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

A stalled AI rollout in a service firm almost always gets diagnosed as resistance, and the diagnosis is almost always wrong. The constraint is hours nobody budgeted.

In a survey of 1,000 employees fielded in March 2026, lack of time was the top barrier to getting better at AI, at 41%. Only 9% said they were learning it because their employer required it. Two-thirds said two hours a week or less would meaningfully improve their skills, and paid time to learn reaches 36% of them.

Owners misread this because their own experience of the tool is nothing like their team's. Section surveyed 5,000 white-collar workers in late 2025: 40% of non-management staff said AI saved them no time at all, against 2% of the C-suite.

I call this the Unfunded Hour. The firm bought the tools, the people want to use them, and nobody gave anyone the hours. The firms that do fund the time mostly spend it on the wrong training.

Why is "I don't have time" an accurate statement?

Because employees are following the scorecard you gave them. Billable hours get measured every week. Nobody measures capability. People follow the scorecard, and it points to billable work.

In a service firm, learning time comes out of work you could sell. The same hours create the work and pay the bills. A software company can train people in spare time without moving client work. A service firm has to take those hours from a person's work or target.

The people who need the tool most have the least room for it. The senior person who would get the most out of the tool is the one whose week is already sold, because being worth selling is what filled the week. The tool becomes one more item on a list that was too long before AI existed, and it stays there.

That is why pep talks fail here. You can tell a team AI matters, and every person can agree, but the calendar still wins on Monday.

The Section survey covered companies with 1,000 or more employees in the US, UK, and Canada. Two-thirds of non-management staff saved under two hours a week, while more than 40% of executives said they saved over eight. That 38-point gap on "no time saved at all" was the largest in the survey.

The research says the same thing in blunter language. Study.com titled its section on this "Barriers to AI Upskilling Are Structural, Not Motivational." Leaders agree in their own data: 58% of leaders and 59% of individual contributors both name time as the biggest barrier to building new skills. Everyone in the building can see the problem. Almost nobody has written it into a plan.

Where do the hours come from?

They already exist, and that is the part most owners miss. Professional-services firms budget about 9.2 guaranteed training days per employee per year, according to SPI Research's benchmark. That training time is already in your budget.

There is more room than that. The share of time billed to clients fell to 66.4% in 2025, the lowest in SPI Research's survey history, according to Deltek's summary of the benchmark. A third of the calendar is already outside client work. Those hours are not missing. They have no assigned use.

So the first move is to move hours already set aside, not ask for a new budget. Nine days is roughly 74 hours a year per person. Ten hours of AI training is a day and a quarter out of that, one seventh of a pool you already pay for. That is a different conversation than asking a partner to write off billable time.

Operators usually ask next whether the hours should be billable. They should not. Learning time billed to a client is a fee for work the client did not ask for, and the first time it shows up on an invoice you will lose the argument and the hours. Book it where the training days already sit, as a firm cost, and let the return show up in cost to deliver.

How much AI training do employees need?

BCG gives us a lower point of comparison. It surveyed more than 10,600 employees across 11 countries and found that regular AI use is sharply higher among people who get at least five hours of training, plus access to in-person teaching and coaching. Only one-third said they had been properly trained. Five hours is roughly what separates a paid license from a person who opens it.

The number I see in delivery is higher. Across a franchise network's delivery data, the pilots that stick get 8 to 10 hours of training over about three weeks. That includes video courses, in-person sessions, Q&A, and live support from the technical team while people are trying it on live work.

Those two numbers describe different outcomes. Five hours helps get someone started. In the pilots that stick, 8 to 10 hours helps a change survive after the enthusiasm wears off. Most firms give zero and then read the result as an attitude problem.

The schedule matters as much as the total. Spread it across three weeks instead of one afternoon. Mix the formats, because a recorded video answers none of the questions people bring. Keep live help available the week they first try it on real work, which is when it breaks and when they decide whether to keep going.

Why not let people figure it out on their own? Because people can feel confident before they can do good work. In the Study.com data, 34% of employees feel confident using AI and 18% consistently produce work that needs little editing.

METR ran a randomized trial in early 2025 where experienced developers believed AI had sped them up by 20%, while the measurement showed they took 19% longer. METR has since said developers are probably faster in 2026 and that its newer data cannot size the change, so treat the slowdown as history. The finding that held is the one about self-assessment: people are poor judges of their own speed. Skill decides who benefits, which is the argument I made about age and adoption.

Why do funded hours still fail?

Because training can improve one task without changing the work around it. Teaching people to prompt a chatbot may make an individual task faster, but the saved time can disappear into correcting the answer and reworking the rest of the job. That is the whole argument of why the extra time AI buys you is already gone.

The Section survey shows where most firms stop. The most common uses people reported were replacing Google searches and generating drafts, with far fewer using AI for data analysis or code generation. That is basic chatbot use, and an AI that only answers questions is a better Google search.

You can see the gain disappear in the numbers. Workday surveyed about 1,600 employees and found 85% saved one to seven hours a week, with much of it offset by correcting errors and reworking what the AI produced. Workday calls it an AI tax, in figures reported alongside the Section survey. Anyone who has rewritten a bad draft twice knows the feeling and did not have a name for it.

Training people on an AI system attached to a designed workflow, one that takes over work the business sells, is a different exercise with a different result. That is the difference between a chatbot and an AI employee. Across the 150+ franchise units, roughly 80% are still running chatbots, 15% have moved to rigid workflows that run on triggers, and 5% are running autonomous agents defined by conditions.

On the Delivery Model Ladder, hours spent on prompting leave you at Stage 1. Cost to deliver drops, but nothing else moves. Hours spent on a rebuilt workflow are what Stage 2 costs. Most firms reading this are at Stage 1 and believe they are further along.

There is a second question underneath: what share of the hours AI already freed has a named use? If no one names a use, the lower cost to deliver turns into spare time that disappears into the day. When the use is named, the hours go one of two places: into work where clients feel the difference and retention answers, or into an offer that sells a finished outcome instead of the hours it takes, where price answers. Both are Stage 2. Neither happens on its own. The training hours have the same problem one step earlier.

Nobody closes this gap by accident. BCG's frontline regular AI users jumped from 51% to 74% in a single year. Over the same period, the share who said they had been adequately trained barely moved, from about one third in 2025 to 36% in 2026. Firms bought tools without buying hours.

What does "resistance" usually turn out to be?

Every time I have been called in to fix a rollout the company described as resistance, the cause was somewhere else. Usually the system was designed wrong, or the tool was too complicated for the job it was doing. Often the training was written once and never updated as the agent changed. Almost always, nobody captured feedback, nobody followed up on results with the team, and nobody said clearly what the firm was trying to achieve.

The clearest evidence on that list concerns training that has nothing to do with the work. In the Study.com data, among employees who received AI training, 56% say practice on real work tasks was the single biggest thing missing from it. Generic prompting exercises teach people to use a tool. They do not teach anyone to do their own job differently.

Pilots that fail tend to be announced rather than run. The Enthusiasm Flood is what the Production Gap looks like from the delivery side when that happens.

Feedback is where most of them quietly die. People are bad at reporting problems, and I am not an exception. I have watched a SaaS tool I use every day break and never told the people who made it.

Now apply that instinct to an internal AI project that needs constant correction in its first month. If nobody tells you what broke, you cannot fix it, and the tool stays broken until people stop opening it. Follow-up, captured feedback, and training that gets updated as the system changes will do more for adoption than any amount of encouragement.

Training builds capability, while a written rule governs use. I have argued before that an AI policy on one page is what stops people using unapproved tools, and that remains true. Hours turn a user into a competent one.

The same survey gives a harder number. 40% of respondents said they would be fine never using AI again, and workers were more likely to describe themselves as anxious or overwhelmed than excited. In the C-suite it was the reverse. That is the part that looks like resistance.

Read that alongside the survey's other finding. Those people never got past google-search replacement, and they are describing life after an untrained rollout at a large company. Indifference like that is what the Unfunded Hour produces on its way out.

AI also brings real anxiety about what it means for the job. Training is where expectations get set and where people get invited to document, test, and improve the thing instead of having it done to them.

Why are managers the hard case?

Operational teams get to business fast, see the benefit quickly, and explore the tool heavily; managers and supervisors take considerably longer and often need dedicated training time.

The reason is practical. A manager's day runs on tracking work: spreadsheets, data entry, running numbers, chasing status. That work is visible and countable, and it makes a person feel productive at the end of a day. It is also the first work an agent takes over.

When AI absorbs the tracking work, managers are left without a written process for the judgment half of the job: feedback, one-on-ones, mentoring, and motivating. The transition needs deliberate attention from leadership.

Without that process, they keep running the people work the way they always did, as a small slice of the week, at exactly the moment it needs to become most of it. This is the cost side of fewer managers doing more managing. Handing tracking to an agent is the right move, but the person on the other side of it usually has no process to fall back on.

What should you fund, and how?

The fix is specific enough to put on a calendar.

1. Take the hours from the training-day pool you already budget. The budget objection ends here, because the money was already spent.
2. Set aside 8 to 10 hours per person over about 3 weeks. Spread it out rather than using one session or one afternoon.
3. Mix the formats. Keep live help available during the week people first use it on real work.
4. Point the hours at one workflow that matters to the business. One workflow owned and finished beats six announced.
5. Give managers their own time and their own material. Their job is changing more than anyone's.
6. Capture feedback, follow up on results, and update the training as the system changes.

Then change what you watch. The number of seats in use tells you nothing. Ask whether the workflow changed.

Two conditions have to hold for any of this to pay off. Someone has to own the workflow from start to finish, or the funded hours produce individual skill without changing how the work runs. The workflow also has to be worth automating, or you finish with a well-trained team using a tool the business never needed.

The test you can run this week

Ask the people who are not using the tool: do they doubt it, or have they had no time to use it?

If they tell you they do not think it works, you have a design problem, a tool problem, or an expectations problem, and more hours will not fix any of them. If they tell you they know it would help and have not had a chance to sit down with it, the problem is the calendar. That answer accurately describes the week you gave them, so check the calendar before you diagnose the culture.

If this was useful, subscribe to the newsletter.