Evidence-first money workflow · Personal Money
Recurring Expense Detector
Detect likely weekly, fortnightly, monthly, quarterly, and yearly expenses from local transaction history with transparent confidence evidence.
- 1Prepare
- 2Analyze
- 3Review and export
Prepare the evidence
Paste the documented CSV schema or choose a local CSV file. Nothing is sent to Nirmion.
Review the analysis
Summary metrics lead back to the rows that support them.
Method and interpretation
How to use Recurring Expense Detector
Use this detector to find subscription candidates, utilities, insurance, memberships, and other repeating expenses in an exported transaction history. It is designed as a review aid: each candidate includes occurrence count, date consistency, amount variation, annualized estimate, and expected next date rather than a black-box recurring label.
Prepare the input
Include ISO date, description, signed amount, and account. At least three observations with the same normalized merchant and account are needed. Choose an amount variation tolerance that reflects the source: fixed subscriptions can use a tight value, while utilities may need a wider review threshold.
Check the worked example
The example contains four Streambox charges about one month apart and three variable electricity charges at a similar monthly interval. Streambox should produce stronger confidence. Electricity can still appear as recurring while its amount evidence explains why review is needed.
Read the evidence
High confidence means both timing and amount stayed tight under the selected tolerance. Medium or low confidence needs manual checking. Annualized cost multiplies the median amount by the detected cadence; next expected date advances the most recent occurrence by the median interval. Neither value confirms a future charge.
Calculation method
Descriptions are normalized for case, spacing, common card words, punctuation, and trailing reference digits, then grouped with account. The detector calculates median amount, median interval, interval range, and amount range. Supported cadence bands are 5–9, 12–16, 26–35, 80–100, and 350–380 days.
Review Lunch Money recurring-item concepts and matching context
Questions this workflow helps answer
Use these questions to confirm that this tool matches the task you need to complete.
- How can I find recurring expenses in a bank-transaction export?
- Which repeated charges belong in a recurring-expense tracker?
- Does a transaction series look weekly, monthly, quarterly, or annual from its dates and amount variation?
Limits and decision boundary
Merchant normalization can merge distinct services or fail when descriptors change completely. Three events are weak evidence for long cadences. Annualized values ignore price changes, free trials, cancellations, taxes, and currency conversion. Review the source statement before acting.
Common mistake
Do not treat every regular-looking charge as an unwanted subscription. Rent, utilities, insurance, debt payments, and intentional savings can all recur. Use the evidence to classify the obligation first.
Your pasted values and selected CSV files are processed in this browser tab. This workflow does not connect to a bank, save a budget, or provide financial, tax, legal, or investment advice.
Questions about this workflow
Why are two charges from one company separate?
The account is part of the grouping key. This prevents similarly named charges on different cards from being combined without review.
Why is an obvious annual bill missing?
The detector needs at least three observations. Three annual events also need dates within the yearly cadence band, so a shorter export cannot establish the series.
Can I cancel a detected expense here?
No. The tool provides evidence only. Cancellation must be completed with the merchant, and the next statement should be checked for another charge.