Recurring expenses are the easiest money to lose and the least interesting to review. A subscription renews. A rate rises quietly at the end of an introductory period. A service you stopped using in March is still billing in November. None of it is urgent enough to deal with on any particular evening, which is exactly how it survives for years.
AI can make the review far less tedious. It should not, however, be anywhere near your bank login.
Why this gets ignored
The individual amounts are small enough to feel unworthy of attention, and the review is genuinely dull — reading a year of statements is a few hundred lines of repetitive text. The cost of ignoring it compounds silently, and the discomfort of doing it arrives all at once. Predictably, it does not get done.
What makes it worth automating is not that it is difficult. It is that it is boring in a way that specifically suits handing off.
The privacy line, first
Some tools ask to connect directly to your bank account. Whatever the merits, that is not necessary for this, and it is not what is being suggested here.
You do not need to hand over credentials, and you should not paste raw statements containing account numbers, full card numbers, addresses or anything identifying. What you need is a sanitised list you build yourself. It takes about twenty minutes and it is the single most important step.
Anything you paste into a general-purpose AI tool may be retained or used depending on that provider’s terms. Read them, and assume anything sensitive should simply never go in.
Building the sanitised list
Open the last twelve months of statements and pull out only recurring charges. For each one, note four things:
- A generic description — “music streaming”, “cloud storage”, “home insurance”. Not the account, not the card.
- Amount and frequency — monthly, annual, quarterly.
- Roughly when it started, if you can tell.
- Whether you have used it in the last 90 days — yes, no, or unsure.
No numbers that identify anything. No names of people. This list is entirely safe to work with, and it is all the analysis actually requires.
Building it by hand has a side effect worth mentioning: about a third of people find at least one thing they had forgotten before any analysis begins.
Categorising and finding the pattern
With the list prepared, the work becomes straightforward. Ask for it to be grouped by category, then sorted by annual cost rather than monthly — because £14 a month reads as trivial and £168 a year does not. That reframing alone changes decisions.
The patterns worth surfacing:
- Duplicates. Two services doing one job. Cloud storage is the classic; so are overlapping streaming subscriptions and two password managers.
- Zombie subscriptions. Unused for 90 days, still billing. Often a trial that converted.
- Creep. Anything that has risen since it started, particularly after an introductory rate ended.
- Bundled overlap. Things you already have as part of something else and are paying for separately.
None of this is analysis a person could not do. It is analysis a person reliably does not do, which is a different problem, and a more solvable one.
Comparing alternatives
Once you have a shortlist, the obvious next question is whether a better option exists. This is where care is required.
Asking for a comparison of typical market alternatives and what to look for is reasonable. Treating any specific price it produces as current is not. Prices, tiers and promotional terms change constantly, and a model will state an outdated figure with complete confidence.
Use it to generate the shortlist and the comparison criteria. Then open the provider’s own page and confirm the live price before you act on anything.
Preparing to cancel or negotiate
The part people avoid is the conversation. This is where preparation genuinely helps.
Ask for the questions worth asking before cancelling: whether a retention offer typically exists, what notice period applies, whether cancelling mid-term forfeits anything, whether an annual plan changes the maths. Ask what a reasonable negotiating position looks like given what comparable services charge.
Then verify the specifics against your actual terms, because your contract is the authority — not a general description of how such contracts usually work.
A short prepared list of questions turns a call you have been avoiding for four months into eight minutes.
Flag first. Verify second. Cancel third.
This sequence matters, and the order is not arbitrary.
Flag. The analysis produces candidates — things worth a closer look. That is all it produces.
Verify. You open the actual statement and the actual terms. Is the charge what you think it is? Is it still active? Is anything else attached to it? This step catches the genuine error case: cancelling something you needed because a line item was misread.
Cancel. Only you do this, and only after the first two steps. No tool should have the authority to cancel, switch or negotiate on your behalf, and nothing here suggests giving it that authority.
A simple recurring-expense audit
- Pull twelve months of statements and build the sanitised list.
- Group by category and sort by annual cost.
- Flag duplicates, unused services, and anything that has crept upward.
- For the top few by annual cost, research alternatives — as a starting point only.
- Verify every candidate against the real statement and the real terms.
- Prepare your questions for the ones worth a call.
- Cancel, switch, negotiate or knowingly keep. All four are valid outcomes.
That last point deserves emphasis. Deciding to keep something after looking at its true annual cost is a legitimate result. The goal is not to cut spending — it is to stop paying for things by accident.
What this will not do
It will not tell you what you should value. If a subscription costs £200 a year and brings you genuine pleasure, no analysis has an opinion on that worth listening to.
It will not know your contract. It will not know a charge is for something you are contractually committed to for another eight months. It cannot see your statement — it only sees what you chose to describe.
And it should never be given the ability to act. The work is finding candidates and organising evidence. Every decision that costs or saves real money stays with you.
Put AI to work for your money.
This sits within the Save pillar at Deliberately Wealthy. The wider model behind it — problem, AI work, verification, human decision — is set out under Workflows.
New workflows, honest results and the approaches that did not survive verification go out in the Deliberately Wealthy Weekly Brief.