Buy Gemini Enterprise seats in waves, not all at once
The license decision usually gets made as one number for the whole company. It goes better as three, each one earned by the wave before it.
Most rollouts start with a headcount. Everyone gets a seat on the same day, the announcement goes out, and the license count matches the org chart. It is the simplest purchase to approve and the hardest one to make work.
Day one for everyone is the expensive way to start
On switch-on day, most of your people do not yet know what the assistant is for in their job. Some will find out quickly. Many will try it twice, get a generic answer, and stop. You pay for those seats every month while they wait for a reason to come back.
The cost is not only the invoice. Low usage in the first quarter gets read as "AI does not work here," and that verdict is much harder to reverse than a seat count.
Choose the first wave by the work, not the org chart
Start with roles where the work is already written down: proposals, contracts, policies, account notes, long email threads. That is where an assistant grounded in your own information earns its keep, because the answer depends on documents only your organization has.
Pick people who will use it daily and talk about it. A first wave of a few dozen committed users teaches you more than a few thousand curious ones.
Let each wave earn the next
Wave two goes to the teams next to wave one, once you know what good usage looks like and can show it. Wave three is everyone else, with training built from what the first two waves actually did rather than from a vendor demo.
Gate each wave on evidence, not on a calendar. If wave one is not using it, adding more seats will not fix that. Find out why first.
Get two things ready before anyone logs in
First, your sharing settings. An assistant grounded in your data answers from what each person can already open. If folders are shared more widely than anyone realized, the assistant will surface that before your next audit does. Tighten access first.
Second, an owner. Someone whose job includes adoption, who watches usage from week one and talks to the people who stopped. Without that person, nobody notices a stalled rollout until renewal.
Decide what success looks like before switch-on
Pick the measure before the first login: weekly active users in the wave, or the tasks you expect it to take over. Write it down. A number chosen after the fact will always look fine.
A seat nobody uses is worse than a seat nobody bought. Phasing the rollout is how you avoid buying the first kind.