The short answer.

When a company uses AI without a plan, its AI use is undirected. Each person works AI out alone, most settle on using it as an assistant to ask questions, draft and summarize, and the larger gain never arrives: handing whole pieces of repeat work to AI, team by team and role by role, because nobody planned it. The company pays for the seats and cannot say what they bought.

The missing piece is a method. A plan supplies it: one page that says which work, in which team and role, moves onto AI first, and who owns each piece. The workflows are then developed from the plan and deployed.

What undirected AI use looks like.

Undirected AI use looks like a company where the tools are switched on, the bill is paid, and each person is left to work out AI alone. In a company of under 200 people you will recognise most of these.

  • Everyone tries things on their own.Each person keeps a few prompts that work for them. Nobody is told what AI is for in their role.
  • Seats are bought and half used.A few people open ChatGPT, Claude, Copilot or Gemini every day. Most open it now and then. Some have not opened it since the first week.
  • A good find never leaves one person’s desk.Someone in finance works out how to prepare the month-end variance notes with AI. The rest of finance does not know, and the method leaves with that person.
  • Five people solve the same task five ways.Five account managers prepare client updates with five different prompts, and the updates read five different ways.
  • Nobody can say what the company is getting.Ask which work AI prepares each week, in which team, and the honest answer is “some people use it for some things.”

Nobody chose any of this. Undirected adoption is what happens when the tools arrive before the method.

Why AI use stops at the assistant.

Without a plan, AI use stops at the assistant because asking, drafting and summarizing is as far as one person can go alone. Handing over a whole piece of repeat work needs decisions no single person can make: which work goes first, what good output looks like, what it starts from, who checks it and who owns it when something changes. Those are team decisions, and in undirected adoption nobody makes them.

That is why AI is not saving most companies much time. An assistant saves a person minutes on a draft. The hours sit in the work that repeats every week: the reconciliation, the client brief, the weekly report, the order exceptions. That work moves onto AI only when someone decides it should, writes down how it is done and gives it an owner. How do we get more value from the AI we already pay for? shows that move on real work.

There is an earlier reason too, and it comes before any of this: most people do not know it is possible. They use an assistant every day and have never seen the place in their own account where reusable instructions are kept. Did you know your business workflows can become something AI runs for you? shows where to look. The stages of AI at work sets out the whole path from there.

Experimenting was the right start.

Experimenting with AI was the right way for a company to start, and experimenting stops paying once it is all there is. Trying the assistants taught everyone what AI does well and where it slips.

After years of experimenting, the situation is different. AI is now a real line in the budget, and it deserves the discipline and focus any spend of that size gets. So, is it fine to let employees figure out AI themselves? For learning what an assistant does, yes. For the return on what the company pays, no: individual trial and error does not decide which team’s repeat work moves onto AI, so that work stays where it is.

No plan is still a plan.

Leaving AI to each person is a decision, and that decision has a cost. The seats renew every year. The good finds stay with the people who made them. Each task keeps being done five ways. The repeat work in every team stays as manual as it was.

Do you need an AI strategy, or can you just experiment? You need a plan, and you keep experimenting inside it. A directed company still tries new things; it tries them on purpose, on work someone chose, and it keeps what works. Installed, used or useful? gives three questions that show where your company stands today.

What directed AI use looks like.

Directed AI use fits on one page. The page names five things.

  1. Your teams.The three to six teams the plan covers first, and the lead of each.
  2. The roles in each team.Named for the work, such as accounts payable or intake, not for a person.
  3. The work that repeats.Five pieces of work each team does the same way every week or month.
  4. What moves onto AI first.One piece of work per team, what it starts from and what it produces.
  5. Who owns it.One person, by name, for each workflow, and one person who owns the whole page.

From that page, each workflow is developed: your best way of doing the work, written precisely enough that an AI assistant prepares it the same way every time, and tested on real cases. Then it is deployed in your company’s own ChatGPT, Claude, Copilot or Gemini, run by your people. A person decides; the workflow prepares. A plan before agents shows a filled-in page.

Illustrative example.

Take a 60-person distributor where everyone has an assistant and each person uses it differently. The operations director writes one page. Four teams go on it: sales, customer service, purchasing and finance. Customer service starts with a prepared reply to each order exception, owned by the customer service lead, who approves every reply before it goes out. Purchasing starts with a weekly reorder brief built from the inventory export, owned by the purchasing manager. Finance starts with the month-end variance notes, owned by the controller, and the person in finance who had already worked out a method with AI writes it down for the whole team. The operations director reviews the page on the first Monday of each month.

Ten questions: directed or undirected?

Answer each question yes or no about your company today. Each no marks a place where AI use is undirected.

  1. Can you name the three pieces of work AI prepares for your finance team each week?
  2. Is there one page that lists your teams and what each should be doing with AI?
  3. Does every team have at least one named piece of repeat work assigned to AI?
  4. Does each of those pieces of work have one owner, by name?
  5. When two people do the same task with AI, do they use the same written instructions?
  6. When someone finds a better way to do a task with AI, is there a place it is written down for the team?
  7. Can you say how many of your paid seats were used in the last month?
  8. Has someone decided which piece of work moves onto AI next?
  9. Does each AI-prepared piece of work have a named person who checks it before it goes out?
  10. Did someone review your AI plan in the last 30 days?

What is different in a company of under 200 people.

In a company of under 200 people, moving from undirected to directed AI use takes a week, not a program. Most published advice is written for large enterprises, with steering committees and platform teams.

  • You can name every team from memory.The plan fits on one page, and the people who fill it in are in the same building.
  • The owner is an operator.The head of operations, the chief of staff or the owner holds the page. IT helps with accounts and access.
  • You start with the seats you already bought.No platform decision comes first. The plan decides what those seats are for.

Where to start.

Start with the plan. How organizations plan their use of AI walks through the three steps: plan it, develop the workflows, deploy them. FullyPowered AI is a better way to do all three, and this is how a company gets 10X the ROI of your existing AI: the same seats, directed at the work that repeats.

FullyPowered Planner is where the plan lives. You set out your teams, their roles and the work between them, and choose what to build first. Your colleagues add their own teams.