The short answer.
Companies get so little from the AI they already pay for because the assistants were switched on and left: every person was given ChatGPT, Claude, Copilot or Gemini, and nobody decided which work it should do or who answers for it. So people chat with the assistant one question at a time, which is the surface of the tool, while the work that fills the week, the work that repeats, runs the way it always did. The value sits in turning that repeating work into workflows, written once, run the same way by a whole team and kept current by one owner. That step never happens on its own, because it needs a decision about which work comes first, and in a company whose AI use is undirected nobody has been asked to make it.
Nothing in that answer is a fault of your people or of the software. Experimenting was the right way to start. The problem is what follows after years of experimenting.
Companies invest in AI two ways, and nobody manages the second.
Companies put money into AI in two ways: they build something new that AI runs, and they give their people an assistant, and only the first of those gets managed. A new AI-run function, such as a service that answers one class of customer question or a system that reads every incoming order, starts life as a project, with a budget and one person who answers for whether it works.
The assistant on every desk started life as a purchase. Seats were bought, a login went round, and the company moved on. It was approved by whoever manages the software bill, not by whoever runs the work, so it arrived with no list of work it is for, no owner and no review other than the renewal invoice.
The two kinds are not equal in reach. A new AI-run function touches the few people who build and run it. The assistant touches everyone with a seat, which in a company of under 200 people is often the larger part of what is spent on AI as well. Whether AI changes how your company works is decided in the second kind, the unmanaged one. The two kinds of AI investment, and the one nobody is managing sets out the difference in full, with a self-check.
Managing the assistant on every desk means giving it what the AI project already has: a list of the work it is for, one person who answers for it, and a regular review.
Everyone can chat, and chatting is not knowing AI.
Being able to type a question into an AI assistant and get a fluent answer feels like knowing AI, and that feeling is the reason the problem stays hidden. A loose request comes back as a well-written reply, so the person asking has every reason to believe they are using the tool well.
The chat box also hides its own depth: it opens on one empty line and a cursor, and everything past that line has to be known about before anyone can use it.
The consequence for a leader is specific. The people closest to the work cannot report a gap they cannot see. Ask a team whether it uses AI and the answer is yes, sincerely. Ask which piece of the team’s work the assistant prepares the same way every week and the answer is often none. A company that measures AI by whether people use it will keep hearing good news while nothing about the work changes. Everyone can chat. That is not the same as knowing AI walks through how a whole company stays at the surface for years.
The AI gap is the spreadsheet gap, and it is easier to close.
The gap between chatting with an AI assistant and using it well is the same gap every company has seen in spreadsheets, and the AI version is far easier to close. In Excel, one person types numbers into cells and adds up a column. Another builds a model once, with formulas and linked sheets, so next month’s report is produced by dropping in new figures.
AI assistants split the same way. The chat user opens a new conversation, pastes something in, asks, reads and copies the useful part. The workflow builder has written down once how a piece of work is done: what it starts from, the steps in order, the reference material, the shape of the result and what a person checks. Each week the builder hands in the new inputs and the assistant prepares the whole piece of work for a person to review.
Two things make the AI gap easier to close. A workflow is written in plain language, so there is no formula syntax or macro code to learn, and the person who does the work today already knows what goes in it. The comparison teaches one more lesson. Nobody closed the spreadsheet gap by turning every employee into a modeller. A few people built the models, and everyone else used them. AI works the same way: you do not need every person to become a builder, you need someone to decide which workflows get built and for which teams. Basic Excel and advanced Excel: the same gap exists with AI sets the two side by side.
What advanced use of an AI assistant actually is.
Advanced use of an AI assistant is everything past the chat box that lets the assistant prepare a whole piece of recurring work, the same way every time, from your own material. The assistants a company already pays for each offer it in their own form and under their own names, and the names change often. The kinds stay the same.
- Saved, reusable workflows.Instructions for one piece of work, written once and run again by anyone in the team who has access. Each run starts from the written method, not from a blank line.
- Projects with standing instructions.A space for one client, one account or one recurring report that holds the files and the instructions that apply every time, so nobody pastes the background in again.
- Connections to business systems.A link that lets the assistant read from the systems the company already runs, such as your accounting system or your ticket queue, and, with a person approving, write back to them.
- Scheduled runs.A piece of work the assistant prepares on a timetable, ready before anyone asks for it, such as a summary waiting at the start of each week.
- Several assistants working together.One assistant splits a large job, hands each part to another assistant with its own instructions, and brings the results back together.
Put the first four around one piece of repeating work and you have a workflow running in the account you already have. A person still reads what it prepared and decides what happens next. What advanced use of an AI assistant actually is covers each kind in more depth, and Did you know your business workflows can become something AI runs for you? gives one example workflow for each of six teams.
Every one of the five kinds needs a decision that one employee cannot make alone. A saved workflow needs someone to agree how the work is done. A project needs someone to own the reference material in it. A connection needs someone to agree what the assistant may read and change. A scheduled run needs someone to own what lands each week. That is why the advanced end sits unused in accounts that are fully paid for: the features are there, and the decisions are not.
Where does one person stand, and where does the company stand?
One person stands on one of six levels of using an AI assistant, and a company stands at one of six stages, and the two do not add up the way leaders expect. A person climbs from asking questions, to giving the assistant their own files, to standing instructions, to reusable workflows the team runs, to workflows connected to business systems and run on a schedule, to several assistants working together. The levels of using an AI assistant at work describes each level and the move to the next.
A company moves through not knowing it is possible, switched on, experimenting, planned, built and managed. The first three stages are undirected: AI is on and each person decides alone what to do with it. The last three are directed: the company decides, on purpose, which work moves onto AI and in what order. The stages of AI at work gives six yes-or-no questions that place a company.
The connection between the two ladders is the part that surprises people. A company can have three or four people on the higher levels and still sit at experimenting, because a company’s stage is set by whether work is shared and owned, not by how good its best users are. The team moves only when that method is written down as a workflow the whole team runs and one person owns it. Training lifts people a level. Only a decision about which work comes first lifts the company a stage, and the move that changes everything is the one from experimenting to planned.
Why isn’t AI saving us time yet?
AI is not saving your company time yet because it is being used on small steps inside the work, while the time in a business sits in whole pieces of work that repeat. An assistant that drafts one email in the middle of a job leaves the gathering, assembling, checking, waiting and handing on untouched. Four reasons keep everyday use from adding up.
- Tasks, not whole pieces of work.One step gets faster and the work around it does not.
- A blank prompt every time.Each person types the background, pastes the records and corrects the first answer, every time, so on small tasks the setup costs as much as the help.
- Checking nobody designed.When a reviewer does not know what the assistant was told or what it left out, the only safe check is to redo the work.
- Freed time with nowhere to go.Small savings scattered across many people disappear into the day, so nobody can point to where the time went.
Read together, the four reasons are one fact seen from four sides: the work was never named. Name one piece of repeating work, write down how it is done, design the check into it and say who runs it, and every one of the four reasons is answered at once. Why isn’t AI saving us time? gives four questions for finding where the time sits in one team’s week.
Why the gap does not close by itself.
The gap between chatting and running workflows does not close by itself because every force that normally gets a company to manage something is missing from the assistant on every desk.
- It was cheap per seat.So it was bought like any other software, by the person who pays the bills, and never handed to the person who runs the work.
- It belongs to everyone.A tool every team uses has no natural owner, and each team lead assumes it is someone else’s job.
- The chat box looks finished.Nothing on screen suggests there is more to set up.
- Good practice is invisible.When one person builds something better than chatting, it lives on that person’s account and leaves when they do.
- Training stopped at the question.People were taught to ask better questions, which is the basic spreadsheet of AI: useful, and a ceiling.
- Nothing breaks.A managed project fails loudly. An unmanaged assistant fails quietly: the work stays as it was and the invoice keeps arriving.
Every item on that list has the same root. Moving work onto AI is a decision about the team’s work, and no single employee is in a position to make it for the team. Using AI is something each person can do alone. Deploying it, so a team runs one method and someone keeps it right, is something only the company can do. What is the difference between using AI and deploying it? draws that line from the team’s side. Until someone is asked to make that decision, the gap stays where it is.
How to see the problem in your own company.
You see the problem in your own company by looking in three places: what you pay for, what each team uses the assistant for, and where each piece of repeating work sits today. The first is a stocktake in three columns: what the company pays for, who has it, and which named piece of work each seat is used for. The third column, filled in by asking each team, is the one that tells you what you are getting. How to take stock of the AI your company already pays for has the full sheet.
The second and third looks are the two sheets below. Copy them as they are.
Sheet one: ten questions for each team lead. Ask them in a conversation, not a survey. Write the answers down in the lead’s words, and never rate the person.
- What does your team do the same way every week or month?Answer: ____________
- Which of that work does the assistant prepare, start to finish?Answer: ____________
- Which assistant does each person use, and on which account?Answer: ____________
- Where are the instructions for that work written down?Answer: ____________
- Does the assistant start from your team’s files, or from a blank line?Answer: ____________
- Does the assistant read from any system you run, or does every input arrive by copy and paste?Answer: ____________
- Does anything get prepared on a timetable, without someone starting it?Answer: ____________
- Who on your team uses AI best, and does anyone else use what that person built?Answer: ____________
- Who checks what the assistant prepares, by name?Answer: ____________
- Who decides which work moves onto AI next?Answer: ____________
Sheet two: the problem map, one page per team. One line per piece of repeating work. For the stage, use the four stages a piece of work goes through: asked (typed from scratch each time), saved (written down once), shared (the whole team runs one version) and sustained (one owner keeps it current). Ask, Save, Share, Sustain explains each stage and the move to the next.
- Team and lead.____________
- Piece of work.____________ Repeats: weekly / monthly. Starts from: ____________
- Stage today.Asked / saved / shared / sustained.
- Where the method lives.One person’s chat history / a saved set of instructions / a shared document / nowhere.
- Who checks the result.____________
- Who owns the method.____________ (write “nobody” where nobody does)
- Next move.____________
Filled in for every team, the two sheets give the problem a shape: blank owners, methods that live nowhere and work stuck at asked.
Illustrative example.
Take a 70-person marketing agency with five teams: account management, creative, media, finance and new business. The agency pays for seats on one assistant for every employee, and a few people pay for a second one themselves. The owner asks each team lead the ten questions.
Account management says it uses the assistant for client emails. One account director has built a project holding a large client’s brand guidelines and past campaign reports, and uses it every week; nobody else in the team knows it exists. The media team writes a weekly performance report for every client from the advertising platforms’ exports, and each media buyer pastes the exports into the assistant in a different way and gets a different report. Creative brainstorms headlines with it, which is ad hoc work and fine as it is. Finance tried it once on a supplier statement and stopped. New business drafts proposals from scratch every time.
On the problem map, every piece of repeating work in the agency sits at asked or saved. Nothing is shared and nothing has an owner. The weekly client report stands out: it repeats every week for every client, it starts from exports someone already pulls, and the head of media already reads each one before it goes out. The agency is at experimenting, with one strong find on one desk and a clear first candidate for a shared workflow. The assistant it needs is already paid for.
What is different in a company of under 200 people.
A company of under 200 people has the same problem as a large one and a far shorter route to seeing it and closing it. Published advice is written for large enterprises, with adoption programs and usage dashboards. A smaller company needs neither.
- The assistant is the main AI investment.A smaller company is less likely to be building a new AI-run function, so the unmanaged kind is where its AI money goes. Managing it is the whole job, not a side task.
- You can see the whole company.You can name every team from memory and ask every lead the ten questions in one week.
- The best find is visible once you ask.In a team of five, one person’s well-built method stands out the moment anyone asks about it, and it can become the team’s first shared workflow straight away.
- The owner is an operator.The person who answers for the assistants is the head of operations, the chief of staff or the owner, not an IT department. IT manages the seats; the owner decides what they are for.
- You stop nothing.The scattered use already in the company is your best evidence of where AI works. Ask about it, write it down and build on it.
Our AI use is all over the place. What do we do first? sets out the first week in a smaller company: take stock team by team, make each team’s best find its first shared workflow, and name one owner for the page.
What closes the gap: plan it, develop the workflows, deploy them.
The gap between what a company pays for and what it gets from AI closes in three steps: plan it, develop the workflows, deploy them.
- Plan it.Decide, team by team, which repeating work the assistant should prepare first, and name one owner for each piece. Planning is the decision nobody was asked to make, and it moves a company from experimenting to planned.
- Develop the workflows.Write down how your best person does each piece of work, precisely enough that the assistant prepares it the same way every time, and test it on last month’s real cases. Free workflow templates in the Workflows Directory are samples and a starting point.
- Deploy them.Put each workflow in the hands of the whole team, in the company’s own assistant, with a person deciding at every step, and check whether it is installed, used and useful. Installed, used or useful? has the sheet for reading that.
That is how a company starts to 10X the ROI of your existing AI: the same seats, pointed at named work, with one person answering for each piece. How do we get more value from the AI we already pay for? follows one team from a loose prompt to instructions all three of its people share. How organizations plan their use of AI sets out the three steps in full, with a one-page plan to copy, and Answers gives short answers to the questions leaders ask along the way.
FullyPowered AI is a better way to do all three. The plan lives in FullyPowered AI Planner: you set out your teams, their roles and the work between them, choose what to build first, and your colleagues add their own teams. We develop the workflows with you, and they run in the ChatGPT, Claude, Copilot or Gemini your company already pays for, run by your own people.
Questions people ask.
- We pay for ChatGPT for everyone. Why does nothing feel different?
Nothing feels different because the assistant was switched on without a list of work it is for or a person who answers for it. People use it one question at a time on small steps, while the work that repeats every week runs as it always did. The change comes when a company names that repeating work and turns it into shared workflows.
- Everyone on my team uses AI every day. Doesn’t that mean we are fine?
Daily use shows that people chat with the assistant, not that it prepares any of the team’s work. Ask which piece of work the assistant prepares the same way every week; if the answer is none, the team is at the surface. Chatting is the right way to start, and it is the beginning of what the assistant does.
- Is the answer more AI training?
Training lifts individual people, and the problem is a decision about the team’s work that no individual can make alone. A company moves forward when someone decides which repeating work becomes a shared workflow, who owns it and who checks it. Training pays off when it is built around those workflows.
- What does advanced use of an AI assistant mean in plain terms?
Advanced use means the assistant prepares a whole piece of recurring work from your own material, not a single reply. It rests on saved reusable workflows, projects with standing instructions, connections to the systems you run, runs on a schedule and, for large jobs, several assistants working together. The assistants a company already pays for offer these in their own form.
- Why is closing the AI gap easier than learning advanced Excel?
An advanced spreadsheet needs formulas and macro code, while a workflow is written in plain language: what the work starts from, the steps, the checks and what good looks like. The person who does the work today already knows all of that. And as with spreadsheet models, a few people build workflows and the whole team runs them.
- Do we need to buy a different AI tool?
Start with the assistant you already pay for, because the gap is in what it is pointed at, not in the tool. ChatGPT, Claude, Copilot and Gemini each hold saved workflows, standing instructions and connections in their paid versions. Decide the work first, and any later purchase has a place to land.
- How do I find out where my company stands with AI?
Make a three-column stocktake of what you pay for, who has it and which named work each seat is used for. Then ask each team lead what the team repeats every week and whether the assistant prepares any of it. Blank owners and methods that live in one person’s chat history show where the gap is.
- Should we cancel the AI seats nobody uses?
Ask what the role’s repeating work is before you cancel an unused seat. A role with work that repeats from records someone can export, with a person who checks it, has a use for the seat that nobody has pointed it at. A role without such work does not need the seat.
- Who should own our AI assistants in a company our size?
One person, by name, who runs the work rather than the software: in a company of under 200 people, usually the head of operations, the chief of staff or the owner. IT manages the seats and access. The owner decides what the seats are for and looks each month at whether that work is being done.