There is a line item in almost every business's accounts that grows by default and shrinks only by decision, and it is software subscriptions. The mechanics are familiar to anyone who has watched them: a tool is adopted to solve a problem, seats are added as people join, the problem changes or the champion leaves, and the subscription continues - because cancelling requires someone to notice, decide and act, while continuing requires nothing. Multiply by every team and every year, and the result is what industry studies keep measuring: organisations typically use their software portfolios far below what they pay for, with substantial shares of licences never or barely touched, duplicate tools covering the same function, and renewals firing automatically on cards nobody reviews. It is not a scandal. It is entropy - and it compounds at renewal prices that have been rising faster than most companies' revenue.
The audit that corrects this is unglamorous and fits in an afternoon for a small business, a week for a mid-sized one. Step one is discovery, and it starts from money rather than memory: pull twelve months of card and bank statements and list every recurring software charge - the canonical finding is subscriptions leadership did not know existed, on cards belonging to people who have left. Step two is usage against seats: for each tool, how many licences are paid and how many humans actually logged in during the last month - most platforms show this in an admin panel; the gap between the numbers is the purest waste in the building. Step three is overlap: cluster the list by function - communication, storage, design, project tracking, notes - and count the clusters served by more than one paid tool; consolidation is harder than seat-trimming, since someone loves each duplicate, but it is where the durable savings live. Step four is action with a default: every subscription is cancelled, downgraded or consciously renewed, and unused means cancelled unless someone claims it by name with a reason.
Two failure modes deserve flagging, because each converts the audit into a different mistake. The first is doing it once: subscription entropy is a flow, not a stock, and a heroic annual purge merely resets the clock. The fixes that hold are structural - a named owner per tool, renewal dates on a shared calendar so decisions precede charges instead of following them, and a lightweight rule that new subscriptions above a threshold get a second signature and an entry on the list. The second failure mode is over-rotation: cutting tools people quietly depended on, or centralising procurement so hard that teams return to the shadow-purchasing that created the sprawl - the same dynamic covered here with unsanctioned AI tools, and it responds to the same remedy: make the sanctioned path easy, not just the unsanctioned path forbidden.
What elevates this beyond housekeeping is the ratio of return to effort, which is difficult to beat anywhere else in the business right now. Recovering a meaningful slice of a software budget requires no negotiation with customers, no new revenue, no risk - it is margin lying on the table, and in a period of compressed margins and rising renewal prices, margin found is strategy. There is also a discipline dividend: a business that knows what software it pays for, who owns each tool and what each is for has, almost by accident, built the inventory that every security review, every AI-governance question and every due-diligence process will eventually ask for. The subscription economy's quiet business model is corporate forgetfulness. Remembering, once a quarter, is the whole countermeasure.