Guide
How to cut wasted AI seats and reduce AI cost
Connect a spend source and a usage source, then let Spendassay audit the two together. It surfaces and prices unused seats, underused seats, tool overlap, and renewal opportunities, each with a proof level and a visible formula. Review the dollar-priced findings, then act through a set of step-by-step instructions your own admins run to reclaim the seats.
What counts as a wasted AI seat?
A wasted AI seat is a paid license nobody uses, or one used far below what it costs. Spendassay flags two types: unused seats with no activity in the trailing period, and underused seats whose usage does not justify the price. Tool overlap counts too, when two paid tools do the same job.
Waste hides across vendors. A Copilot seat sits idle after someone changes teams. A Cursor license renews for a contractor who left. Two teams pay for ChatGPT and Claude to do the same work.
Spendassay names each type with a dollar figure and a proof level:
Unused seats: paid licenses with no measured activity in the trailing window.
Underused seats: activity far below the seat's cost, priced against a peer cohort.
Tool overlap: duplicate paid tools serving the same function.
Every finding shows its proof level (counted, compared, or estimated) and exposes the formula behind the number, so finance can check the math.
How does Spendassay find wasted AI seats?
Spendassay audits two data sources together: a spend source that shows what you pay, and a usage source that shows what gets used. It matches seats to real activity, then prices the gap. The audit is read-only and metadata-only. It never reads source code, prompts, or message content.
Connect a spend source first. This can be the vendor's own billing, a corporate-card feed through Ramp, or a CSV export. It answers what you are paying for.
Connect a usage source next. GitHub, Cursor, Okta, and Google Workspace show who actually uses each tool. API token spend from OpenAI and Anthropic covers metered use. CSV import works when you have no live integration.
Spendassay joins the two. A seat you pay for with no matching activity becomes an unused-seat finding. A seat with thin activity becomes an underused-seat finding, benchmarked against a peer cohort.
The audit stays buyer-side. It grades no vendor's own homework, reads only metadata, and works at the team level. It never builds individual league tables.
How do you act on the findings to reduce AI cost?
Open the findings list, sort by recoverable dollars, and confirm the seats you want to reclaim. Each finding links to a set of step-by-step instructions: the affected seats, the admin-console steps, and drafted internal comms. Your own admins run the step-by-step instructions in the vendor tools. The audit tracks recovery but never writes to those systems.
Findings are ranked by dollars, so you work the largest waste first. The proof level and formula sit next to each number for sign-off.
For seat reclamation, the step-by-step instructions list the exact licenses, the steps to remove or downgrade them in the vendor admin console, and a draft note to the affected team.
For renewals, Spendassay builds a negotiation package: current spend, a cited public benchmark, the math for cutting to the size you need, and drafted asks for the vendor account manager.
Spendassay's audit is read-only. It never writes to, routes, or throttles vendor systems. Your admins execute every change; the audit confirms the seat closed on the next sync and banks it to a Recovered statement.
How much AI spend can you expect to recover?
For a typical mid-market team, Spendassay surfaces 10 to 15 percent of seat and subscription spend as recoverable within 30 days. The exact figure depends on how many idle seats, underused licenses, and overlapping tools the audit finds. Every recoverable dollar is priced with its proof level so the target is defensible to finance.
The 10 to 15 percent target is a starting range for a first audit, not a guarantee. Teams with recent headcount changes or several overlapping tools usually sit at the higher end.
The recoverable number is a sum of individual findings, each with its own dollar value and formula. You can trace the target back to the specific seats behind it.
Recovery gets tracked, not assumed. A finding only counts as banked once the seat actually closes on a later sync, so the Recovered statement reflects real savings your admins delivered.
How do you keep AI seat costs down over time?
Leave the connectors in place and re-run the audit on each sync. New idle seats surface as they appear. Renewals get flagged before they lock in, and quality-trend findings show when a downgrade hurt output. The Recovered statement gives finance a running record of what the audit put back into the AI cost report.
Seat waste is recurring, not one-time. People change teams, contracts auto-renew, and new tools get bought without review. A standing audit catches each as it happens.
Renewal opportunities surface ahead of the renewal date, so you negotiate with usage evidence and a benchmark in hand rather than after the fact.
Quality-trend findings watch for regressions, so a seat you cut or a tool you consolidated shows up if it dents team output.
The result is a CFO-credible AI cost report: what you spend, what gets used, what was recoverable, and what your admins actually banked.
Step by step
- 1
Connect a spend source
Link what you pay for. Connect vendor billing, a corporate-card feed through Ramp, or upload a CSV of your AI subscriptions and token invoices. This gives the audit the cost side of every seat and tool.
- 2
Connect a usage source
Link what actually gets used. Connect GitHub, Cursor, Okta, or Google Workspace for seat activity, and OpenAI or Anthropic for API token use. CSV import works when you have no live integration. The audit reads metadata only, never code or message content.
- 3
Review the dollar-priced findings
Open the findings list and sort by recoverable dollars. Review unused seats, underused seats, tool overlap, and renewal opportunities. Check the proof level and formula on each finding, then confirm the seats you want to reclaim.
- 4
Act through the step-by-step instructions
Open the step-by-step instructions for each confirmed finding. It lists the affected seats, the admin-console steps, and drafted comms. Your own admins remove or downgrade the licenses in the vendor tools, or send the renewal negotiation package. The audit never writes to vendor systems.
- 5
Track the banked recovery
Let the next sync confirm each seat closed. Spendassay marks the finding banked and adds it to the Recovered statement, giving finance a running record of AI spend put back into the AI cost report.
Common questions
Does Spendassay cancel or downgrade the seats for me?
No. Spendassay's audit is read-only and metadata-only. It finds and prices the waste and gives you a set of step-by-step instructions, but your own admins execute every change in the vendor tools. It never writes to, routes, or throttles vendor systems.
Can I use it without connecting live integrations?
Yes. CSV import works with zero live integrations. Upload a spend export and a usage export, and the audit matches seats to activity the same way. Live connectors like GitHub, Ramp, and Okta just keep the audit current automatically.
How does Spendassay know a seat is really wasted?
It joins your spend source with your usage source, then prices the gap. Each finding carries a proof level, counted, compared, or estimated, and exposes its formula, so a paid seat with no matching activity is flagged with the exact dollar value and the math behind it.
Does it rank or monitor individual employees?
No. Spendassay reports at the team level and never builds individual league tables. It reads metadata about seats and usage, not source code, prompts, or message content, so you can audit AI spend without surveilling people.
Run the audit on your own AI spend
Connect read-only sources and get a CFO-credible AI cost report. Free to start, about 10 minutes to connect.