The short answer.
AI is not saving your company time because it is being used on the small parts of the job. An assistant saves a person a little on a draft or a summary, while the time in a business sits in the work that repeats every week, and that work moves onto AI only when someone decides it should, writes down how it is done and gives it an owner.
You pay for ChatGPT, Claude, Copilot or Gemini, and people use it. Four things keep that use from adding up: it is used for tasks, not whole pieces of work; every person starts from a blank prompt; the checking was never designed; and the time that does come back has no destination. Each has a fix, and all four fixes are one move: point the AI you already pay for at named, repeating work, on purpose.
AI is used for tasks, not for whole pieces of work.
Most AI use in a company is task use: one person asks an assistant for one step, such as a first draft, a summary or a rewrite, and then carries on with the rest of the work by hand. The step gets faster; the work around it does not.
A piece of work is everything from the moment a request arrives to the moment the finished result leaves the team. Most of its time is in finding the records, putting them together, checking them, waiting for someone else and passing the result on. An assistant that drafts the email in the middle of that chain leaves every other link untouched.
The time moves when the whole piece of work moves: the workflow takes in the records, prepares the full output in your order, marks what is missing and hands a person something ready to decide on.
Every person starts from a blank prompt.
When AI use is undirected, every person starts each job from an empty box. They type the background, paste in the records, explain what good looks like and correct the first answer. That setup is real work, and it is repeated by every person, every time.
On a small task the setup costs more than the assistant saves, and people stop bothering. The same setup written once, as a workflow the whole team runs, is paid for one time. How do we get more value from the AI we already pay for? shows that move from a loose prompt to shared instructions on one team’s queue.
Does checking AI’s work cancel out what it saves?
Checking AI’s work cancels out the saving when nobody designed the check. If the reviewer does not know what the assistant was told, what it looked at or what it left out, the only safe check is to do the work again, and then nothing was saved.
A designed check is different. The workflow’s instructions say what it must use, what it must never do and what it hands to a person. The output shows where each fact came from and marks the gaps instead of filling them. The reviewer then checks a short, known list: the flagged items, the figures against their source, the decision that is theirs. A person still decides, without redoing the preparation.
Measuring whether that check is quick enough is part of How to tell whether AI is paying back, which counts the checking in full.
The time that comes back has no destination.
Time that AI frees in small, scattered pieces across many people disappears into the day. A little back on one email and a little on one summary never adds up to a block anyone can point to, so nobody can say where it went.
Time becomes visible when it comes back in one place: one piece of work, for one role, prepared every time it comes round. That returns a recognisable block to a named person, and a named block can be given a job before it arrives. What will your team do with the capacity it gains? has the worksheet for giving it one.
What to do about each reason.
Each reason has one fix, done in this order.
- For task use: choose whole pieces of work.List the work each team repeats every week or month, from request to finished result. Repeatable work and ad hoc work shows how to sort a team’s week.
- For the blank prompt: write it down once.Capture how your best person does that piece of work, as instructions an assistant follows the same way every time, with the records it starts from. The Workflows Directory holds free workflow templates to start from; they are samples, made to be changed.
- For undesigned checking: design the check with the workflow.Name the reviewer, by name. Write down what they check and what the workflow must flag for them. Test it on last month’s cases, where you already know the right answer.
- For time with no destination: give it a job first.Before the workflow goes into daily use, write one line saying what the person who runs this work does with the time it frees.
Every fix needs an owner: one person who keeps the workflow current and decides when it changes.
Four questions to find where the time is in one team’s week.
Ask these of one team, with its lead and the person who does the work best. The answers name the first piece of work to move onto AI.
- What does this team do the same way every week or month?List each piece of work from the request arriving to the result leaving the team.
- Where in that work does the time go?For each piece: the gathering, the assembling, the checking, the waiting or the handoff.
- Who already checks the result, and what do they look for?One name per piece of work, and the three or four things they always check.
- If this work arrived prepared, what would that person do next?One line, naming the work the time goes to.
Illustrative example.
Take a 45-person insurance brokerage. Its account managers use an assistant to tidy renewal emails, and the owner cannot see where any time went. The operations lead asks the four questions of the commercial renewals team. The work that repeats is the renewal review for each client before the policy comes due. The time goes on gathering: last year’s policy, the claims history, the client’s changes and the insurers’ new terms, assembled by hand into a comparison. The senior account manager checks every comparison for missing exclusions and changed limits. With the review prepared, she would call clients before their renewal date instead of after. The brokerage chooses the renewal review as its first workflow, names her as its owner and the person who decides what goes to the client, and writes her check into the instructions as the items the workflow must flag.
What is different in a company of under 200 people.
In a company of under 200 people, the time is easier to find, because the people who do the work and the people who decide are a conversation apart. Most published advice on this is written for large enterprises, with usage dashboards and adoption programs. You need neither.
- You can ask the four questions this week.Every team lead is reachable, and so is the person who does each piece of work best.
- One workflow is visible.In a team of four, one piece of work prepared every week changes how the whole team’s week runs, and everyone sees it.
That is how a small company starts to 10X the ROI of your existing AI: the seats you already pay for, pointed at the work that repeats.
Start with one team’s week.
Finding where the time is comes first in any plan for AI, and the plan decides which work moves onto AI, in what order and who owns it. We have a plan. How do we build the first AI workflow? takes the next step. How organizations plan their use of AI sets out the whole method in three steps: plan it, develop the workflows, deploy them. 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.