page.

Before the meeting, each lead walks the team’s week through the five ways. Design’s calculation checks rest on an engineer’s judgment and stay with the engineers. Business development’s proposals differ every time, so they go on the plan as later. Project accounts prepares a monthly invoice for each project from timesheets and the project tracker, but the finance manager is changing the accounting system and says not now. Project management sends each client a weekly progress report from the project tracker and site notes; it gathers, checks, drafts and repeats, a project manager reads every report before it goes, and the head of project management says yes to owning it. Project management goes first.

The office manager takes the notes in FullyPowered AI Planner, and the page names the review day: the first Thursday of each month. The managing director tells the whole firm what is decided, that every engineer who signs off on work today still signs off on it, and that ideas go to one channel the office manager answers each Friday. A senior engineer who doubts the idea writes the first rule of the standard: every figure in a report traces to a line in the project tracker.

The head of project management sits with the senior project manager whose reports clients like best and opens last month’s reports, including one that went out with a wrong date. The capture sheet comes from those cases, and every past report is rerun after each change. When the report is in daily use, the next line is ready: project accounts receives the progress each week and invoices from it, so the invoice draft becomes its first workflow once the new system is in.

What is different in a company of under 200 people.

A company of under 200 people starts with AI by conversation, not by program, and that makes it faster. Much of the published advice is written for large enterprises, with steering committees, long assessments and a platform decision first.

  • The person who decides is in the room.The owner settles the order of work on the spot.
  • Every lead is a conversation away.Leads know the repeating work first-hand, without a survey.
  • The expert is often owner and checker.Write all three roles down, and name who covers when that person is away.
  • One announcement reaches everyone.Nobody hears a second-hand version first.
  • Attention is the limit, not budget.People with time to own a workflow run out before ideas do, which is why two or three in development matters most here.

That is where 10X the ROI of your existing AI comes from in a smaller company: the seats you already pay for, pointed at named work, one decision at a time.

Questions people ask.

For short answers across the whole subject of planning AI, Answers takes the questions leaders ask in order.

I run the company and I feel behind on AI. Where do I start?

Start with one decision: a one-page plan, written with your team leads, that names your teams, the work each repeats, the first piece each moves onto AI and who owns it. Then one team builds one workflow, tests it on real cases and puts it into daily use. You need a page, a first team and an owner.

What are Ask, Save, Share and Sustain?

They are the four stages a piece of work goes through with AI. At Ask, someone types the request from memory; at Save, the method that worked is written down; at Share, the whole team runs one version; at Sustain, one named person owns it and keeps it current. Starting means moving one piece of work up a stage.

How do I get a better answer from ChatGPT or Copilot right now?

Write four asks into the request: do the whole routine part, check the result against a standard you name, set out the options for the decision you make, and add the view of a role you do not employ. A request that carries them comes back finished and checked, not as a first draft to rewrite.

How do we decide which work to give AI first?

Walk one team’s week with its lead and mark each repeating piece of work with the ways AI helps: gather, check, draft, think it through and repeat. Weekly work that touches three or more, starts from records someone can export and already has one person checking it is where a first workflow comes from.

What work should we keep away from AI?

Leave work that happens a few times a year, work whose value is one person’s judgment in the moment, and work where a mistake would leave the building unseen off a first plan. The assistant can still help with it one request at a time. In every workflow, decisions about people and commitments on the company’s behalf stay with a person.

Which team should go first with AI?

The team whose week repeats most, whose work starts from records someone can export, where one person already reviews the output, and whose lead agrees to own the first workflow. The keenest team and the team with the biggest problem are often the wrong first choice, and both go on the plan with a date.

What do we tell employees about AI?

On the day the plan names a first team and workflow, the person who runs the company tells everyone what is decided, what each team is asked to do, what stays the same and where to bring ideas. What stays the same is that a person decides and a workflow prepares. Each team lead repeats the message to their own team that week.

What comes after the first AI workflow?

The workflow that receives the first one’s output, or the same piece of work in another team, chosen from the plan once the first is in daily use. The first owner sits beside the second owner through the build. Keep no more than two or three workflows in development across the company at once.