What happens after the AI tool is switched on

"We rolled out an AI tool and it hasn't changed much." If that sounds like your firm, the cause is usually what happened after launch, because a tool left to spread on its own gets used unevenly, each person in their own way. Someone has to change how the work is done, show people how, and check back after a month, and that needs a budget, just as the tool does.

The same tool, very different results

Microsoft's own researchers gave Copilot licences at random to more than 6,000 workers across 56 firms. In some firms three in four people used it; in others, fewer than one in ten. Every firm had the same tool, so the gap came from the firms: their work, their people and what they did once the licences arrived. Where you land in that range depends on what you do after launch.

The UK government ran a large trial of its own, giving Copilot to 20,000 civil servants for three months in late 2024. Its findings report says "driving adoption required substantial effort" and that training "was essential to achieve benefits of the tool". Each department supported its own users, and how well it went differed from one department to the next. So even with the same tool across government, getting people to use it well took real work, and training was essential to getting anything back.

Left alone, people use it unevenly

Doing nothing can look reasonable, because staff are already using AI without being asked. Deloitte found that 63% of UK working adults say they use generative AI for work, and around half of those say they have had no formal training. That means AI is probably already in your building, used by people who were never shown how. When people teach themselves, a few get faster and nobody can say what the business gets back.

When use spreads like that, the time it saves tends to disappear. In BCG's 2026 survey, 66% of frontline workers who use AI regularly said they get "limited or no guidance on what to do with the time they save". BCG says that time "leaks out of the organization unless it is tracked and deliberately reinvested". So a tool can save your staff hours every week and still leave nothing you can point to. If nobody tracks the time saved, nobody can say afterwards where the hours went or whether the tool paid.

Three jobs that need a name against them

The first job is changing the work itself. BCG estimates that 70% of the value from AI comes from people and processes, and the rest from algorithms, data and technology. So most of what you get back depends on changing how the work is done, and someone has to own that. If a tool now drafts supplier letters, someone still has to check them, and someone has to decide who checks and against what, then write the new way down.

The second job is showing people how, inside their own work. Roche, the drug company, made AI training compulsory for every employee, then ran regular, voluntary sessions where colleagues showed how they were using it. In its regulatory team, around half of staff say they now save four hours or more a week, up from almost nothing before. The people showing others how came from the same teams, so they already knew the work. So choose the people who show others how from the teams that do the work.

The third job is checking back. A month after launch, someone sits down with the team and asks what they use the tool for, where it gets in the way, and where the saved time has gone. In a US Gallup survey, staff who strongly agree their manager supports their use of AI are 2.1 times as likely to say they use it a few times a week or more. Yet only 28% of employees in organisations bringing AI in strongly agree. A review meeting a month in is a cheap way for a manager to show that support, and it is the moment you find out whether the tool is paying.

What to budget for

Budget for training and follow-up as seriously as for the tool, and put it in the plan from the start. That means a named person for each of the three jobs, hours set aside in their week to do it, and a review meeting in the diary for a month after launch. If that budget is not there, wait before buying, because a tool without it gets used unevenly and the time it saves goes unrecorded.

If your tool is already live and not doing much, the same three jobs apply now: pick the names, change one process, and book the meeting.