A good day of AI training for staff looks much the same in most agencies. By four o'clock half the room has built something in ChatGPT or Claude they're pleased with, and someone's already planning to use it on next week's pitch.
Then Monday lands, a client wants the deck a day early, and the inbox has filled up while everyone was in the session. By Friday nearly everyone's back to working the way they did before.
I've watched this happen more times than I'd like. In my experience most of what people pick up on the day has faded within a week, and it happens after excellent sessions too. Whether anything sticks gets settled in the weeks after the training, which is the part almost nobody plans for.
Why AI training for staff fades so fast
Hermann Ebbinghaus mapped the basic pattern in the 1880s. He learned lists of nonsense syllables, relearned them after different gaps, and found that forgetting is steepest straight after learning and slows down from there. That's the forgetting curve. In 2015 Jaap Murre and Joeri Dros ran the whole experiment again, with one of them spending about 70 hours learning and relearning lists, and their results came out close to Ebbinghaus's original data.
Nonsense syllables are about as far from agency work as you can get, so I wouldn't lean on his exact numbers, though the shape of the curve carries over. Something new that doesn't get used soon after you learn it slips away quickly, and an AI workflow someone watched once in a workshop is exactly that kind of new thing.
The training transfer research points outside the room
Researchers call this transfer of training: whether what people learn on a course shows up in their actual work. A lot of that research builds on a model Baldwin and Ford published in 1988, which sorts the causes into three groups. There's the person being trained, the design of the training, and the work environment they go back to.
When Rebecca Grossman and Eduardo Salas went through the transfer research for the factors with the strongest and most consistent links, the work environment ones were transfer climate (roughly, whether the workplace makes it normal to use what you learned), support, the opportunity to perform, and follow-up. If you run an agency, every item on that list happens back at people's desks once the training's over, which puts most of it in your hands.
The surveys on AI at work point the same way. Microsoft and LinkedIn's 2024 Work Trend Index found only 39% of people who use AI at work had been given AI training by their company, and 78% of AI users were bringing their own tools. BCG's 2025 survey of more than 10,600 people found regular use was sharply higher among employees who got at least five hours of training and had access to in-person training and coaching. I'd underline the coaching, because it's the support from the transfer research turning up again in the AI data.
Learning has to happen on live client work
Of the four, opportunity to perform is the one I'd put first. It sounds obvious, but think about where the practice happens on a typical training day. It's on a made-up brief in a room away from the work, and afterwards you're left to work out alone how it applies to the real brief on your desk, usually on a Tuesday with the client chasing. Honestly, most people never get round to it, and I don't blame them.
So I'd flip it round and do the learning inside a job that's already on the desk. When the practice brief is the real brief, there's nothing to translate back afterwards, and the first go at the new way of working happens while the trainer's still there to help.
New habits take longer than a week
You'll have heard that a habit takes 21 days. That figure traces back to a 1960 book by a plastic surgeon, and the best-known real study puts it a lot longer. Phillippa Lally's team at UCL recruited 96 people who each picked one new daily behaviour, things like eating a piece of fruit with lunch. On average it took 66 days for the behaviour to feel as automatic as it was going to get, and the range ran from 18 days to a predicted 254.
Those were simple things done once a day. Changing how an account team writes a monthly client report is a much bigger shift, so I wouldn't expect it to settle any faster. One finding in there I do find reassuring: a single missed day had little effect on how the habit built up, while lots of missed repeats added up over time. Which is a good argument for regular follow-up, if that makes sense.
Put the three bits of research side by side and a quarter starts to look like the right length. Thirteen weeks covers Lally's 66-day average with room to spare, and it keeps support in place through the weeks when the new way of working is most likely to slip.
A short plan for making AI habits hold
If you're about to sign off AI training for your team, this is the shape I'd push for.
- Time the work before you start. Pick the workflows you want to change and time them as they run today, on real jobs. Without a before number you'll never know what the training was worth.
- Train on live client work. Choose one or two workflows the team does every week, like client reports or first drafts of proposals, and do the learning inside those jobs while they're happening.
- Give every workflow an owner. That's one named person who checks it's being used and fixes it when it breaks. Managers make a big difference here, and I've written about how manager support changes AI adoption separately.
- Write the new workflow down. If the new way only lives in the head of whoever was in the session, it disappears the week they're on holiday. There's more on that in why you can't fix a process that only lives in somebody's head.
- Keep following up for the whole quarter. A short, regular check-in on what's being used and what's slipping is the follow-up the transfer research keeps pointing at, and a one-off session can't include it.
- Time the work again at the end, then once more after you've stepped back. The second check tells you whether the habit holds when nobody's watching.
The question I'd take into your next training conversation is what happens in week two and week ten, after the trainer has left. That's where a new habit holds or fades, so it's where I'd want to see a plan and a person checking in.
If you want to see what that looks like with an agency team, here's how we run it: discovery first, then three months of training on live client work, then measuring what stuck.