The AI works. Nobody uses it. What went wrong?
The pilot was a success, the launch email went out, and three weeks later everyone is back on spreadsheets. The missing piece was never the technology.
Three weeks after launch
The project was a success by every measure on the plan. The pilot hit its accuracy target, the demo impressed the board, and the new AI assistant for the sales team went live with a company-wide email and a short video. A month later, a manager checks the usage dashboard. Eleven of sixty people logged in this week. Most are back on their old spreadsheets and their own notes.
No one was hostile. There was no revolt. The tool simply never became part of how people work. The project team, puzzled, goes looking for bugs and finds almost none. The technology was fine.
A tool that works but is not used is a tool that does not exist.
This pattern is so common that it deserves its own name in project plans: the adoption gap. Closing it takes a different kind of work from building the software, and it is usually the work that gets cut first.
Why people quietly go back
People rarely reject a new tool out loud. They drift away from it for reasons that are perfectly rational from where they sit.
- They are not sure what it is good for. A vague 'it will help you with sales' does not tell someone what to do on Monday morning.
- They tried it once, got an unhelpful answer, and concluded it does not work. Nobody showed them how to ask better.
- They worry about looking foolish, or about what the tool says about their own job.
- The old way is familiar and quick when you are busy, and the new way costs effort before it pays back.
None of these is a technical fault, and none can be fixed by an update. They are human reasons, and they need a human response.
What a useful training plan looks like
Good training is not a one-hour webinar with slides. It is built around the real tasks people do every day. Instead of showing every feature, it picks three or four moments in a working week, such as preparing for a client meeting, writing a follow-up or checking a contract clause, and practises exactly those.
People learn fastest with their own material. Letting them bring a real email they have to write today, then working through it with the tool, makes the value visible in ten minutes. A short, written 'how to ask' guide with a few examples that worked saves more support time than a long manual nobody reads.
Include an honest section on limits: what the tool is bad at, when to double-check, and what should never be pasted into it. Staff who understand the limits trust the tool more, not less.
Find your champions
In every team there are one or two people who are curious, a little ahead of the others, and respected by colleagues. These are the champions, and they matter more than any announcement from management. When a respected colleague says 'this saved me an hour yesterday', people listen.
Give them early access and some time to experiment. Ask them to share what works in the team chat, and to flag where the tool falls short. Their feedback improves the product, and their enthusiasm spreads in a way no training session can match. Recognise their effort publicly, so that being a champion is seen as valued work and not a favour.
Budget for it from day one
Training is not an extra to be added if there is money left. It belongs in the plan alongside development and testing, with its own time and its own owner. A rough rule of thumb from teams that have done this well is to set aside a meaningful share of the project effort for adoption, rather than hoping the interface will explain itself.
Define success in terms of use, not delivery. Instead of 'launched in March', aim for 'seventy per cent of the team uses it weekly by June, and median time on the target task is down by a third'. Review those numbers monthly, and treat a dip as a prompt to talk to people, not to blame them.
The best AI project is not the one with the most advanced model. It is the one that people actually open on a Monday morning.
Keep reading
Teaching teenagers to work with AI, not to hide from it
Schools are debating whether to ban AI. The more useful question is what students should be able to do that a chatbot cannot. Here is why building beats banning.
Read articleHow to pick your first AI use case, before someone picks it for you
After a workshop, the whiteboard holds forty ideas for AI. Most will disappoint. A simple scoring method, and a few ground rules, helps you start with the one that will actually succeed.
Read article