Getting started

Is imported expense data accurate enough? Why you still review

Is imported statement data accurate enough to trust? The statement is the bank's record, but parsing can slip, so here are the three mistakes worth catching.

Editorial illustration for Is imported expense data accurate enough? Why you still review

Imported statement data is as accurate as the bank's own statement on the money itself, because the amounts and dates come from that record. What can slip is the reading of the file and the guessing of categories. That is why a review step is a feature, not a flaw, and why a quick check of three things, sign, date and category, catches most of what matters.

What can be trusted, and what cannot?

Split the job into three layers, each with its own risk.

LayerWhere it comes fromHow reliable
Amount and dateThe bank's statementThe bank's truth, if the file is read correctly
Reading the fileTurning a PDF, CSV or Excel layout into rowsCan slip on odd layouts, wrapped lines or number formats
CategoryRules, history and optionally AIA good first pass, but a guess for new merchants

No importer can promise that every statement layout will read perfectly. Layouts vary and banks change them. Because rows are shown before they are saved, you get a chance to compare them with the original.

Which three mistakes are worth catching?

  1. Sign: a credit shown as spending, or a refund treated as a purchase. This one distorts totals the most, because a single large mistake swings a month.
  2. Date: a row that lands in the wrong month, or a day and month swapped, so that 03/11 reads as 11 March. It quietly moves spending between months and breaks budgets.
  3. Category: a merchant put in the wrong bucket. It matters less for the grand total but a lot for the reports built on categories.

Everything else is usually cosmetic. A cut-off merchant name can be fixed later with an alias or a rule, and a missing note matters little.

How long should a review take?

Less than you fear. Say a statement has 140 rows. You do not read every narration; you look at the biggest five debits, any credit, the first and last date, and the rows marked as possible duplicates. Then skim the categories for anything that looks wrong. Five to ten minutes is realistic once your rules cover the repeat merchants, and the first import will take longer.

Why does categorisation need checking separately?

Categorising happens in stages in SpendixAI: your own rules, then your history and similar merchants, then an AI model for what is left, if AI is on. A rule is firm. History is a strong hint. An AI model is a best guess from a normalised merchant name and an amount band. Anything it is unsure of, or everything unmatched if AI is off, is held for review. A correction you make becomes history, so the next import does better, which is the real payoff of reviewing.

Is review a reason not to bother?

No, it is the reason the result is trustworthy. A tracker that quietly saved every row without showing you would be faster for a month and wrong for a year. A short look at each batch catches the few errors that matter, while the machine does the typing that you used to do.

A pocket review list

  • Does the row count match the statement?
  • Are debits and credits the right way round on the biggest rows?
  • Are the first and last dates inside the range you downloaded?
  • Are there any possible duplicates, and have you checked them against the bank?
  • Do the top five categories by amount look sensible?
  • If something is strange, can you download the same statement as CSV or Excel and try again?

If a number never reconciles with your bank, take the question to the bank, because the statement is the primary record.

SPENDIXAI JOURNAL

Useful ideas start with a clearer picture of your own spending.

Explore your reports