Here's a situation I see a lot. A company buys Microsoft Copilot or a Claude enterprise plan for hundreds of seats. IT rolls it out. There's a kickoff email, maybe a lunch-and-learn. Three months later someone pulls the usage report, and almost nobody is opening the tools. Leadership asks the question I keep hearing: why did we buy this if people aren't using it?
I think the answer starts with what we mean by adoption. A lot of people picture adoption as every employee becoming an AI expert, writing clever prompts all day, and reinventing their job around a chatbot. That picture sets the wrong target, and chasing it burns money and morale. Adoption that sticks looks more like AI showing up inside the workflows and culture so people keep doing their jobs, with better information and less grind, without having to retrain on a skill set they haven't practiced in decades.
My work at Spej sits right in that gap. The companies that get past flat usage usually run two lanes at once: equip the people doing the work, and redesign the revenue workflows so AI lives in the system instead of only in a few personal chat windows.
Handing out seats rarely moves usage
The bottom-up move is familiar. Give people doing the work a license, encourage them to build skills and small agents other teams can reuse, and hope usage spreads from the ground up. That instinct is solid. The people closest to the work often know where the friction is better than anyone in a strategy deck.
If you only hand out tools and walk away, though, usage often stays extremely low. Sometimes the pushback is loud. Sometimes people just never log in. Either way, the spend sits on the books and the dashboard stays flat. For most of those teams, access was never the hard part. Time to experiment, clarity about when the tool helps, and a path that fits the day they already have usually were.
A few power users don't make the company AI-native
Even when bottom-up works well for individuals, it tends to stay stuck in point solutions. Someone in finance gets fast at summarizing reports. Someone in ops builds a personal helper for email. Those people get more productive, and that helps. What you usually don't get is a company-wide shift in how work runs. The gains stay isolated to how those employees work, while the processes that bring in revenue look the same as they did before the licenses arrived.
That gap explains a lot of the ROI frustration I hear. Leaders expected an operating change. What the purchase mostly delivered was personal productivity for the people who already liked experimenting.
Put AI in the workflow, and keep the job recognizable
The other lane is what we mean by top-down at Spej. Look at the outputs that actually bring in revenue, and re-engineer those workflows so AI runs inside the shared process across the company. Personal tools still help. The bigger operating gains show up when the workflow itself carries AI.
The version that fails is the mandate. Leaders hear "top down," picture a forced new way of working, and roll out a system that asks everyone to relearn their job at once. People resist, morale drops, and the unofficial workarounds come back.
The version that works leaves the job familiar. The person in the seat still does work they recognize. Tedious steps get absorbed into the system. Judgment steps get better inputs earlier. You integrate AI so the day feels like the same job with richer information, and you avoid asking every employee to become a prompt engineer overnight. That is also how you protect morale. Making AI a second career on top of the one they already have is a fast way to lose people.
Leaders should own the redesign
One of the patterns I push back on is dumping the burden of business transformation onto the whole staff. Training and equipping people still matter. Leaders also have to own how work flows, especially the non-technical CEOs, COOs, and CFOs who understand the business at a high level. Handing the whole problem to "the AI guys" and hoping the P&L moves rarely works.
When companies try only one lane, either seats and training without process redesign, or process redesign without caring how people actually work, they end up with disconnected approaches. I see a lot of missing-ROI frustration in that split. Both lanes together is what tends to stick: train and equip people, and redesign the revenue workflows so AI is in the system while the job still feels like their job. Part of what we try to do at Spej is help companies get there without turning the entire workforce into AI experts.
What to do this week
Pull the usage report for whatever seats you bought. Then pick one workflow that touches revenue or cost and sit with the people who run it. Ask where the day slows down, where they invent unofficial shortcuts, and what would have to be true for an AI-assisted version of that work to feel easier than what they do now. Write down what "good" looks like for that workflow before you buy another round of licenses or schedule another all-hands demo.
If usage is flat, treat that as information about how the work is designed, how much time people have to experiment, and whether leadership owns the process change. Flat usage rarely means your team can't learn AI.
If you would like help connecting the seats you already bought to workflows that actually change, that is the work we do at Spej. Talk with the Spej team.
