Back to Blog

Before You Renew AI Licences, Count the Seats That Work

Active is not working. A four-week seat count, a worked hypothetical renewal and one action for Monday.

An old painted wooden key board on a deep blue wall, rows of hooks each holding a key. Most keys are dull and untouched; a few are polished bright and worn smooth by use, and one worn key in the foreground catches the light. Headline: Active is not working. Credit: Dr. Jonah Tebaa.
Direct answer

What should you count before renewing AI licences?

Before renewing AI licences, Dr. Jonah Tebaa advises organizations to use the Working-Seat Count framework rather than vendor login dashboards. Instead of accepting discounts on full seat tiers, firms tally actual task completions over four weeks, classifying seats into working, occasional, or idle bands. In a hypothetical illustration of forty seats, dropping twelve idle seats outperformed a fifteen percent vendor renewal discount, lowering the cost per working seat from $109.09 to $76.36.

The renewal notice said 40 seats, and the vendor's page said 28 were active. A 15 percent discount was on offer if the owner signed for twelve months. Before signing, he asked each team lead one question: how many times did you do the work this tool is for, in the last four weeks?

Everything in this article is a hypothetical illustration, with invented numbers. The firm, the owner and the vendor are made up, and none of it is a client story. The method is the one I use when an owner asks me whether an AI renewal is worth signing as offered.

What "active" counts, and what it leaves out

A renewal conversation usually starts from the vendor's number. If your dashboard counts logins, "active" means a person opened the tool at some point in the period. It does not say whether they did the thing you bought the tool for. A person can log in every morning, glance at a suggestion, and close the window. A login-based count can flatter a bill, and I would not assume that yours does or does not. I would simply not sign on it.

What you pay for is a seat that does the task. That task is different in every firm: a client summary drafted, a meeting note filed, a first-pass design produced. So the first job is to name it, in one line, for each tool. Once it is named, it can be counted, and a count is something you can divide into a bill.

The discount is the second thing worth looking at slowly. A discount lowers the price per seat. It does nothing to the share of seats that work. If three in ten seats do the work, a 15 percent discount gives you a cheaper bill for the same three in ten.

The Working-Seat Count in five steps

This is my working method, not research. It takes a few hours if the figures can be had, and the order matters.

  1. Define the working task in one line per tool. For example: "a client summary drafted" or "a meeting note filed". The count is of that task, not of logins or messages.
  2. Collect four weeks of per-seat counts. Ask the vendor for per-user task or usage figures, or have each team lead tally from their own team. If a seat cannot be counted, say so. An uncountable seat is not yet a working seat.
  3. Set the line before you look. Working means at least 8 times in four weeks, about twice a week. Occasional means 1 to 7. Idle means 0. You may choose a different line for a task that is monthly by nature, but fix it first, so the result does not bend your choice.
  4. Divide. Cost per working seat equals the monthly bill divided by the working seats. Also record the working share: working seats divided by total seats.
  5. Decide per band. Idle seats go. Occasional seats move to a shared pool or a lower tier, or get a named reason to stay. Working seats stay, and they are the number you negotiate from. Test any discount against the smaller order before you accept it.

A worked example: 40 seats, one renewal

Every figure below is a hypothetical number for illustration. A services firm holds 40 AI seats at an assumed $30 per seat per month. The bill is 40 x 30 = $1,200 a month. The vendor offers 15 percent off for a twelve-month commitment on all 40 seats: 1,200 x 0.85 = $1,020 a month.

The vendor's page shows 28 of 40 seats active, which is 28 / 40 = 70%. The team leads' four-week counts say something different:

BandSeatsShare of 40Monthly cost at $30Assumed value
Working (8+ times)1127.5%$330$1,540
Occasional (1-7 times)1742.5%$510$680
Idle (0)1230%$360$0
Total40100%$1,200$2,220

The checks: 11 + 17 + 12 = 40. The dashboard's 28 "active" seats are the working and occasional seats together, 11 + 17 = 28. The working share is 11 / 40 = 27.5%. Cost per working seat today is 1,200 / 11 = $109.09, against a $30 sticker price per seat.

The value column is an assumption, not a measurement. I have put $140 a month on each working seat, $40 on each occasional seat and $0 on each idle seat, all net of checking. That gives 11 x 140 = 1,540 and 17 x 40 = 680, so 1,540 + 680 = $2,220 a month. Three options, per month:

  • A. Renew all 40 at $30. Cost $1,200. Net value 2,220 - 1,200 = $1,020.
  • B. Take the discount, 40 seats at $1,020. Net value 2,220 - 1,020 = $1,200. Cost per working seat 1,020 / 11 = $92.73.
  • C. Drop the 12 idle seats and keep 28 at $30. Cost 28 x 30 = $840. Net value 2,220 - 840 = $1,380. Cost per working seat 840 / 11 = $76.36.

The discount does help against renewing as is: it is better than A by 1,200 - 1,020 = $180 a month. But C beats B by 1,380 - 1,200 = $180 a month as well (and 1,020 - 840 = 180 on the cost side). Over a year that is 180 x 12 = $2,160. Put another way, the 12 idle seats cost 12 x 30 = $360 a month, or 360 x 12 = $4,320 a year under option A. The cheaper-looking offer costs more than the smaller order.

Notice the other thing the twelve-month commitment does. It locks the idle seats in for the year. A discount that is conditional on the full count is, in effect, a price for keeping seats nobody uses.

When the discount is the wrong lever

The discount is the wrong lever when the problem is the count, not the rate. In the example, no price per seat can fix a working share of 27.5%. Only the order size can.

The 17 occasional seats are the honest complication. They are not worthless. In my assumptions each is worth $40 a month and costs $30, so on paper they pay for themselves a little. They cost 17 x 30 = $510 a month for $680 of assumed value. The owner may keep them, move them into a smaller shared pool, or ask the vendor for a lower tier. That is a separate negotiation, and it should start from a known number of working seats, not from the vendor's 28.

Two cautions. First, the method does not guarantee a saving. If the work is real but nobody has been told to use the tool, the answer may be training and not cancellation. Second, prices, minimums and notice periods vary by vendor and contract. Read the clauses on minimum seats, commitment term and notice before you act on any count. The 40 seats, the $30, the 15 percent, the $140 and $40, and the 11, 17 and 12 split are assumptions to show the method, not research.

I wrote earlier about choosing the first task to automate by how often it repeats. That is a decision made before you buy. This one is made after, on the seats you already pay for.

What to do on Monday

Pull four weeks of per-seat task counts before your next renewal date. If the vendor cannot give them, ask each team lead to tally. Sort the seats into Working, Occasional and Idle against a line you set before you looked. Divide the bill by the working seats, and write that one number at the top of the renewal memo, above the vendor's figure and above the discount.

Then price the discount against the smaller order, and decide which of the two you would rather defend. If you want a second pair of eyes on the count, write to me.

Related evidence: Wikipedia defines shelfware as software that is paid for but goes unused, including money spent licensing more seats than are needed, which it notes cuts into the bottom line. (Wikipedia's entry on shelfware, software paid for and left unused)

Zylo's 2026 SaaS Management Index reports that, even as waste reduction improves, organizations continue to uncover unused seats, overlapping tools and feature tiers that add cost without clear value, and that embedded AI makes usage harder to predict. (Zylo's 2026 SaaS Management Index on unused software seats)

Frequently asked questions

Is a login not good enough to call a seat active?

A login shows access, not work. If your dashboard counts logins, a seat can be "active" while the task you bought it for is never done. I count the paid-for task instead, one line per tool, such as a client summary drafted or a meeting note filed, and I divide the bill by the seats that did it.

Why four weeks and eight uses?

Four weeks is short and recent enough to reflect how people work now, yet long enough to cover a month's rhythm. Eight uses is about twice a week. That line is my working choice, not research. Change it if your task is monthly by nature, but fix it before you look at the numbers.

What if I cannot get per-person numbers from the vendor?

Ask each team lead to tally the task for their own people over four weeks. A simple list is enough. If a seat still cannot be counted, treat it as unproven, not as working, and shrink the order accordingly. You can add seats back later once someone shows what they do with them.

Should I cut every occasional seat?

No. Occasional seats still deliver some value. Sort them instead: a shared pool, a lower tier, or a named reason to stay. In my hypothetical, 17 occasional seats cost $510 a month for an assumed $680 of value, so the decision is a separate negotiation, not an automatic cut.

Written by Brian, Dr. Jonah Tebaa's AI partner, on his behalf.