Nobody was careless
It is tempting to treat forgotten subscriptions as a discipline problem — someone should have been watching. That framing is both unkind and wrong, and it produces the wrong fix: a memo about being careful, which changes nothing.
The truth is structural. A subscription is designed to be forgotten. That is not a criticism of the companies selling them; it is what a subscription is — an arrangement that continues without further decisions. Every property that makes them convenient also makes them invisible.
So the fix is structural too: not more vigilance, but one session a year in which the data is arranged so that the invisible becomes obvious. That is what this piece is about.
Why they are structurally invisible
Four properties, each individually reasonable, that combine into something nobody sees.
Each charge is small. Below the amount that makes anyone look twice. Twenty subscriptions at modest amounts would demand attention as one line and get none as twenty.
Continuing requires no action. An invoice gets approved by someone. A subscription renews because time passed. There is no moment at which a person is asked whether this should continue.
They are spread out. Company card, current account, a second card, occasionally a personal card and an expense claim. Any single-account review misses most of them, which is why most reviews find less than they should.
The pattern only exists across time. This is the deepest one. Look at a single transaction and nothing marks it as recurring. A statement in date order is precisely the arrangement in which a repeating pattern cannot be seen.
Where they come from
Five origins account for nearly everything anyone finds, and none of them involve anybody doing something wrong.
| Origin | What happened | Why it survives |
|---|---|---|
| The trial that converted | Signed up to evaluate, forgot to cancel | First charge looks like a decision that was made |
| The project tool | Bought for a job that finished | The project ended; nobody owned the aftermath |
| The departed colleague | Signed up under their name | Nobody left recognises the merchant |
| The duplicate | Two teams bought the same thing | Each team assumes the charge is theirs |
| The upsell nobody noticed | Plan auto-upgraded on usage | The amount changed, not the payee |
Row three is the most expensive and the most awkward. When the person who set something up leaves, the charge outlives every trace of why it exists. Nobody currently at the company can identify it, so nobody cancels it — and it can run for years.
Row four is the most common and the most instructive, because it is a communication problem wearing a cost disguise. The second team did not waste money; they did not know the first team had already solved the problem.
Where they hide
Knowing the origins tells you where to look, and the answer is broader than most people search.
The company card is the densest hiding place. Most software goes on a card, and card statements are reviewed for fraud rather than for necessity — a legitimate charge from a real vendor passes every check anyone applies.
The current account holds the older ones: direct debits and standing orders set up years ago, often for services that predate whatever system you use now.
The second card nobody thinks of. The one issued for travel, or for a specific supplier, or to a person who no longer works there but whose card was never closed.
Payment processors.A charge that appears as a processor’s name rather than the merchant’s is effectively anonymous. These need the card statement’s fuller description to identify at all.
Expense claims. Subscriptions on personal cards, reclaimed monthly. These are genuinely hard to find in bank data and deserve their own review — see expense report to Excel for getting those into a table.
The one-session audit
An afternoon, done once a year. Five steps, and the first is the only one that takes any real time.
1 · Two years, every account
Current accounts, all cards, both years. This is the step people cut short, and it is the step that determines what the audit can find.
2 · Group by payee
Merge the name variants — the same merchant rarely appears identically twice. Ambiguous ones get shown rather than merged.
3 · Keep what repeats
Anything appearing at a regular interval: monthly, four-weekly, quarterly, annual.
4 · Sort by annual cost
Not by monthly amount. The annualised figure is the number that actually prompts a decision.
5 · Three columns
Who owns it, is it used, keep or cut. Twenty minutes with the list in front of you.
Then: record the end dates
An annual plan usually runs to the end of its term. The saving is real and it is not immediate.
Step four deserves emphasis because it is where the psychology turns. A modest monthly figure invites a shrug; the same subscription shown as its yearly cost gets a decision. Sorting by annualised total is the single most useful thing you can do with the list — the mechanics are on subscription detection.
The annual ones are the point
If you take one thing from this piece: annual subscriptions are where the money is, and they are the ones a twelve-month review cannot find.
In a single year of data, an annual charge appears exactly once. One occurrence is not a pattern — it is indistinguishable from a one-off purchase, and every method that relies on repetition will pass over it silently.
They are also individually much larger than monthly ones, and they renew with the least friction of anything you pay for. A monthly charge offers twelve chances a year to notice it. An annual charge offers one, and it usually arrives during a busy month.
Hence two years of statements, not one. It is more work at the conversion step and it is the difference between an audit that finds your small recurring costs and one that finds your large ones.
What people actually find
The list is remarkably consistent across businesses of very different kinds.
| Finding | Typical action |
|---|---|
| Two tools doing the same job | Consolidate onto one, once both teams agree |
| Seats for people who left | Reduce the licence count — often the biggest single saving |
| A tool nobody has opened in a year | Cancel, after checking the admin panel |
| A plan that auto-upgraded | Downgrade, or negotiate the tier you actually need |
| A service from a finished project | Cancel; note the notice period |
| A charge nobody can identify | Investigate, then cancel and see who objects |
| A price that rose quietly | Renegotiate — the rise was never discussed |
The second row is worth its own attention. Per-seat pricing means a team that shrank keeps paying for the size it used to be, and nothing in the billing tells you — the charge stays exactly the same, which is precisely why it never gets questioned.
The quiet arithmetic
The reason this accumulates is not carelessness but a mismatch of scale. Subscriptions are priced to feel small monthly and are paid annually forever, and those two facts are almost never held in mind at the same time.
| How it is presented | How it is paid | What gets compared |
|---|---|---|
| Per user, per month | Total, every month, indefinitely | A coffee |
| Billed annually, save 20% | One charge, renewed silently | A rounding error in the month it lands |
| Starts free | Converts on a date nobody diarised | Nothing — it was free when decided |
| Scales with your team | Rises without a new decision | Last month, which was similar |
Nothing in that table is deceptive. Per-seat monthly pricing is a perfectly honest way to sell software, and the annual discount is genuine. The problem is that the unit in which a decision is made — a small monthly figure, once — is not the unit in which the money is spent.
Which is why sorting by annualised cost changes behaviour so reliably. It is the same subscription, presented in the unit that matches the commitment, and people who shrugged at the monthly figure make a decision within seconds of seeing the yearly one.
The fourth row is the subtlest. A tool priced per seat rises as you hire, without anyone ever approving an increase. It is the only category of spending that grows automatically with success, and it is examined less often than almost anything else in the business.
Seats are usually the biggest finding
Everyone expects to find a forgotten tool. The larger number is almost always hiding inside tools that are actively used and correctly chosen — simply paid for at the wrong size.
Per-seat pricing has a ratchet. When someone joins, a seat is added, because the tool is needed on their first day. When someone leaves, the seat is rarely removed, because nothing forces the question and the charge does not change in a way anyone notices.
Over a few years of ordinary turnover, a team of fifteen can be paying for twenty-two seats across a dozen tools. Nothing looks wrong on any statement: the merchant is legitimate, the amount is stable, and the tool is genuinely in daily use.
Finding this requires a step the bank cannot give you: open each significant tool’s admin panel and compare the licensed seat count to the list of people who actually work there. It takes a couple of minutes per tool and it is routinely the largest single line of the whole audit.
The same ratchet applies to storage tiers, contact limits and usage plans that were upgraded during a busy period and never revisited when the spike passed.
Making it not personal
There is a way to run this audit that produces defensiveness and a way that produces cooperation, and the difference is almost entirely in how the first conversation is framed.
The failure mode is presenting the list as evidence — here is what has been wasted, who signed up for this. It is understandable and it guarantees that next year nobody will help you, because helping means volunteering something that looks like a mistake.
The framing that works is structural, and it is also true: nobody was careless, subscriptions are designed to continue without decisions, and the reason we do this annually is that no individual could reasonably have caught it. That is not diplomacy — it is an accurate description of why the list exists.
Practically, ask people what they use rather than what they signed up for. “Which of these do you rely on?” gets honest answers. “Why are we paying for this?” gets justifications, and justifications are how a tool nobody opens survives another year.
Share the outcome too. A team that hears the total recovered is far more willing to flag something themselves next time, and self-reported cancellations are worth more than anything an annual audit will find.
Two things the bank cannot tell you
The audit produces a list quickly, and then stalls on two questions that no amount of transaction data answers. Knowing that in advance saves an afternoon of looking for the information in the wrong place.
Whether anyone uses it.A payment says nothing about whether the tool was opened. This lives in each service’s own admin panel, usually as a last-login column, and checking it takes a couple of minutes per tool. It is also the single most decisive piece of information in the whole exercise: a tool with no logins in eight months needs no further discussion.
What it is for. Merchant names are frequently unrecognisable — a holding company, a product name that changed, or a payment processor standing in for the actual vendor. Searching the name usually resolves it in seconds; when it does not, the card statement often carries a longer description than the bank line does.
When neither works and nobody recognises the charge, cancelling and waiting to see who objects is a legitimate method. It feels reckless and rarely is: a service genuinely in use produces a complaint within days, and one that produces silence for a month was answering the question on its own.
Do that one at a time rather than in a batch, and not during a period when the business cannot absorb a small disruption. The point is to learn something cheaply, not to prove a point.
Deciding what to cut
The list is the easy part. Three questions per row make the decisions quick and stop the exercise stalling.
Who owns it? A named person, not a team. If nobody claims it after asking twice, that is itself the answer.
Is it used?Check the tool’s own admin panel, not your impression. Last-login dates take two minutes to find and are frequently surprising — including for tools everyone assumed were essential.
What breaks if it stops? Sometimes nothing. Sometimes an integration nobody remembers building. Asking before cancelling costs a minute; finding out afterwards costs a morning.
Do not treat cancellation as the only outcome. For tools you genuinely use, an annual commitment, a smaller seat count or an older plan tier are usually available and almost never offered unprompted. The audit tells you exactly where that conversation is worth having.
Why once a year is the right frequency
Quarterly sounds more diligent and is worse. Three reasons, and they all point the same way.
Annual renewals only come round once. The largest charges in the audit appear one month a year. A quarterly review catches the same monthly subscriptions four times and the annual ones no more often than an annual review does.
The second pass finds almost nothing. The first audit clears several years of accumulation. Three months later there is nothing to find, and a review that reliably produces nothing is one that quietly stops happening.
Goodwill is finite. Asking teams which tools they rely on is a reasonable annual request and an irritating quarterly one. Spend that willingness where it produces something.
Attach it to the budget cycle, when the recurring baseline is being established anyway. The two exercises need the same data and answer adjacent questions, and doing them together costs barely more than doing either alone.
Cancelling is the hard part
Finding them takes an afternoon. Cancelling them takes longer than anyone expects, and it is where these projects quietly die.
Some cancellations are two clicks. Others require finding the login of a person who left, resetting a password on an email address that no longer exists, or emailing a support desk that replies in three days with a retention offer. A handful require written notice.
Budget for that. Give one person the list and a deadline, rather than distributing it across whoever is nominally responsible — a shared list is a list nobody finishes.
Record the actual end date next to each one, because annual plans normally run to the end of the paid term. Booking the saving from the cancellation date rather than the term end is how a cost-cutting exercise ends up reporting money that has not arrived — and that is worse than not measuring it at all.
Not having to do it again
The audit is worth repeating annually regardless. But four habits stop the list growing between times, and none of them require a policy document.
One card for subscriptions. A dedicated card makes the whole category visible in one place instead of scattered across three. It is the highest-leverage change available and it costs nothing.
A named owner at sign-up. Recorded when the subscription starts, not reconstructed two years later. One line in a shared sheet.
Cancellation in the leavers checklist. Alongside returning the laptop and disabling the accounts. This single line eliminates the most expensive category of all.
A calendar reminder before each annual renewal. Not after. A reminder a month before the renewal date turns an automatic charge back into a decision, which is exactly what it stopped being.
Recurring costs are also the cleanest possible baseline when the budget is rebuilt — the method is in the budget-building guide, and tracking against it afterwards on budget versus actual.
The one you will hesitate over
Every audit produces at least one row that nobody can decide about: a tool used by one person, occasionally, for something that matters when it matters. It is too expensive to ignore and too useful to cancel confidently.
The mistake is to leave it undecided, because undecided means it renews. A row that has been discussed twice and cancelled neither time has effectively been approved, without anyone taking responsibility for approving it.
Three ways out, all better than deferring. Move it to the cheapest tier that still covers the case. Ask the one person who uses it what they would do instead, which is usually a shorter answer than expected. Or cancel it and agree openly that you will re-subscribe if it turns out to be needed — most services make that trivially easy, and a month of absence answers the question definitively.
What makes this worth the paragraph is that the hesitant row is where the process most often breaks down. Not because the amount is large, but because unresolved items accumulate: next year the same row appears, with a year more history and the same absence of a decision.
An honest caveat
This exercise is satisfying, which makes it easy to overrate. Two things are worth saying plainly.
It is a one-off gain, not a strategy. The first audit finds accumulated waste from several years. The second finds a fraction of that. If your costs are genuinely too high, subscriptions are rarely the reason — they are simply the most enjoyable thing to cut, because nobody defends them.
Cutting a tool people rely on is expensive. The saving is visible and the cost is not: someone spends a week working around its absence, and nothing records that. Check usage before cancelling, and ask before assuming.
With those two caveats, it remains one of the highest-return afternoons available in a small business — largely because the alternative is paying indefinitely for things nobody would choose to buy today.
If there is a single sentence worth keeping from all of this, it is that the fix is a layout, not a discipline. Nobody needs to become more vigilant about small recurring charges — vigilance is precisely what a subscription is designed not to require. What is needed is one afternoon a year in which the same data everyone already has is arranged so that repetition is visible instead of hidden, and the annual figure is shown instead of the monthly one. Everything else follows from those two changes, and neither of them is about anyone trying harder.