How to Stop Paying for Unused AI Licences
An AI licence is bought per person and used per task. That gap is why many companies pay for seats nobody opens, and why cutting the least-used ones can switch off exactly what you should keep.
Carlos Andrés Ramírez ·
An AI licence is bought per person and used per task. That gap is why many companies pay for seats nobody opens, and why cutting the least-used ones can switch off exactly what you should keep.
Start with how they are bought. Most enterprise AI tools are sold by the seat: every person with access is a line on the invoice. They get bought in bulk, when leadership decides the company needs AI, and assigned by org chart: a whole function, a whole grade. Nobody asks, at that point, which specific task each seat is meant to serve.
Months later the renewal arrives, and finance or IT procurement asks the reasonable question: of all this, what gets used? The vendor dashboard answers with what it has, which is logins, sessions and messages. With that data the temptation is to cut the seats with the least activity. It is a good first step for finding money and a poor basis for deciding.
The symptom
How do you tell which AI licences nobody uses?
Some signals show up in any admin console and almost always point to money leaving with no return. None is enough alone, but together they draw the list of licences worth reviewing.
- People who have not logged in since the licence was assigned.
- Licences held by someone who changed role or left the company and were never released.
- Whole teams with only a few sessions a month, spread thinly across many people.
- Two tools doing the same job, bought by different functions, each with its own block of seats.
- Named licences that one person effectively shares with colleagues who have no access.
The problem underneath
Usage frequency measures habit, not value
Because frequency of use measures habit, not value. Picture an accounting analyst who opens the AI tool once a month, on close day, and saves a full day of reconciliations. On the dashboard that person looks inactive. Next to them, someone who opens it twenty times a day to rephrase emails looks like the model user, and their work has not changed. Cut by activity and you switch off the first licence and renew the second.
The underlying mechanism is that the seat was assigned before the task was defined. With no defined task, nobody can say what is lost by cutting, and that uncertainty weighs more than the saving. So many companies prefer to overpay than to risk removing something somebody needs. It is a rational decision given the information they have, and it is fixed with better information, not with more discipline.
A licence with no task attached cannot be cut with judgment or renewed with judgment. It can only be cut or renewed blind.
BECOME
The test
Renew, move or cut is decided by the task, not by the user
Three questions, in this order, for each licence or group of licences. The first clears the obvious cases, the second protects what is critical, and the third turns the saving into a decision.
- What task does it serve?
- Ask each function head to name the specific task the licence supports. If they cannot name one, the licence is a candidate to cut or reassign with little risk.
- What happens if it is missing that day?
- For the ones with a task, ask what happens if the tool is unavailable on the day it is needed. If the answer is a delay or a real cost, such as a close, an audit or a delivery, usage is low but the licence is critical and gets protected.
- Can it be shared or moved?
- If the task is occasional, one licence rotating among two or three people may be enough. If use is daily and growing, you may need more. The saving comes from matching the number of seats to the tasks, not from a flat cut.
This exercise takes days, not months, because it runs function by function with the person who knows the task. What changes is the renewal conversation: instead of negotiating a price per seat on an inherited number, the company arrives with a seat count it can defend one by one.
Before the next renewal, pull the list of inactive licences from the console and send it to each function head with one question: what task do you need it for? Whatever nobody claims within two weeks gets released. Whatever someone defends with a concrete task stays, with the name and the reason written down.
Frequently asked questions
How do we stop paying for AI licences nobody uses?
Tie each licence to a specific task and to an owner who can defend it. Those nobody claims within a reasonable window get released, and those that are defended get documented with their reason. The saving comes from matching the number of licences to real tasks, and the review repeats at every renewal so they do not pile up again.
Is it a good idea to cut the AI licences with the least activity?
As a first filter, yes. As a final decision, no. Frequency of use measures habit, not value: a licence used once a month at an accounting close can matter more than one opened twenty times a day. Review each candidate with the person who knows the task before cutting anything.
How often should AI licences be reviewed?
Before every renewal at a minimum, and lightly each quarter to release the seats of people who changed role or left. A small periodic review stops the build-up from growing into a large, risky cut on renewal day.
Who should decide which AI licences to keep?
The head of the function that uses the licence decides, supported by finance or IT procurement, who provide usage data and cost. The person who knows the task knows what is lost by cutting, and the person who owns the budget knows what it costs to keep. Neither decides well alone.
Let's review which AI licences are worth renewing
From the idea to the operation
Adoption is not communicated: it is designed with the teams who will operate the capability, and measured against a baseline.
About the author
Carlos Andrés Ramírez — Transformation Director
Specialist in business transformation and reinvention. Director of Specialised Programmes and lecturer in Artificial Intelligence at UPC's Graduate School.