For procurement and FinOps — the team that watches cloud and software spend
Bring evidence to the renewal, not a target
We count which AI licenses went unopened, which vendors overlap, and which contracts renew next. Your ask becomes a dollar figure with a source, not a percentage you hope they meet.
Free Snapshot, no card. We can only read. Nothing writes back to your systems.
"We would like a better price" is a position. "Forty-eight percent of the seats on this contract went unopened for 60 days, here are the counts by month, and we are cutting the commitment to match" is evidence. Vendors negotiate the first. They concede to the second.
We produce the second. A seat is a paid license for one person, and an idle seat is one nobody has opened in 60 days (the definition, and how we count it). We read seat activity per vendor straight out of the admin consoles, plus where two tools do the same job, usage charges nobody planned for, and the date each contract comes up.
Counted, not estimated. A vendor rep can check it against their own console in a minute.
Seeing spend by vendor is the minimum. The question that recovers money is which seats bill without being used — across every vendor at once.
The renewal calendar, and what you can already claim
Evidence goes stale and renewals are scheduled. Miss the window and a 12-month commitment renews itself at the number you were trying to change.
We keep a live calendar of every AI contract you hold: which vendor, which date, how many seats you committed to, how many went unused in the last 60 days, and how many you can cut. Cutting seats is not the same as cutting price. You pay for fewer licenses, at the same price each. Where a contract allows that at all, it usually allows it only on the renewal date. So the date belongs on a calendar, not in someone's inbox.
Ninety days out, the finding is already sitting there: this contract, this many idle seats, this much money, this deadline.
How to negotiate AI renewals from evidence · The hidden cost of idle AI seats
Cutting seats and cutting price are two asks. Blend them and you pay for it.
Cutting seats reduces the number of licenses. A discount reduces the price of each one. Separate levers — and a blended ask lets the vendor answer the expensive one with the cheap one.
Here is the math on 100 Cursor Business seats at $40, where only 60 are in real use.
Same 20%, different order, $15,360 a year apart. Set the seat count with evidence before price enters the conversation.
Free template
The evidence pack you send the vendor
A one-page structure for the document that does the work. Seats committed against seats used, with the dates you counted. The idle count in dollars. Overlap with other tools. The seats you want to cut, the discount you want, and the date you need an answer by. Fill it in by hand, or let us fill it in for you.
No email wall. If it is useful without us, take it.
This sits next to Vendr and Tropic, not instead of them
Benchmark platforms have something we do not: the prices other companies actually paid. Vendr's marketplace lists a median Cursor contract of $111,408 a year across 76 documented purchases, ranging from $19,200 to $395,040. Buyers typically get 15–25% off by committing for a year, and 25–40% when they combine levers. Worth having. But read it for what it is: those buyers all chose to hire a negotiation service, so they are not the whole market.
What price data cannot tell you is how many of those seats your company opened last month. A benchmark gives the going rate. Seat use gives the count. A credible ask needs both.
"Comparable companies pay $X per seat, and we need Y fewer seats than we committed to."
Bring the benchmark for the price. Bring us for the count.
What it costs, per renewal
Observe is $16 per engineer per month, per engineer covered, never per employee. At 200 engineers: $3,200 a month, $38,400 a year.
Take one contract at Vendr's median Cursor value of $111,408 a year. Cut the seat count by 20% and it drops to $89,126. Take 15% off that smaller commitment and it drops to $75,757. That is about $35,650 recovered from one renewal, and the seat cut keeps paying every year after.
One renewal like that nearly covers the year; a second covers it comfortably. Most companies with 200 engineers have more than two AI contracts.
What you walk away with
A renewal calendar with a dollar figure on each date. A per-contract evidence pack you can send without rewriting it. Every number we show carries a proof level, so you know how much guessing sits behind it.
Renewal findings are counted. That is the point.
Free account, no card. Connect billing and one vendor console, and the first renewal finding lands the same day.
Questions procurement asks
Will the vendor dispute the seat numbers?
They can check them. Seat activity comes from the vendor's own console, so the source is their system. Disputes tend to be about definitions: "active" meaning opened, versus logged in, versus billed. That is why the dates and the definition are printed on the evidence pack.
What if the contract does not let us cut seats?
Then the evidence goes to price and scope instead: a lower seat count next term, a shorter commitment, or caps on usage charges. It still moves a lever, just a different one.
Does this cover usage charges, or only seats?
Both. Seats, API tokens and infrastructure land in the same view. A token is the unit AI vendors bill for — roughly a few characters of text. See the AI spend management guide.
What connects to what?
GitHub, Anthropic, OpenAI, Cursor, Ramp, Okta and Google Workspace, plus CSV import. Every connector reads and none of them write.
How far ahead do I need to start?
Sixty to ninety days before the renewal date. Enough history to make the idle seat count unarguable, and enough runway to use it.
Know what the next renewal is worth before the reminder email arrives
Pull the calendar, see which contract is closest, and find out what you can already stop paying for.
Snapshot is free, no card. Observe $16 per engineer per month. Control $40 per engineer per month.