Written by Tim Cakir
Most companies buy the licenses, run the training day, and then wonder why nothing changed.
I have watched this from the inside more than a hundred times. The tools work. The people are willing. And three months later the usage reports look the same as they did before anyone signed anything.
The reason is simple. Adoption is a habit problem. You cannot buy a habit.
The two people in every team
Sit in on any team eight weeks into an AI rollout and you can sort everyone into two groups within an hour.
The first group opens the tool before they know exactly what they want. They type something messy. They get back something wrong. They read why it went wrong and they go again.
The second group waits until they can write a perfect request. Which means they mostly do not open it at all.
Same license. Same training day. Same manager. The gap between those two people after three months is enormous, and none of it came from the software.
Watch what someone does with a bad first answer. That single behavior predicts more than any skills assessment you can run.
What HR usually measures, and why it misses
Most rollouts get measured on license activation and training completion. Both are easy to collect and neither tells you anything.
A person can activate a license, sit through the session and tick the box. Then change nothing about how they work on a Tuesday. Your dashboard will show 100% completion. Your operating reality will show nothing.
The number that matters is how many people used it on a task they were already doing, without being asked. That is harder to collect. It is also the only signal that predicts whether any of this survives the quarter.
Leadership presence beats leadership sponsorship
Here is the pattern I trust most, because I have watched it hold across very different organizations.
At Source Capital, a lower-middle-market private equity firm, we trained 123 leaders across three cohorts. The Managing Partner, Tom Harbin, did not sponsor the program from a distance. He sat in and personally co-built the tool that won the internal competition.
Screening a deal at that firm used to take a couple of days. It takes about seven hours now. More than 50 organizations across their portfolio and peer network were touched by what those cohorts built.
None of that came from a mandate. It came from people watching a Managing Partner be visibly bad at something new in front of them. Then watching him get good at it.
That is the part you cannot delegate. A sponsor sends a memo. A participant makes it safe to be a beginner.
Twelve weeks, not one day
A one-day workshop produces a good mood and no behavior change. I know because I ran them for years before I stopped.
The format that works is spaced out. We run twelve weeks, one live session a week, around two to three hours of work in between. The length is the mechanism. It gives people enough failed attempts to get past the first bad answer. It also gives them time between sessions to try things on real work rather than exercises.
Baseline, from training alone and before anyone builds a tool or an automation, is about five hours per person per week. At a 450-person software company we worked with, eight participants documented between 33 and 51 hours saved per week between them.
Those numbers do not come from the software being clever. They come from eleven more weeks of practice than a workshop gives you.
What to do differently on the next rollout
Four things, in order of how much they matter.
One. Put a senior person in the room as a participant. Not to open the session. To do the exercises badly in front of everyone. This is the single biggest item on this list. It costs nothing.
Two. Stop measuring completion. Ask a different question at week eight: what did you do this week that you would have done differently in June? If people cannot answer, the program is not working, whatever the dashboard says.
Three. Give it weeks, not a day. Behavior changes through repetition on real work. Any format without spacing is a communications exercise wearing a training badge.
Four. Make people show their reasoning. The fastest learners I see are the ones who read why the model answered the way it did, rather than grabbing the output and moving on. Build that into how teams share work. Have someone explain a result out loud before it gets used.
The uncomfortable part
If your rollout did not change anything, the tools are almost certainly not the reason. Two companies with identical licenses get completely different outcomes, and the variable is daily practice.
That is good news, because it is the part you control. It is also harder than buying software, which is why most organizations quietly keep buying software instead.
Start with who sits in the room. Everything else follows from that.
Author Bio:
I’m Tim Cakir, founder of AI Operator, and I spend my days helping people understand AI and actually enjoy working alongside it.
There’s a lot of extreme views about AI. Some claim its a magic pill that will solves all our problems and make us rich. Others see it as existential threat to humanity that will take away our purpose.
Here’s what I think:
AI is a tool – an extremely powerful tool that removes the boring, repetitive parts of work, frees up our time, and allows us to focus on more creative, meaningful projects.
I’ve trained teams at companies like Google, Recharge, and ScaleUp, showing them how to bring AI into everyday workflows, in ways that actually stick. My favorite moment is when an AI skeptic suddenly “gets it” and starts imagining all the ways it could help them.
I’m on a mission to help more people see AI as an opportunity, not a threat, and give them the tools to make it part of their own success story.
And outside of work? You’ll probably find me sailing, DJ-ing, making music with my wife – or building with AI.