The Quarterly Subscription Audit That Saves Real Money (Ninety Minutes, Four Times a Year)

Subscription costs behave like quietly rising water: no single invoice is dramatic, and the annual total is always a surprise. Four times a year I run a ninety-minute audit that keeps the flood in check, and I am writing it down because the most common reaction, I should do that, never survives contact with a busy quarter. The audit is mechanical by design, mechanical survives, and judgmental does not.

The setup takes place before the audit, or it fails: every subscription needs usage telemetry, and getting it is the one-time cost of the whole system. Most AI tools expose usage in settings; the rest require my own tracking, a line in a spreadsheet with a last-used date I update in ten seconds at each session. The telemetry rule eliminates the audit's classic failure mode, deciding from memory and vibe, because memory always overestimates the tools you liked and underestimates the ones you quietly depend on.

The audit itself runs in four passes, ninety minutes, calendar-blocked. Pass one, the usage sort: every subscription gets its ninety-day session count, no discussion, just numbers. Pass two, the five-or-fewer club: anything under five sessions in ninety days goes on the cancellation list, subject only to pass three. Pass three, the seasonal check: the real exceptions are tools that are legitimately seasonal, the tax-season accounting tool, the once-a-year design crunch, and those get paused or downgraded to on-demand rather than cancelled outright, because the goal is matching spend to usage, not punishing tools for having seasons.

Pass four is the one that finds the most money, and nobody does it: the overlap check. Take the tools that survived and ask which pairs do the same core job. My audits have found, repeatedly, two assistants doing one assistant's work, two storage services where one sufficed, a specialized tool whose single valued feature its general sibling had quietly added. Overlaps never announce themselves at purchase; they develop as tools grow into each other, which is exactly why only a recurring audit catches them. The rule for overlap resolution is unfeeling: the one that integrates with more of my actual workflow stays, the other goes, sentiment is not a usage statistic.

The numbers across two years: the audits pay for themselves the first time and keep paying, the annual savings have funded better tools rather than just shrinking the bill, and the second-order effect turned out to be the real prize, a stack where everything earns its place quarterly develops no guilt, no zombie subscriptions, and no dread of the billing page. Ninety minutes, four times a year, entirely mechanical. The subscriptions you forgot you had are counting on you not doing this.