Your Team Isn't Resisting AI. Nobody Gave Them the Hours.

When a company asks me to help with an AI rollout and tells me the team is resisting it, I don't start by trying to change the team's attitude. I want to know what happened when they tried to use it, and where in the week they were supposed to learn.

Because in the rollouts I've been called in to fix, what got called resistance turned out to be something else, a tool that was too complicated, a system designed around the wrong work, training that didn't keep up, or nobody giving people the time and help to get through the first attempts.

You can give everybody a login and tell them AI matters, but if the client work stays the same and their targets stay the same, you haven't given them a way to learn it. You've added another thing to a week that was already full.

I call that the Unfunded Hour, and if you're the owner, this is a decision you can do something about.

Think about the person whose work is already booked. The client is waiting, the manager can see whether the job is finished, and the hours spent on it show up in the numbers the firm reviews.

Learning the new system has no place in that plan. Even if the person believes it will help later, they still have to choose what to do now, and the work with a deadline wins.

That doesn't mean the team is incapable of learning, it means they are following the priorities you've made visible. If you want a different choice, some of the work or some of the target has to move.

The research gives you a reason to take that answer seriously. In Study.com's March 2026 survey of 1,000 US employees, 41% named lack of time as a barrier to getting better at AI, the most common answer.

Two-thirds said two hours a week or less would meaningfully improve their skills, while 36% said their employer provided paid time to learn. Those are employee reports, not a measured training prescription, but they describe a gap you can check against your own calendar.

And this wasn't simply a group waiting to be ordered into training. Only 9% named an employer requirement as their main reason for learning, with improving their work and personal curiosity much more common reasons.

Study.com sells training, so it has an interest in this problem. Still, the time issue appears elsewhere, Skillsoft reports that 58% of leaders and 59% of individual contributors named lack of time as the biggest barrier to building new skills.

You may be thinking that AI has already saved you so much time that your team should be able to find a couple of hours. Let me show you why I wouldn't use your own experience as the answer for everyone else.

Section surveyed 5,000 white-collar workers at companies with at least 1,000 employees, in the US, UK and Canada, in late 2025. As the Wall Street Journal reported, 40% of non-management staff said AI saved them no time, compared with 2% of the C-suite.

Those are large companies, not a direct measurement of a firm your size, and the answers are self-reported. But they should stop you from assuming the person running the business and the person doing the client work are getting the same benefit from the tool.

Two-thirds of non-management staff in that survey saved less than two hours a week or no time at all, while more than 40% of executives said they saved over eight. An owner can honestly think the tool is wonderful while a team member honestly cannot see where the promised time went.

So where would I find the learning hours? I would look at the training time the firm already has before asking for a new budget.

SPI Research's 2025 professional-services benchmark reports 9.2 guaranteed training days per employee in its 2024 figures. On an eight-hour day, that's about 74 hours a year, and ten hours of AI training would use roughly one seventh of that pool.

That is the benchmark, not a claim that your firm has 74 unspent hours waiting in an account. Check your own plan, see which hours are available and which commitments can move, then put the training where people can actually attend it.

The same caution applies to utilization, the share of working time billed to clients. Deltek's summary of SPI's benchmark puts it at 66.4% in 2025, but the remaining time is not automatically spare time.

People still sell work, run the firm, handle admin and meet commitments that don't go on a client invoice. You need to look at that work rather than announce that a third of the calendar is free.

If there is no training time to reassign, then funding it means making a real capacity decision. I would rather you face that cost before the rollout than quietly leave people to pay it with evenings and weekends.

And I would book general AI learning as a firm cost, not pass it to a client as if it were delivery they asked for. Any client-funded training needs to be part of an explicit agreement, not a surprise line on an invoice.

Now, how much AI training do employees need? I can give you the pattern I've seen, and a research finding beside it, but they are not the same kind of evidence.

BCG's 2025 AI at Work survey, covering more than 10,600 people across 11 countries and regions, found regular use was sharply higher among employees who received at least five hours of training and had access to in-person teaching and coaching. Only about a third said they had been properly trained.

That is an association, not proof that five hours causes somebody to become a regular user. It does tell you that buying access and preparing people to use it are different things.

Across the franchise delivery work I see, the pilots that stick usually get 8 to 10 hours of training over about three weeks. That includes video courses, in-person sessions, questions and answers, and live support from the technical team while people try it on real work.

I'm describing a pattern from those pilots, not promising that ten hours will make any tool work in any firm. The job you picked and the system you built still matter.

But notice the shape of the training, because the hours alone aren't the whole story. People learn something, use it, find the part that doesn't behave as expected, and have somebody available to help before they give up and return to the old way.

A video can help with the first part. It cannot answer the question someone brings after trying the system on their own client work, and one afternoon doesn't give you that cycle across several attempts.

You also need to check what people can do, not just ask whether they feel confident. In the Study.com survey, 34% reported confidence using AI, while 18% said they consistently produced high-quality output with little editing, and both figures are self-reports rather than an outside assessment.

There is a useful historical warning from METR's early-2025 developer experiment. Experienced developers believed AI had made them about 20% faster, while the measured result was that they took 19% longer on the tasks in that study.

METR's February 2026 update says those older speed results should not be treated as the current effect, and selection problems prevent a reliable estimate from its newer data. I keep the example for the gap between feeling faster and being faster, not to claim your team will slow down by the same amount.

That is also why skill and benefit are different from who adopts first. Opening the tool is a start, it isn't the result you're paying for.

And even funded training can stop in the wrong place. Teaching someone to get a nicer draft out of a chatbot can help that task without changing the rest of the job, and the saved time can disappear into checking, rework or other work nobody planned to change.

That's the problem in the extra time AI buys you is already gone. For this rollout, keep your attention on the piece of work you're trying to change, not just on how well people can prompt.

The Section respondents most commonly described using AI for search replacement and drafts, with fewer using it for analysis or coding. The same Wall Street Journal report described a separate Workday survey of about 1,600 employees, in which 85% reported saving one to seven hours a week, with much of the benefit offset by correcting and reworking AI output.

Those are separate surveys, so you can't combine them into a calculation of lost time. They do give you a useful question for the training plan, will people learn the whole new way of doing the job, including checking the result, or only the fast first step?

An AI system attached to a defined role and workflow asks more of the firm than a chat window does. Across the 150+ franchise units in the delivery data I review, the recorded split was roughly 80% chatbots, 15% rigid workflows and 5% autonomous agents defined by conditions, not a picture of the whole market.

On the Delivery Model Ladder, the move that matters here is from making individual tasks faster to changing how the work runs. Training needs to prepare people for that change, who checks the output, when the system stops, and what the team does with the time it frees.

BCG's 2026 survey reported frontline regular use rising from 51% to 74%, while the share reporting adequate training was 36%, still close to the roughly one third reported in 2025. Adoption can move much faster than the support around it.

I've also watched pilots get announced rather than run. People see a demonstration, receive some instructions, and then the people responsible for the system wait for somebody to report a problem.

Let me tell you why I don't trust that last step. I have seen a SaaS tool I use every day break and never reported the bug to the people who made it, so I can't expect everybody else to be a perfect source of feedback either.

If your new AI system needs constant correction and nobody tells you where it failed, it can stay broken until people stop opening it. Go back to the team, look at what happened on real work, update the training when the agent changes, and make reporting a problem easy.

In Study.com's survey, 56% of the people who had received AI training said practice on real work was the biggest thing missing. Generic examples won't tell your team how to handle the client task waiting for them after the session.

And don't confuse capability with permission. A written AI policy tells people which tools and uses the firm allows, while training helps them do the permitted work well, you need both.

There is a counterpoint worth keeping in the room. In the Section survey, 40% of all respondents said they would be fine never using AI again, and non-management workers were more likely to describe anxiety or overwhelm than excitement.

That isn't proof that missing training caused their feelings, and it doesn't disappear because another survey found curiosity. The populations and questions differ, and someone can want to get better at a tool while wishing it wasn't part of the job.

So ask about the job anxiety too. Explain what you're changing and invite people into documenting, testing and improving the work, rather than treating every objection as an attitude you have to overcome.

Managers often need their own version of that conversation. In my experience, operational teams tend to engage quickly when they can see a direct benefit, while managers and supervisors can take considerably longer and need dedicated training time.

The tracking work is familiar, reading spreadsheets, running numbers, checking status. When an agent takes that over, the manager is left with more of the judgment work, feedback, mentoring, one-on-ones, helping people improve, and often no written way to do it.

You haven't made the manager less capable, you've changed what their week needs to contain. Leadership has to help them learn and organize that work, which is the harder part of fewer managers doing more managing.

For your next rollout, pick one workflow worth improving and name the person who owns it from start to finish. Find the training hours, move the other commitments that would consume them, and spread the learning across real attempts with help available.

Use the 8 to 10 hours over about three weeks as a starting point from my experience, then adjust to what the work requires. Give managers material for the part of their role that changes, and keep someone responsible for following up on problems and results.

If the workflow isn't worth changing, more training won't make it worth changing. And if nobody owns the whole job, the training can improve individual skill while leaving the operation where it was.

Before you diagnose resistance, ask the people who aren't using the system what happened when they tried, and whether they had time to try properly. If it doesn't help with their work, fix that problem, and if they haven't had the hours, change the week you gave them.