What It Looks Like When Activation Actually Works
Ten months ago, a client came to us wanting to plug into everything happening with AI. They could feel the momentum building around them and didn't want to be left out of it, but there was no shared language yet for what counted as a real AI solution versus a novelty. If you'd asked ten employees to describe what the tools were good for, you'd have gotten ten different answers, because there wasn't yet a shared way to talk about it.
Getting from there to somewhere real took ten months of real work. Recurring meetings, department by department. Education sessions we ran more than once, because ideas worth using well are worth teaching more than once. Broad communication paired with one-on-one conversations, because some people needed the big picture and some people needed someone to sit with them and work through their own week. We were honored to be in the room for every version of that, the company-wide sessions and the single conversation with one person still figuring out where AI fit into their day. That consistency, showing up for it again and again and meaning it, is what actually determines whether anything sticks.
Last week I sat in a meeting where ten people from that same company took turns presenting one slide each: what they built, what it saved, and what they're doing next. Nobody was told what to build. Nobody was reading from a script we wrote. A small legal team walked through how they used AI to run an entire acquisition in house, something that used to trigger outside counsel and a company-wide fire drill, and saved well into six figures doing it. Someone on the content team, who isn't an engineer, built a tool with a few AI-assisted afternoons and some light scripting that cut a recurring task down from most of an afternoon to a few minutes, for dozens of coworkers who rely on it daily. Someone from the infrastructure team showed a tool that replaced a patchwork of manual requests and a spreadsheet with a single web app managing hundreds of pieces of hardware. Someone from the data team mentioned, almost as an aside, that the whole team had picked up building AI solutions on their own, and had logged more than a dozen of them so far this year.
I didn't say much in that meeting. I didn't need to. I sat back and watched people who'd started this ten months earlier with no shared vocabulary for any of it now stand up and teach each other. That's the moment I keep coming back to. Not because of what it says about us, but because it's rare enough that I know how much had to go right underneath it.
None of what got presented is our win to claim. That's the point.
What was actually presented
The range mattered as much as any single number. Nine departments took a turn, spanning legal, content, marketing, partnerships, sales, day-to-day operations, software engineering, infrastructure, and data and analytics. A couple of the smaller teams reported savings in the four-to-six-thousand-hour range for the year. A few of the larger, more technical teams reported figures well above that, into the tens of thousands of hours. The smallest team in the room, a handful of people, posted the single biggest dollar figure of the day: a six-figure amount saved on one project by using AI in place of outside help that would normally have been required. More than one presenter said their own number was probably an undercount, because nobody on a team that size has time to log every AI solution people quietly build for themselves.
Add it up, even conservatively, and you land somewhere north of seventy thousand hours saved this year, spread across at least seventy separate AI solutions that people built and are actively using, not handed down from a central team but built by whoever needed them. A couple of departments also flagged real revenue on the table if a project currently in progress gets finished and turned into something sellable, worth six or seven figures by their own estimate. That one belongs in a different column. It hasn't happened yet.
What activation actually looks like
When it works, the meeting stops looking like training and starts looking like what I saw last week. People stop asking permission and start asking each other questions. A marketing lead takes what her team has been learning in these AI Operator sessions and runs her own lunch and learn for the broader department, not because anyone asked her to, walking people through the same frameworks and adding a hands-on exercise where they build an agent themselves. It lands well. She comes back reporting a room full of good questions and half the team finding out they already had ChatGPT access they didn't know about. A sales rep explains, unprompted, how the outreach automation his department built has become the template other departments now want to copy. An engineer who no longer writes code with his fingers, because he talks to Claude instead, tries to explain why that isn't as strange as it sounds anymore.
The real measure of whether an AI transformation took hold is whether people have stopped needing you to hand them the next idea, not whether they can recite the framework you taught them.
It's also why the meeting opened the way it did, before a single slide went up. The first thing on the agenda was a real conversation about a company-wide AI usage policy, prompted by a legitimate concern someone had raised about a tool that trains on the data you feed it. Token costs are climbing. Departments are building tools without knowing the others exist. Nobody in that room needed convincing to try AI anymore. The open question had already moved past that, to how everyone keeps up with how much they're already using.
Over ten months, I watched a room go from needing convincing to needing to be managed. Most companies never get there, because most never get past the first conversation. This one did, department by department, meeting by meeting, for the better part of a year. The credit for that belongs to the people who did it. I just got to watch.