Subscription Audit Scanner

Point it at a transaction export and it finds the subscriptions you forgot you're paying for — including the ones too small or too generically named to spot by eye.

Date: 2026-08-11 Type: Utility Theme: Budgeting Status: Idea

What it is

A command-line Python script that reads a CSV of card or bank transactions, groups charges by merchant, and picks out the ones repeating on a roughly monthly schedule. It doesn't just list them — it flags the specific reasons a charge is worth a second look: the amount is small enough to hide in a statement, the merchant name is too generic to place (PAYPAL *INST XFER tells you nothing about what you're actually paying for), or the price crept up since the first charge.

Who it serves

Someone who wants to run a subscription audit but doesn't want to manually scroll three months of statements circling recurring charges by hand — the approach NerdWallet and most personal-finance guides still recommend as the default method.

The money problem it addresses

C+R Research's 2022 subscription survey — still the benchmark figure cited across 2025–2026 coverage — found that people estimate they spend $86/month on subscriptions when the real number, tallied line by line, averages $219/month. Three-quarters of respondents said recurring charges are easy to forget, and 42% admitted to paying for something they'd already stopped using. The gap isn't because people don't care; it's that a $2.99 iCloud charge and a $9.99 line item labeled "PAYPAL" don't announce themselves on a statement full of groceries and gas. This script does the circling for you, and adds the two checks a manual read tends to skip: did this price go up since I signed up, and is this charge actually identifiable.

How it works

1. Reads a JSON config pointing at a transaction CSV (date, merchant, amount, category).
2. Normalizes merchant names — strips trailing store codes like #4521 — and groups charges by merchant.
3. Keeps a merchant only if it recurs at least min_occurrences times (default 3) with an average gap close to 30 days. This is what keeps groceries and gas out of the results even when the same store shows up five times in a quarter — the charges are frequent, but the amounts and spacing are irregular, not subscription-like.
4. For every merchant that survives that filter, checks the first charge against the latest one and flags it if: the price rose past price_hike_threshold_pct (default 15%), the amount sits under small_amount_review_threshold (default $5), or the merchant name matches a generic-keyword list (PAYPAL, SQ *, GOOGLE *).
5. Prints a report: every recurring charge with its monthly amount and flags, a monthly/annual total, and the same totals narrowed to just the flagged subset.

What's in this folder

• requirements.md — the full behavior spec: inputs, detection rules, exit codes, and the known limitations (monthly cadence only, exact-match merchant grouping).
• script.py — the scanner itself. Standard library only, Decimal for all money math, type-hinted, exits non-zero on a missing or malformed config/CSV.
• sample-config.json — the detection thresholds and file paths the script reads.
• sample-transactions.csv — three months of synthetic transactions: eight real recurring subscriptions mixed in with groceries, gas, and one-off purchases that look repetitive but aren't.
• sample-output.txt — the actual report the script prints when run against the sample data (captured from a real run, not typed by hand).

How to run / read it

python3 script.py --config sample-config.json

No setup beyond Python 3.10+. It reads sample-transactions.csv (named in the config), prints the report to stdout, and — if output_report is set in the config — writes the same text to a file.

Estimated impact

Against the sample dataset, the script surfaces $88.64/month ($1,063.68/year) in recurring charges across eight subscriptions, five of which get a review flag: a $2.99 iCloud charge, a $3.99 note-taking app, a $4.25 news subscription, a $9.99 charge labeled only "PAYPAL," and a Spotify plan that rose 20% ($9.99 → $11.99) partway through the window. That flagged subset alone is $33.21/month, or $398.52/year — the specific slice most likely to survive an unaided skim of a bank statement. On real data the number moves with what's actually in the account; NerdWallet's own writer found $1,470/year running this exact exercise by hand.

Good to know

This is a detection tool, not a bill-pay or cancellation service — it never touches a live account, and it only sees what's in the CSV you give it. It's tuned for monthly subscriptions specifically; an annual renewal (Amazon Prime, most insurance) needs a longer history window and a different cadence setting to catch, and the sample data deliberately includes one to show it getting correctly left out. The review flags are heuristics — small amount, vague merchant name, price jump — not proof that a charge is unwanted. Rates, offer terms, and subscription prices referenced here move over time; check your actual statement before canceling anything.

Sources

• https://www.crresearch.com/blog/subscription-service-statistics-and-costs/ — the $86 estimated vs. $219 actual monthly spend gap, the 74% "easy to forget" figure, and the 42% who pay for an unused subscription.
• https://www.nerdwallet.com/ca/p/article/finance/how-subscription-audit-helps-save-money — the manual statement-scan audit method (pull 2–3 months of statements, list every recurring charge) and the $1,470/year real-world result it grounds the impact estimate against.
• https://www.substract.co/blog/how-to-find-all-your-subscriptions — the specific blind spots (sub-$5 charges, cryptic merchant names, PayPal-masked charges) that shaped which review flags the script checks for.