The short answer.
Keeping AI use undirected costs a company five things, and none of them arrives as a line on a bill: seats that buy no work you can name, the same setup repeated by every person, rework where one person’s output meets another’s, methods that leave with the person who found them, and decisions nobody made. Between teams it costs more, in handoffs that stop matching, customers who hear two answers and checks each person assumed someone else did. For the owner and the team it costs a steady confusion about what is allowed, what is working and who answers when something goes wrong. A one-page plan that names the work, the team and the owner ends each of those costs.
The articles behind this guide each count one part of that cost. Read together, they establish something none of them says on its own: every cost of undirected AI is an empty space where a name should be. An idle seat is a role with no named work. A prompt rebuilt five times is a method with no named home. Rework is a standard nobody named. Lost know-how is a workflow with no named owner. A skipped check is a piece of work with no named checker. The unease in a team is a set of rules nobody wrote under anyone’s name. So the cost of leaving AI undirected is not a pile of separate problems. It is a short list of blanks, and filling them in is the plan.
Writing that plan is the first of three steps set out in How organizations plan their use of AI: plan it, develop the workflows, deploy them.
What happens when a company uses AI without a plan?
When a company uses AI without a plan, each person works AI out alone, settles on asking, drafting and summarizing, and the repeat work in every team stays as manual as it was. A few people open ChatGPT, Claude, Copilot or Gemini every day, others now and then. Ask which work AI prepares each week, in which team, and the honest answer is “some people use it for some things.”
Experimenting was the right way to start. People learned how to ask and where an assistant slips, and the people who explored are the ones a company needs next.
Exploring stops short for a plain reason: the decisions that move whole pieces of work onto AI belong to the company, not to any one person. The line between the two kinds of decision is where every cost in this guide begins.
- What one person can decide alone.How to phrase a question, which draft to keep, whether to use the assistant on today’s email.
- What only the company can decide.Which piece of a team’s repeat work moves onto AI, what good output looks like for it, who checks it before it leaves, who owns it when something changes, and what information may go into the assistant at all.
Left undecided, the second list does not stay empty. Each person fills it in privately, differently, and without telling anyone. What happens when a company uses AI without a plan? sets out the full pattern, and Is it OK to let employees figure out AI themselves? separates the freedom worth keeping from the direction a company owes its people.
What does undirected AI cost, kind by kind?
Undirected AI costs a company in five kinds, and the subscription, the only one with a number on it, is the smallest. The other four are spread thin across every person and every week, in pieces too small for anyone to add up, which is exactly why they last.
- Seats that buy no named work.The company pays for an assistant for everyone, and nobody has said what the seat is for in each role. The renewal arrives with no answer to what it bought.
- The same setup, repeated by each person.Five people each work out how to prepare the weekly report, typing the background and correcting the first answer. The company pays for the same setup five times, every time.
- Rework where outputs meet.Two people prepare the same kind of document with different prompts, and the two versions use different figures or leave out different things. Someone downstream notices and the work is done twice.
- Methods that leave with people.The best method in a team sits in one person’s chat history. When that person changes role, goes on leave or moves on, the team starts again from a blank prompt.
- Decisions nobody made.Which team goes first, which work moves onto AI, who checks the output, which account company information belongs in. Each open decision is settled by whoever happens to act, or left open.
The fifth cost is the one that holds the other four in place. A seat stays idle because nobody decided what it is for; a method stays private because nobody decided it belongs to the team. Close the decisions and the other costs lose their footing. What does it cost a company to use AI without a plan? takes each cost in turn, with a cost check to run with each team lead.
The fourth cost arrives without warning: a method that lives with one person is safe only until the day that person is away. When the know-how sits with one person sets out the questions to ask.
Why companies get so little from the AI they already pay for follows the first cost, the seats, to the work they could be doing.
The complications show up between teams.
When each person uses AI their own way, the trouble shows up between teams rather than inside them: handoffs stop matching, customers hear two answers to one question, and checks get skipped because each person assumed another person checked. A private method works well at its own desk; the trouble starts when its output reaches a team that never saw the prompt.
- Handoffs that no longer match.A team starts preparing its weekly summary with AI, the summary changes shape, and the one figure the next team relied on is folded into a paragraph.
- Two answers to one customer.One team asks its assistant from an old price list, another from the current order system, and the customer hears both, each sent in good faith.
- Checks that fall between two people.One person passes on a draft assuming the next person reviews it; the next person assumes it was reviewed already. The work goes out unread.
The connecting point is where the cost lands. Effort AI saves in one team is spent again in the next, by people who no longer trust what arrives and check it all themselves. So every team can report that AI helps while the business feels harder to run, because the work between teams has no owner. What goes wrong when everyone uses AI their own way? includes a handoff sheet that gives each handoff one format, one source for the facts and one named checker.
Why does AI feel chaotic, for the owner and for the team?
AI feels chaotic in a company because nobody has said what it is for there, so every person decides alone, every day, and the owner and the team each carry a different part of that open question. The word describes the experience, never the people in it.
The feeling sits differently in each chair.
- For the owner.A worry nobody else can see. You approved the spend and cannot say what it does for the business, so every renewal lands on the same unanswered question.
- For the team leads.One person races ahead, another never touches it, and the lead has no basis for saying what the team should do. Every answer feels like a personal opinion.
- For the people doing the work.Three questions stay open for each person: am I allowed to use it for this, is my way the right way, and what does it mean for my role. Nobody has answered them, so each person carries them alone.
That unease has a cost of its own. Good finds stay hidden in case they are not allowed. People draft with AI and then redo the draft by hand, unsure it will pass. The people most careful about quality hold back longest, so a team’s best judgment is the last thing to reach its AI use.
The useful thing about the feeling is that it is specific. Each part of it points at one decision nobody has made: what AI is for in each team, which shared method a team runs, what the rules are, what each role still decides, and who answers for the whole. Why does AI feel chaotic at our company? turns those five into sentences a leadership team can fill in and share, and Why is my team uneasy about AI, and what helps? sets out the three answers to write down for each team.
Who answers when AI gets something wrong?
In a company without an AI plan, nobody answers when AI gets something wrong, because nobody was ever told the work was theirs to stand behind. In a company with a plan, every workflow carries two names before it runs: an owner who keeps the workflow right, and a checker who decides on each piece of work before it leaves the team. A person decides; a workflow prepares.
Without those names, a mistake lands on whoever typed the prompt. A wrong figure reaches a client, and the person who prepared it feels accused of a job they were never given. Nobody was careless. Whose work it was, and who should have checked it, get decided for the first time in the middle of a mistake, defensively.
The after-effects cost more than the error. The person who was blamed stops using AI, or stops saying so. The method that produced the error still lives in someone’s chat history, so it runs again. And one bad afternoon becomes the story everyone tells about AI, which keeps the repeat work manual for another year. Of all the costs in this guide, the unowned mistake is the one that can stall the rest of a company’s AI use at once.
With an owner and a checker named in advance, the same mistake becomes a correction: the checker fixes the piece of work, the miss goes in a log, the owner fixes the workflow, and the checker’s list gets one more line. Nobody asks who typed the prompt. Who answers when AI gets something wrong? includes a sheet to fill in before the first mistake.
How do we know our AI use is undirected?
You know AI use is undirected by what you hear when you ask about the work: the same task described three ways in one team, a best method that lives in one person’s chats, new hires told to play around, and leads who answer with a list of tools instead of a named piece of work. Each sign shows up in one question to the right person.
The signs fall into three groups, and each group points at a different first move. Signs about method say the work has no shared workflow, signs about rules say the ground rules are unwritten, and signs about direction say nobody owns the plan. How do I know if our AI use is undirected? lists ten signs to check and what each count means.
Questions to ask each team lead, to copy. Ask them in person, one lead at a time, and write the answers down as given. Record the work, never rate the person.
- Which piece of your team’s repeat work does AI prepare every week?______
- When two people on your team do that work with AI, do they use the same written instructions?______
- Where is your team’s best AI method kept, and whose is it?______
- Who checks work prepared with AI before it leaves your team?One name, or “nobody yet”: ______
- Which work arrives from another team already prepared with AI, and does it still match what you need?______
- What AI question has your team been waiting for someone to answer?______
- If AI got something wrong on your team tomorrow, who would fix the method?______
Every blank you cannot fill is one of the empty names from the short answer. Note the team beside each one; the team matters as much as the total.
Not having a plan is still a plan.
A company with no AI plan still runs on one: the default plan, under which each person works AI out alone, and that default makes the same decisions a written plan would, without anyone signing them. It decides which work moves onto AI, who owns a method that works, what good output looks like, who checks the work, what is allowed and what the seats are for, and in each case the answer is whoever happens to act.
Nobody chose the default. It arrived with the seats, and for the first stretch of experimenting it was a reasonable arrangement. After years of experimenting, though, AI is a real line in the budget, and it deserves the discipline and focus any spend of that size gets.
Seen this way, every cost in this guide is the default plan reporting back. The idle seats, the rebuilt prompts, the mismatched handoffs, the unease and the unowned mistake are its results. Replacing it stops nothing: people keep experimenting, inside a plan. Not having an AI plan is still a plan sets out the six decisions the default makes and how to take each one back.
A cost ledger to copy. One page, one row per kind of cost, filled in from the answers your team leads gave. In each row write four things: where you see the cost today, who carries it, the decision that would end it, and who makes that decision, by when. Write “Unknown” where nobody knows, with the name of the person who will find out.
- Seats that buy no named work.______
- The same setup repeated by each person.______
- Rework where outputs meet.______
- Methods that would leave with a person.______
- Handoffs between teams that no longer match.______
- Work that leaves with nobody named to check it.______
- Questions people carry alone.______
- Decisions nobody made.______
Write the name of the person who owns the ledger at the top, and the date you will look at it again. A ledger with a name at the end of every row is the first draft of a plan.
One firm counts what it is leaving undirected.
Illustrative example.
Take a 70-person civil engineering firm with four teams: design, surveying, project management and the business office, which covers finance and administration. Everyone has an assistant on the firm’s account, and the managing principal cannot say what it does for the firm. She asks the operations manager to put the questions to each team lead and fill in the ledger.
In project management, each project manager writes client progress reports with their own prompts, so clients receive reports in several shapes, and the principal on each job rewrites them. Setup repeated, and rework. In surveying, one senior surveyor has a careful method for turning field notes into the site conditions summary the design team works from. Nobody else knows it, and the surveyor is due to lead a second office. A method about to leave with a person. In design, several engineers have quietly stopped using the assistant after a poor answer, and nobody has said what AI is for in a design role. Unease, and seats with no named work. In the business office, nobody has decided whether project budgets may go into an assistant at all, so the finance lead avoids it entirely. A decision nobody made.
The principal makes the decisions in one meeting with the leads. Project management goes first, with one shared workflow for the client progress report, owned by the senior project manager, and the principal on each job checking every report before it goes to the client. The surveyor writes the site conditions method down and owns it as surveying’s first workflow. The principal answers the budget question in writing for the whole firm. Design’s role line says plainly that AI prepares and engineers decide, and design waits on the plan with a date. The operations manager owns the page and reviews it with the four leads on the first Monday of each month.
What is different in a company of under 200 people.
In a company of under 200 people, each cost of undirected AI lands on fewer shoulders, so each one weighs more, and each one is quicker to end. Much of the published advice on AI is written for large enterprises, with steering committees, usage dashboards and governance programs, and a company your size needs none of them to stop paying these costs.
- One departure can empty a team’s know-how.When the one person with the method leaves a team of four, the team loses it entirely. A large company has colleagues to rebuild it; a smaller one has a blank prompt.
- The owner sets the tone.When the owner says what AI is for, the unease in a team lifts quickly. Silence from the top is what keeps it.
- The evidence is a conversation.You can ask every team lead the questions above yourself and fill in the ledger from their answers, without a survey or an assessment.
What changes in the first month with a plan.
In the first month with an AI plan, very little work has moved onto AI yet, and the questions that kept coming back have written answers: the owner knows what is being built and by whom, each team lead owns one named piece of work, and everyone knows what is allowed and where their ideas go. How the work feels has changed, from confusing to clear.
- Week one.The leader and the team leads decide which teams go first, what each builds first and who owns it.
- Week two.Each team lead knows one thing the team is building, and the leader tells everyone what is decided, what stays the same and where ideas go.
- Week three.The first workflow is being built from how the work is done today, and the team sees a real draft of its own work.
- Week four.The owner of the plan holds the first review. The question stops being “are we behind?” and becomes “is the first workflow ready, and what comes second?”
What changes in the first month after a company gets an AI plan? follows those four weeks in detail. The first move is to stop nothing: each team’s best find becomes its first shared workflow. Our AI use is all over the place. What do we do first? sets out that first week. A company whose first push faded starts the same way, from the uses that survived it; We tried AI and it did not stick. What now? covers the restart. Why plan how your company uses AI? sets out what is different the week after the page exists, and once the plan is in place, How to roll out AI across a company, and tell whether it is working takes it team by team.
The plan that ends these costs is the ledger with every name filled in, kept current. FullyPowered AI Planner is where that 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. That is where 10X the ROI of your existing AI comes from: the same seats, directed at named work, on purpose.
Questions people ask.
For short answers across the whole subject of planning AI, Answers takes the questions leaders ask in order.
- What does it actually cost us to let everyone use AI their own way?
It costs five things that never arrive as one line on a bill: seats that buy no named work, the same setup repeated by each person, rework where two people’s output differs, methods that leave with the person who found them, and decisions nobody made. Between teams it also costs mismatched handoffs and skipped checks. A one-page plan that names the work, the team and the owner ends each one.
- Is it OK to let employees figure out AI themselves?
Yes for learning, and no as the whole approach. Exploring is how people learn what an assistant does, so keep that freedom. The company still owes each role three things no individual can supply: the piece of work to hand over, a written standard for it, and a named person who checks it.
- Why does AI feel so chaotic at our company?
AI feels chaotic because nobody has said what it is for in the company, so each person decides alone, every day. The feeling describes the experience, not the people, and it points at specific decisions nobody has made: what AI is for in each team, which shared method a team runs, what the rules are, what each role decides and who answers for the whole. Each decision settles one part of the feeling.
- Who is responsible when AI gets something wrong?
Without a plan, nobody is, because nobody was told the work was theirs to stand behind. With a plan, every workflow has an owner who keeps it right and a checker who decides on each piece of work before it leaves the team. Name both before the first mistake, and the first mistake becomes a correction rather than a crisis.
- How do I know if our AI use is undirected?
Ask each team lead about the work. If you hear the same task done several ways, a best method kept in one person’s chats, new hires told to play around, or a list of tools where a named piece of work should be, the use is undirected. That describes how AI is used, never the people using it.
- We don’t have an AI plan. Isn’t that just keeping our options open?
No. Without a written plan, a default plan runs instead, under which each person works AI out alone, and it decides which work moves onto AI, who owns a good method and who checks the output. Write down what the default decides today, then make each of those decisions on purpose, with one owner for the page.
- We tried AI and it faded. Is it worth trying again?
Yes, as a plan rather than another round of licenses or training. A first push fades when no work was named, nobody owned it and nothing was built for the next person to reuse. Start from the uses that survived, turn the best one into one team’s shared workflow with a named owner, and set a date to review it.
- What changes in the first month after we write an AI plan?
Guessing turns into knowing. The leader can say what is being built and by whom, each team lead owns one named piece of work, and people know what is allowed and where ideas go. Little work has moved onto AI by the end of the month, and the first workflow is being tested on real cases.