Every AI subscription your company pays for renews on schedule, whether anyone used it or not. Seats are the clearest example: a fixed price per person, per month, per tool. They accumulate quietly, and they almost never get cleaned up.
This makes seats different from every other line on the AI bill. Usage-based spend at least tracks work - when nobody asks the model anything, the meter stops. A seat is the opposite: the meter runs whether or not anyone shows up. That is what makes idle seats both the dumbest kind of waste and the easiest to fix.
Where idle seats come from
- Trials that became subscriptions. Someone tested a tool with five seats. Two people use it. The other three renew monthly.
- People who left. Offboarding checklists cover email and laptops. AI platforms rarely make the list.
- Tool hoppers. A team switched from one assistant to another. The old seats were never cancelled.
- Just-in-case seats. Bought for a project that ended, kept for a project that never started.
None of these are anyone's fault. Seats get added in thirty seconds on a busy day, usually for a good reason. Removing one requires somebody to notice, check, and ask - and no calendar reminds them to. That asymmetry does all the damage: adding is instant, removing is a chore, so the count only ever moves in one direction.
The math is not small
AI seats run $20-60 per month each. Six idle seats at $40 is $2,880 a year - for nothing. In a 100-person company using three or four AI platforms, finding 10-20% of seats idle is completely normal. This is usually the fastest 5-15% anyone cuts from an AI bill, because it requires no behavior change at all.
A worked example: the 100-person company
Put the numbers above together and the invoice writes itself. Take the 100-person company from earlier, running three AI platforms. Not everyone has a seat in every tool - say 70 paid seats in total, at the $40 midpoint of the $20-60 range. That is $2,800 a month in seat licenses, $33,600 a year.
Now apply the typical finding of 10-20% idle: 7 to 14 seats. At $40 each, the company pays $280-560 a month - $3,360-6,720 a year - for software nobody opens. Split the difference and call it ten seats: $400 a month, $4,800 a year, recovered by sending a handful of emails.
The other half of the example is time. Idle seats do not bill once - they renew. A seat that went idle in March and gets caught in next January's budget review has billed ten times for nothing. Catching the same seat in April costs one wasted month instead of ten. That is the whole argument for a monthly check rather than an annual one.
Notice also what the cleanup does not cost. Nobody loses a tool they use. No workflow changes. No policy gets written. That is why idle seats come first in the 40% playbook: it is the one fix where the saving is real on day one and the pushback is near zero.
How to find them
Every major AI platform has an admin page with per-seat activity. Look for last-active dates. The manual routine:
- Once a month, list every seat across every AI platform you pay for.
- Flag anything inactive for 30+ days.
- Ask the owner one question: keep it or lose it?
- Reassign before you cancel - someone else may be paying for a personal subscription out of pocket.
Do the whole pass in one sitting if you can. Partial audits die quietly: two tools get checked, the third holds the worst offenders, and the browser tab closes for good. One afternoon, every tool, one list.
How to check this in your own company this week
The monthly routine above is the maintenance plan. If you have never audited seats at all, the first pass fits in one afternoon:
- List every tool. Pull the last two months of card statements and invoices and write down each AI product billed per seat. Most companies find at least one they forgot they were paying for.
- Export the members. Open each tool's admin page and pull the member list with last-active dates. If you run ChatGPT Enterprise or Microsoft Copilot, our guides show exactly where those numbers live.
- Flag 30+ days. Sort by last-active date, count everything idle for a month or more, and multiply by the seat price. That figure is your headline.
- Send the one-question email. "You have not used this in six weeks - keep it or lose it?" Give it a one-week deadline and treat silence as a no.
- Fix offboarding. Add one line to the leaver checklist - "remove AI seats" - and the biggest source of idle seats closes for good.
Write the result down, even if it is embarrassing. The number from this first pass becomes the baseline that proves the cleanup worked - and the argument for keeping the check monthly instead of letting the seats grow back.
"What if they need it back next month?"
This is the objection that keeps idle seats alive. Nobody wants to be the person who took a tool away from a colleague. Three answers take the sting out.
First, releasing a seat is not deleting an account. Every major tool lets an admin re-add a member in minutes, so the worst case of being wrong is a short wait, not a lost capability.
Second, the 30-day bar is generous. Someone who has not opened a tool in a month is not depending on it. And when a supposedly heavy user shows up in the flagged list, it usually means they quietly moved to a different tool - which is worth knowing anyway.
Third, reassign before you cancel, exactly as the routine above says. Freed seats often have a waiting list: the colleague expensing a personal subscription, the new hire whose request is stuck in procurement. A seat that moves to someone who uses it stops being waste without shrinking anyone's toolkit.
And if an owner pushes back with a real reason, perfect - that seat is not idle, it is just quiet. The point of the audit is not to take tools away from people. It is to stop paying for the ones nobody wants.
Or let the software nag you
The reason idle seats survive is that nobody owns the monthly check. Optimize watches seat activity across all your tools and flags idle ones automatically, with the dollar amount attached. The awkward cleanup conversation becomes a one-click approval.
Connecting takes about 15 minutes and is read-only - Optimize never cancels anything on its own. The first report lands within 48 hours with idle seats already flagged, and most companies make their first cut inside two weeks. Seats are usually the fastest 5-15%. The bigger cuts come from putting the right model on each job, and they take a little longer.