How Long Does AI Training Take? Two Clocks, Not One

AI training time is two separate measurements: time-to-first-competence, the point at which somebody can complete a real task with an AI tool without help, and time-to-habit-change, the point at which they reach for it without being asked.

For most office roles, first competence takes five to ten hours of structured practice spread across two to four weeks. Habit change takes considerably longer, with the best behavioural evidence putting the median at 66 days of near-daily repetition. Budget against one clock and report against the other, and the numbers will never reconcile.

Nearly every published answer to this question gives a single figure, because nearly every published answer is written for an individual learning AI as a career move. The person signing off a programme for 80 people is asking something else: how many weeks until this changes how the work gets done, and what am I paying for in the gap. Here are both numbers, where they come from, and how far they move by role.

Why does AI training have two timelines?

AI training has two timelines because learning a tool and changing a working habit are different processes running at different speeds. Competence is a skills problem and it resolves in hours. Habit is a repetition problem and it resolves in weeks.

The distinction decides what you buy. A two-day intensive on Microsoft Copilot buys competence for the people in the room. It buys nothing on the second clock, because the second clock does not start until those people are back at their own desks, doing their own work, with the tool open.

Research from the London School of Economics' Inclusion Initiative and Protiviti, published in October 2025 and based on nearly 3,000 workers and 240 executives surveyed globally, found that 93% of employees who had received AI training used AI in their job against 57% of those who had not, and trained employees saved 11 hours a week against five for the untrained. That difference is the second clock, running.

How long does it take to become competent with an AI tool?

Five to ten hours of structured practice, spread across two to four weeks, is enough for most office workers to complete a real task with ChatGPT, Claude or Copilot unaided. That is the honest range for the first clock, and it is shorter than most people expect.

It feels longer because most people never reach it. Slack's Workforce Index, a survey of 17,372 desk workers across 15 countries conducted in August 2024, found that 61% had spent under five hours in total learning how to use AI, and 30% had had no training at all. The workforce is not slow. It is untrained.

Two things move this range. Format is one, and live cohorts compared with self-paced courses close the gap faster because somebody is there at the moment the tool does something strange. Task specificity is the other. A finance analyst learning to draft variance commentary reaches competence faster than the same analyst learning "AI" in the abstract, because there is one output to be right or wrong about.

Competence here means something narrow, and it is worth stating out loud: the person can do one named task, to the standard their manager would accept, without asking for help. It does not mean they know what the tool is bad at. That takes longer, and it is the more expensive thing to skip.

How long before AI use becomes a habit?

Habit change runs on a clock measured in weeks and months. The most-cited study on the question, by Phillippa Lally and colleagues at University College London, tracked 96 volunteers performing a chosen daily behaviour for 12 weeks and found that the median time to reach 95% of maximum automaticity was 66 days, with individuals ranging from 18 to 254 days.

Two caveats, because that figure gets quoted as though it were a law of physics. The study covered simple daily behaviours such as eating a piece of fruit with lunch, and a multi-step AI workflow is a harder thing to automate, so treat 66 days as a floor rather than a target. The second caveat is more cheerful: missing a single day did not measurably damage the process. Nobody has undone anything by skipping a Tuesday.

For planning, that puts habit change at roughly eight to 12 weeks of regular use for a straightforward daily task, and considerably longer where the task is occasional. A monthly board pack cannot become a habit in eight weeks, because eight weeks contains two repetitions. Frequency, not enthusiasm, sets this clock.

What does "trained" mean when somebody actually measures it?

In most organisations "trained" means attended, which is the weakest definition available. It records that a calendar invitation was accepted.

The gap between attending and applying has been measured. Training professionals from 150 organisations, surveyed by Alan Saks and Monica Belcourt for the journal Human Resource Management in 2006, estimated that 62% of employees applied training material on the job immediately, 44% at six months and 34% one year later. Those are self-reported estimates from the people who ran the training, which if anything makes them generous. Roughly a third of what you buy is still in use twelve months on.

Three definitions are worth separating, because each takes a different length of time to reach and only the third belongs in a board update:

• Attended. Same day. Measured by a register.

• Competent. Two to four weeks. Measured by watching somebody complete a real task unaided.

• Habitual. Eight to twelve weeks at minimum. Measured by tool telemetry, or by asking a line manager what has changed about how the work now arrives.

Agree which of the three the programme will be judged on before it starts. Otherwise it gets judged on the first and reported as a triumph, or judged on the third at week four and written off before the clock has finished running.

How long does AI training take by role?

AI training takes longest where the work is least repetitive and most heavily governed. The ranges below are planning assumptions we scope programmes against rather than findings from a study, and they assume structured sessions with support in between, not a single workshop. Fuller detail on designing that structure sits in our guide to how to train your team in AI.

The last row is the one people argue with. Executives are trained first and use AI least, because a leadership week does not contain enough repetitions of any single task for a habit to form. If leadership adoption matters to you, it needs a different design from the one used for the wider workforce, usually built around one recurring artefact such as the weekly board summary.

Why do AI training budgets get set against the wrong clock?

Budgets get set against the wrong clock because the thing procurement can buy is training days, while the thing that produces the result is elapsed weeks. Those are different units, and only one of them arrives with an invoice attached.

The effect is predictable. All of the money lands in weeks one and two, and the eight to twelve weeks that decide whether anything changed have no budget line, no owner and nothing in the diary. The correction is not more training. It is moving perhaps a fifth of the budget out of delivery and into the reinforcement window, where it pays for follow-up clinics, the internal champion's time, and whoever answers the questions in the team channel at 4pm on a Thursday.

Two numbers are worth agreeing before anything is signed. First, the date you will measure, which should sit at least eight weeks after the final session. Second, what you will measure, expressed as a change in how work arrives rather than a satisfaction score. Departmental order and who goes first are a separate question, covered in our walkthrough of rolling out AI training across 100 people.

If you are scoping a programme and want a second opinion on the timeline before committing a quarter to it, talk to us. Bring the roles you want trained and the date you have already promised somebody a result. Usually one of the two is wrong, and it is better to find out now.