Somewhere this month, a CEO ends a meeting by handing somebody AI. This time it's you (congratulations, however that happened). Most new owners start by shopping for tools, and that's the wrong first move.
We've spent the past year asking companies the same four questions about who owns AI. The same pattern keeps showing up, often in the same words.
Those 90 days decide whether you own a goal or a plan with money behind it. The second compounds every quarter; the first stays where it started.
In our research that gap matters more than anyone's individual skill. Ownership does the work here, not enthusiasm.
Days 1-15: take a census
Sit down with fifteen people across teams and watch them work, screen share open, no survey. Fifteen shows you the pattern and still finishes inside two weeks.
Usage data answers most of this faster, once you have it. On day one you don't, and the export takes longer to arrange than the conversations do.
Watch for the specifics: somebody pastes a customer email into Claude, then hand-edits the reply for six minutes. Somebody else keeps a prompt in a note on their phone.
You'll probably find what we keep hearing. Leaders tell us "we didn't have a sense of who was doing what." Adoption is lopsided: a few people are great, some will tell you "I'm never opening Claude," and the middle is thinner than anyone expects.
Two weeks in, you'll start feeling like you should have announced a strategy by now. Spend the first two weeks counting instead. Who uses what, for which tasks, and where the clever work hides.
Resist announcing something in week one, a new tool or a kickoff meeting. Don't do it. The announcement ages badly the moment the census contradicts it.
Days 16-30: get a baseline you can defend
Now for the boring part: seats bought, seats opened, and seats used daily. Most companies measure exactly two things: seats bought, and whether usage went up. Your baseline only needs to beat that by one level.
Worth saying: if you run four assistants across three admin consoles, and most companies do by now, this takes a week of asking IT nicely rather than an afternoon in a spreadsheet.
Two more entries belong in it, and neither takes long. First, place your company on the maturity matrix by answering four questions in order, where the first no is your level:
- Owner. Does someone own AI adoption? That's you now, so this one's a yes.
- Plan. Is there a funded plan behind the goals?
- Team. Is a dedicated AI ops team in place?
- Wiring. Has that team wired the whole org?
Second, write down your real power users. The top of the token bill rarely matches that list.
You can't meet people where they are until you can find out where that is. An owner we talked to said that about their own rollout, and it's the whole job description at this stage. Everything after this depends on knowing what's actually happening, which is why the census comes before the strategy.
You'll be tempted to build a dashboard now. A first version is about a week of work and it's a fine month-two project. Not yet.
Days 31-60: turn goals into a funded plan
Most owners stall here, on the second of those four questions: what's the plan, and what's the budget. Answer it and everything downstream gets easier, because you stop asking permission one piece at a time.
A real plan stays small and concrete. It names two or three workflows where your power users already win, gives an owner real hours, carries a budget line, and moves one number.
Goals are not a plan. A plan has a budget, named workflows, and a number it moves. Bring that to whoever owns the money with the baseline behind it, and you're asking for an investment, with evidence attached.
Choose the number carefully
Pick the number before you pick anything else, and make it a business number. Time saved and messages sent count activity. They feel like progress, and they crumble in a room that owns a P&L.
A business number lives in a system your company already trusts: pipeline velocity at the stage where your deals stick, cost per ticket for your highest-volume type, days from signup to first value, renewals in one segment.
The number you choose is what separates hopeium from results. Choose one, get its owner to agree it's the right one, and let it narrow which workflows make the plan.
One tempting shortcut is the all-hands workshop. In our research those don't stick, because the people who volunteer to present are the people already using AI. Put the same hour into two power-user workflows and you'll have something to show.
Days 61-90: make one win repeatable, then report
Back to the census. Pick one thing a power user does that the rest of the team doesn't, and turn it into something everyone runs. This is where those fifteen conversations pay off, because you already know who has the best version of it.
Measure it before and after, even crudely. Hours per week and drafts per rep are fine as week-to-week proxies, as long as they connect to the business number you chose in month two. A proxy that doesn't won't survive the readout.
One repeatable win beats ten pilots. It proves the plan works, and leadership remembers that slide.
Then give the readout: the baseline you built, the win you made repeatable, and the ask for next quarter. Don't promise ROI you can't trace yet. The owners who keep their credibility claim only what the data supports.
The whole sequence, on one page
A good idea that arrives too early still costs you a quarter. Everything on the right earns its place, once the step before it lands.
| Phase | Do | Save for later |
|---|---|---|
| 1-15 · Survey the field | Talk with fifteen people, screen shares open. Write down who uses what, for which tasks. | Tool announcements, policies, kickoffs |
| 16-30 · Get usage data | Count seats bought vs. opened vs. daily. Answer the four questions: owner, plan, team, wiring. List your real power users. | The dashboard build |
| 31-60 · Fund the plan | Choose one business number (stage velocity, cost per ticket type, time-to-onboard, renewals). Pick 2-3 power-user workflows that move it. Put hours and a budget line against them. | The all-hands workshop |
| 61-90 · Prove one win | Make one power-user win repeatable. Measure before and after. Report: baseline, win, and ask. | ROI claims the data can't carry yet |
Why the order matters
Do the census first and the baseline almost writes itself, because you already know who to count. Do the baseline before the plan and the budget conversation starts from numbers you both trust. Run them in order.
None of it gets complicated. Every step hides a faster-looking alternative right next to it, and that alternative usually turns out to be the dashboard.
By month four you'll have pulled the same numbers by hand a few times, and you'll be reporting one of them on a schedule. Whoever keeps doing that pull is making a real choice about where their week goes.
Go find your fifteen people.