What your subscriptions are actually costing you
Your monthly total is wrong. Annual billing and the five-year view together make subscriptions far more expensive than any single month suggests.
Drafted by an AI agent from the transcripts of AI Magic 2033, and checked line by line against what was actually said. The ideas, the examples and the working are Sean's. The first draft is not.
The monthly figure is the wrong unit. A $12-a-month subscription does not cost $12 — it costs $720 over five years. That is before the two or three services that bill annually, which look like nothing in eleven out of twelve months until they surface, usually at the worst possible time.
Running a bank statement through AI does not just identify what you spend. It reframes the question.
Annual billing hides inside a monthly view
Most people track subscriptions mentally in monthly terms. The streaming service is $8, the fitness app is $14, the cloud storage is $10. The problem is that anything billed annually looks like nothing in eleven out of twelve months. When the annual charge arrives, it does not feel like a subscription. It feels like a surprise.
A 96-transaction statement — a single month of a real household's spending — can contain 15 recurring charges. Two of them will be billed annually. On the day the statement was exported, those two annual charges appeared as a one-off hit rather than the monthly commitment they represent. Dividing them by 12 and adding the result to the monthly subscription total gives the actual picture. Almost no one does this before they run a statement through an AI.
What 96 transactions actually contain
Not every line is a subscription. Groceries, childcare, a coffee — those are spending choices made in the moment. Subscriptions are different: they are charges you agreed to once, which have been deducting automatically, sometimes for months you cannot remember authorising.
Pulling a single month of bank data through an AI and asking it to sort recurring charges from one-off spending produces a list that surprises most people. Charges for apps you stopped opening. A fitness service you paused but never cancelled. A cloud storage tier you upgraded for a project two years ago and forgot entirely. The AI does not judge any of these, which is part of why it works — it simply shows you the full list without the rationalisation you would apply reading through it yourself.
The number that changes which ones you cancel
The fifteen recurring charges in the course example come to A$227.18 a month when you average the yearly ones out. That sounds manageable — less than a dinner out for two.
A$472.84 actually left the account that July. The difference is two services billed annually — a security subscription at A$89 a year and a home cover at A$179 — both of which happened to renew in the month being read. Neither is expensive. Both are invisible eleven months out of twelve, and then one of them is the largest single line on the statement.
That gap is the whole reason this exercise is worth doing. The monthly number is the one you carry in your head. The annual renewals are the ones that make a month feel inexplicable.
At the averaged rate, the fifteen come to about A$2,726 a year, or A$13,631 over five years — arithmetic on the monthly figure, not a prediction. That is the scale the decision should be made at. A service that genuinely saves you an hour a week earns its keep at that price. One you open twice a year does not.
No single subscription looks catastrophic, but collectively there's subscription…
Chapter 3.1 · ~1598s
The transcript captures it better than any polished summary could. The problem is never any one charge. It is the drift.
Why a single month misleads you
The AI flags something important when analysing a single statement: one month cannot establish a normal monthly average. A July statement that contains two annual renewals will make subscriptions appear more expensive than a typical month. A January statement that contains none will make them appear cheaper.
An honest monthly cost requires looking at 12 months of statements, identifying which charges repeat monthly and which repeat annually, then calculating the true monthly equivalent for each. That is a more demanding exercise — but it is also the only one that gives you a number worth acting on.
At minimum, when reviewing a single statement, ask the AI to flag any charge that looks like it might be annual. The size and date patterns make this detectable even from one month of data.
Which subscriptions to cancel first
Not all subscriptions carry the same weight. The most useful filter is: did you actively use this service during the month? If not, it is costing you money for the right to use something you are not using.
The harder ones are subscriptions you do use, but rarely. A design tool you open once a fortnight. A workspace suite with features you have never touched. These are worth auditing against a free tier or a cheaper plan rather than an outright cancellation — the downgrade often preserves what you actually use.
What the AI cannot tell you is which services matter to you. That is always a judgment call. What it can produce, clearly and quickly, is the exact monthly-equivalent cost of each service, and what cancelling the three lowest-value ones would return per year. That is a useful number to have before the decision.
What usually goes wrong
Most people cancel the cheapest subscriptions first, because they are easiest to justify cutting. The calculus is usually wrong. A $5-a-month service you use daily is worth keeping. A $25-a-month service you have not opened in two months is not — even if cutting it feels like a more significant decision.
The other consistent error is running a subscription audit while ignoring annual charges entirely. If your review only captures monthly direct debits, you will miss the annual ones until they renew. The two annual charges in a 96-transaction statement represent real ongoing cost, but they will not appear in the monthly total if you only count what shows up each month.
The actual point of doing this
The exercise is not really about cancelling things. It is about seeing the real number rather than the one that feels small because it appears once a month.
When you see that the same fifteen services are A$227 in an average month and A$473 in a month where the annual ones land, the question stops being "should I cancel this one?" and becomes "which of these do I want to still be paying for in five years?" That is a different decision, and a faster one, because it is being made at the right scale.
Drawn from chapter 3.1 of AI Magic 2033.
