Why nobody uses the AI seats you bought

62 seats, four people a day

62 of us had a seat, and four people used one on a normal working day.

That number on its own reads as a rollout nobody wanted, and I had two explanations ready before I looked. I checked both, and both were wrong.

Those rows are cumulative rather than separate groups, so the 48 contains the 39 and the 39 contains the four. 48 of the 62 had opened something within the month, and six had never opened anything at all. That is not refusal. Most people had the tool, went in, and came out again.

Session length says what they were doing in there. For most users a session ran about one message: a question in, an answer out, back to the work. Two people worked in longer sequences with follow-ups, and two out of 62 is not a habit.

The connectors were wired and the tool was approved

I was sure the connectors to our own systems were not hooked up. They were. I was sure one of the tools had never been approved for use. It had been.

Both assumptions would have made this a rollout problem, and both fell over on inspection. What was left is the harder explanation: people could see what the tool did and could not see what it was for in their own week.

25 of the 62 seats were not sales or marketing at all

Roughly 25 of the 62 seats belonged to product, so 62 is not the number of GTM people with a seat.

That matters before anybody reports a figure per function. A judgment about our marketing team needs the marketing seats, and the analytics were organization wide. Quoting four of 62 to a CMO without that split hands them a mixed number and lets them read it as theirs.

I also had to withdraw my own first reading of one team. The marketing capabilities looked unused, the population was four or five people, and some of them had only just been released. That produced a rule I now apply everywhere: no adoption judgment before 30 days of availability, and a documented release date per capability, which almost nobody keeps.

Counting people who can use it says nothing about what runs

The reflex behind this whole exercise was to count the people who can work this way and read the total as organizational maturity.

Those are two different quantities. A capable person with no process to start gets you a capable person, working faster on the parts a chat window can reach. Nothing about the way the work runs changes, and nothing survives that person going on holiday.

Four people out of 62 on a normal day, and a session that is one message long. What that counts is the number of processes there are to start.

For us the answer was none.

From six months inside a B2B GTM organization with more than 60 AI seats and 36,000 accounts in its CRM.

Scoring the organization on its weakest gate

The version that held up scores the organization on seven dimensions and takes the lowest score rather than the average.

Mandate, access, skills and roles, work redesign, data foundation, governance, measurement. A good average with one of those at zero describes an organization that cannot run anything, so the average flatters and the lowest one is the answer. That run landed one step above the bottom of a five-step ladder. Access and skills were not what held it there, because the connectors were wired and the tool was authorized.

One governance detail is worth carrying, because it was the only surprise on the risk side. Legal approval was in place, the works council position was open, and our tooling made 442 writes to company systems in 30 days. Read only was no longer the whole description, and nobody had updated it.

What to do instead

1. Read the distribution, not the daily number. Daily, weekly, monthly and never are four different problems with four different answers, and a single figure hides which one you have.
2. Read session length. One message per session means the tool is being used as a search box. That is real value, and it is not a process.
3. Split the seats by function before anybody reports a number. A mixed denominator turns into a wrong judgment about a specific team.
4. Count the processes that start without a person thinking of them. In most organizations the honest answer is zero, and that number explains the seat usage better than any survey of attitudes.
5. Score the organization on its weakest dimension. One gate at zero is what the organization is, whatever the other six say.
6. Check what the tooling actually writes before calling anything read only. The description and the behaviour drift apart quietly, and the governance conversation runs on the description.

The seat usage and the silent process are the same story from two ends. When something does start on its own, it can stop and still look healthy, and a campaign whose follow-up was switched on and never sent is the same failure with a marketing budget attached.