A subscription audit finds the tools nobody opens anymore, the ones worth cancelling outright. Idle AI spend is a quieter, harder-to-spot cousin of that problem: capacity inside tools you genuinely keep and use, seats nobody has logged into in months, or a committed-use pricing tier sized for volume you never actually reached, sitting there fully paid for and doing nothing. It survives a normal subscription audit precisely because the tool itself is legitimate and in active use, just not at the scale you are paying for.
How idle capacity hides inside a tool you actually use
A 25-seat Claude Team plan where only 16 people log in regularly is not a wasted subscription in the way an unused tool is, it is a genuinely useful tool sized wrong. A committed-use API tier purchased for a projected volume that never materialised is the same pattern in a different form: real value, wrong size. Both survive a 'do we still use this' audit because the answer is yes, while the more useful question, 'are we using all of what we're paying for', never gets asked.
Pull a per-user login report for any seat-based AI subscription and compare active users against paid seats.
Check committed-use or volume-tier pricing against actual usage for the last three months, not the projection used to justify the original tier.
Flag any seat unused for 60 days as a candidate for downgrade or reallocation, not automatic cancellation, since roles change.
Revisit sizing decisions quarterly rather than only at annual renewal, since team composition and usage patterns shift faster than most contracts do.
A worked example
A Perth engineering firm running a 30-seat Claude Team subscription found, on pulling the usage report for the first time in over a year, that only 21 seats had logged in during the previous quarter. The remaining nine belonged to staff who had left, moved to roles that no longer used the tool, or had simply never adopted it after initial rollout. Downsizing to 22 seats, with a small buffer for onboarding, cut the monthly bill by roughly $270 without affecting a single person who was actually using the tool. The unused capacity had been sitting on the invoice, unnoticed, for the better part of a year.
Committed-use tiers carry the same risk in a different shape
A business that negotiates a committed-use pricing tier for API access, locking in a volume discount against projected usage, takes on the risk that projected usage does not materialise. If actual usage lands well under the committed volume, the business is paying the discounted rate on capacity it never draws down, which can work out more expensive than standard pay-as-you-go pricing would have been at the lower actual volume. Reviewing committed-use tiers against real usage each quarter, rather than assuming the original projection still holds, catches this before a full annual commitment locks the mismatch in.
Why this is a different fix to a subscription audit
A standard subscription audit asks whether a tool is still used and cancels the ones that are not. Idle spend needs a second, more granular question asked of every tool that survives that first cut: is the size of what we're paying for still the right size for how we actually use it. That second question catches money a simple cancel-unused-tools exercise never finds, because the tool in question is genuinely still in active, valuable use, just oversized for the team actually using it.
Reallocating instead of always cancelling
Not every idle seat should be cancelled outright. A seat sitting unused because a role is currently vacant, or because a new starter has not yet been onboarded, is a timing issue rather than genuine waste, and cancelling it only to repurchase it a month later costs more in admin overhead than it saves. The useful discipline is distinguishing genuinely idle capacity, unused for months with no clear reason, from temporarily idle capacity tied to a specific, known cause, and only acting on the first category immediately.
Building the habit without a big project
Most Australian small businesses can run this check in an afternoon for their two or three largest AI subscriptions: pull the usage report, compare against paid capacity, and right-size where the gap is meaningful. Doing this once a quarter, alongside the usual subscription review, catches idle spend while it is still a modest amount rather than letting it compound silently across a full contract year. For a business that has never checked, the first pass is usually the one that finds the most.
