Your AI Training Ended in Week One. So Did Adoption.

Key Takeaways

• Launch training is the start of adoption, and most companies treat it as the finish. Half of trained employees in one large survey say their AI training isn't ongoing.
• The tools move faster than any course. A curriculum written at rollout is out of date within a quarter, so training has to ship on the same cadence as the product.
• Hours and coaching matter more than a polished kickoff. BCG found people with five or more hours of training are much more likely to become regular users.
• Managers decide whether training sticks. If a team lead never asks "how did you use AI on this?", the habit fades.
• Ongoing support has four parts: office hours, champions, a curriculum that refreshes as the tools ship, and follow-up where usage lags.

Most AI rollouts have a great first week. Accounts get provisioned, somebody runs a lunch-and-learn, a few people build something clever and post it in Slack. Then the calendar fills back up. By week six, the people who were already curious are still using the tool, and almost everyone else has quietly gone back to the old way.

The survey data says this is the normal outcome. In Lenovo's 2026 Work Reborn research, a survey of 6,000 full-time employees at large companies across 12 countries, 51% of employees who received AI training said it isn't regular or ongoing. 42% said it isn't effective. And 31% of AI users said their employer gives them no training at all. (Lenovo sells AI hardware and services, so read it with that in mind. The pattern matches what we see.)

I think most companies are treating AI training like a software launch. Run the session, check the box, move on. That model worked fine for tools that didn't change much after you bought them. It breaks for AI, because the tool changes every month and the skill is a habit you build over weeks.

What fixes it is support that keeps going long after the kickoff.

Week One Teaches the Tool. Week Six Decides the Habit.

Launch training does one job well. It gets people logged in, shows them what the tool can do, and explains the rules. That's necessary. It's also the easy part.

The hard part comes later, when someone sits down with a real task on a busy Tuesday and has to decide whether to try the AI way or the way they already know. The old way wins almost every time, because it's familiar and the deadline is real. Nobody fails at AI adoption in the workshop. They fail at the desk, three weeks later, alone.

That's the gap a one-time course can't close. A workshop can show you a good prompt. It can't be there when your prompt doesn't work on your actual spreadsheet and you have 20 minutes before a meeting.

Plan for the dip

Expect usage to spike after launch and then fall. That's normal. What matters is whether you have something scheduled for the moment it falls: a follow-up session, an office hour, a champion who checks in. If the plan ends at the kickoff, the dip becomes the new baseline.

The Tools Change Faster Than Any Curriculum

Here's the part I think people underestimate. Claude, ChatGPT and Copilot all ship meaningful features every few weeks. New connectors, new ways to share work across a team, new agent features. A curriculum written at rollout describes a product that no longer exists by the next quarter.

So a team trained in month one is working with a picture of the tool that gets less accurate every month. They don't know about the feature that would have saved them an afternoon. And nobody is going to read release notes on their own. (I barely do, and this is my job.)

We had a version of this problem inside PressW. AI moves fast and our team shares a lot of good material across Slack, Notion and everywhere else, so people were losing hours trying to keep up. We built an internal tool that pulls it all together and sends a summary every Monday, and it saves an estimated 4 hours per person each week. Even a team of AI engineers needed a system to keep learning. A finance team or an operations team needs one more.

Refresh the curriculum on a schedule

Tie training to the product roadmap and to your own rollout plan. When a new feature lands, or when you deploy a new skill or workflow, run a short session on it. Keep each one small and specific. Thirty minutes on one thing people will use this week beats a two-hour refresher on everything.

Hours and Coaching Beat a Polished Kickoff

There is decent evidence that the amount and shape of training matters more than the launch event.

In BCG's AI at Work 2025 survey of more than 10,600 workers across 11 countries, employees who got five or more hours of training were much more likely to become regular AI users, especially when that training was in person and included coaching. The same survey found only 36% of employees feel adequately trained.

Five hours is not a lot. But almost nobody gets five hours in a single kickoff, and nobody would retain it if they did. The way to get there is spread out: an hour at launch, then short sessions, office hours and one-on-one help over the following weeks.

Coaching is the part I'd underline. A trainer standing at the front of a room teaches the tool. A coach sitting next to someone working on their own task teaches the habit. That second thing is what changes behavior.

Spread the hours out

Instead of a half-day workshop, plan five or six short touchpoints across the first two months. Each one should end with the person having done real work in the tool, on their own task, with help nearby.

Managers Decide Whether Training Sticks

You can run perfect training and still lose, if managers aren't on board.

The same BCG survey found that regular AI use among frontline employees has stalled at 51%, and only 25% of frontline workers say their leaders give them enough guidance on AI. I don't think those two numbers are a coincidence.

People take their cues from their manager. If a team lead never uses the tool, never asks about it, and never makes room in the schedule to learn it, the team reads that correctly: this isn't a priority. If a manager starts a review with "how did you use AI on this?", the team reads that correctly too.

This is where a lot of training programs go wrong. They train the individual contributors and skip the people who set the norms.

Brief leaders before you train teams

Give managers their own session first. Cover what the tool is good for in their team's work, what you expect them to ask for, and how adoption will be measured. Then ask them to show up to their team's training. It costs an hour of their time and it's the cheapest lever you have.

Support Is the Part of Training People Actually Use

When I think about what makes AI training stick, I keep landing on one idea: the training is the small part. The support around it is the big part. Lenovo's survey found half of employees say better training would help them get more value from AI at work. I read that less as "we want more courses" and more as "we want help when we're stuck."

Here's how I'd break down the ongoing layer. Four pieces, each covering a different way adoption fades.

None of these are complicated. Most companies skip them anyway, because they don't look like a project. They look like overhead. That's the mistake.

Follow the usage data, not the attendance sheet

Attendance tells you who showed up. It tells you nothing about who changed how they work. Look at active use by team, week over week. When a team drops, go back to that team specifically. That's a much better use of a trainer's time than another all-hands session. We set this up through PressW's AI Usage Analytics, so training follow-up is driven by who actually stopped using the tool.

Shadow AI Fills the Gap When Support Runs Out

There's a cost to getting this wrong that goes beyond wasted licenses.

The same Lenovo research found that up to a third of employees use AI without IT oversight, and 22% of AI users say their employer doesn't provide AI tools at all. People want to use AI. If the approved tool feels hard and nobody helps them with it, they'll open a personal account in another tab and paste in whatever they're working on.

That's a security problem, but I'd frame it as a support problem first. People take the path of least resistance. Your job is to make the approved path the easy one.

Make the sanctioned path the easiest one

Put your usage policy into the training, in plain language, with real examples of what's fine and what isn't. Then make sure the approved tool is set up well enough that people don't need a workaround: the right connectors, the right access, the right skills already loaded. Good Claude deployment and good training are the same project.

Train Once, Then Keep Showing Up

I'm convinced the companies that get real value from AI over the next few years won't be the ones with the best launch day. They'll be the ones that kept showing up in month three, month six and month twelve.

What I'm less sure about is the right long-term model. How much of this support should live inside the company versus with a partner? How small can a champions program be before it stops working? When the tools start teaching people how to use themselves (and they're getting closer), what's left for a trainer to do? I don't have clean answers yet. I suspect the answer to the last one is "the judgment part," but we'll see.

What I'd do this week: pull up your AI rollout plan and look at what's scheduled after the kickoff. If the answer is nothing, that's where your adoption went. Our AI Training work is built around that ongoing layer: role-based sessions, office hours, champions and a curriculum that changes as the tools do. If you want to talk through what that could look like for your team, Book a Discovery Call.

FAQ