To verify a total an AI assistant gave you, run five checks: count the rows, add the amounts in a spreadsheet, reconcile with your statement's closing balance, spot-check three lines, and make sure the category totals add up to the grand total. If any check fails, the summary is not safe to use. The whole routine takes about ten minutes.
What does a sample answer look like?
Imagine you asked for a March summary and got this: 'You spent ₹58,400 in March. Food ₹14,200, Rent ₹22,000, Travel ₹6,800, Shopping ₹9,100, Other ₹7,500.' It reads well. The checklist will show whether it is right.
What are the five checks?
- Row count. Count your debit rows in the sheet with =COUNT($C$2:$C$200) and ask the assistant how many it read. A mismatch means skipped or invented rows.
- Sum. Add the debit column yourself with =SUM($C$2:$C$200) and compare it with the headline figure.
- Closing balance. Opening balance plus credits minus debits must equal the closing balance printed on the statement.
- Three spot checks. Pick one large, one small and one random line, and confirm each is in the assistant's list with the right amount and category.
- Category totals. Add the category figures; they must equal the grand total.
How do the checks work on the sample?
The category figures add up as 14,200 + 22,000 + 6,800 + 9,100 + 7,500 = ₹59,600. But the headline said ₹58,400, so the answer disagrees with itself by 59,600 − 58,400 = ₹1,200. Next the closing balance. Say the statement shows an opening balance of ₹41,500, credits of ₹65,000 and a closing balance of ₹46,900. Then 41,500 + 65,000 − 46,900 = ₹59,600 should be the debits, and that matches the category total, not the headline.
| Check | Result on the sample | Verdict |
|---|---|---|
| Category totals vs headline | 59,600 against 58,400 | Fail |
| Closing balance | Debits implied: 59,600 | Matches categories |
| Conclusion | Headline wrong by ₹1,200 | Ask the assistant to recount |
The failure is useful: it points to the headline, not the whole answer, so you can ask for a recount rather than start over.
What if the statement balances but the categories look off?
A correct total can hide wrong categories, because moving ₹5,000 from Shopping to Food does not change the sum. Spot-check the largest lines in each category and look at anything labelled Other, which tends to hide errors. If you want a measured view, take a random sample of 30 lines and mark each right or wrong.
How big a gap is acceptable?
For a total built from one statement, the right answer is no gap at all: the debits should match to the rupee. If you are combining items from different dates or currencies, a small difference can come from rounding or exchange rates, and you should be able to name where it arose. A gap you cannot explain is a gap to chase. Write down the gap and its explanation, even a boring one such as one row entered twice, so the same slip is easy to recognise next time.
How do you make this routine quicker next time?
- Keep a sheet with the formulas already in place; paste new rows into it.
- Keep opening and closing balances in two cells, with a check cell that says OK or Check.
- Save the checklist as a note beside the prompt you use each month.
If you track spending in SpendixAI, the reports are built from your saved transactions, so you can compare an assistant's number with the figure on a report and download a CSV to run the same sums.
One rule to keep: a figure you have not checked is a draft, and a decision should wait for the final version.
Useful ideas start with a clearer picture of your own spending.
Explore your reports


