To compare this month with last month using AI, paste two category tables side by side, ask for changes above a threshold you set, then ask the assistant to sort each change into one-off or recurring using notes you supply. Small moves are noise, so the threshold is the most important part of the prompt. Anything under ₹500, say, should be ignored.
Paste two months in one table
These are hypothetical months.
| Category | Last month (₹) | This month (₹) | Note |
|---|---|---|---|
| Groceries | 9,800 | 10,400 | None |
| Fuel | 3,200 | 4,600 | Long drive for a family visit |
| Eating out | 3,900 | 5,200 | None |
| Health | 600 | 2,800 | Doctor visit and tests |
| Shopping | 2,000 | 7,500 | Festival clothes |
| Utilities | 3,100 | 2,900 | None |
Last month totals ₹22,600 and this month ₹33,400, a difference of ₹10,800. Ask the assistant to confirm both totals first.
Prompt 1: the biggest changes
Percentages need care with small bases. A rise from ₹600 to ₹2,800 is about 367%, calculated as 2,200 divided by 600, but the rupee figure is the better guide. Ask for both and read the rupees first.
Prompt: "Compare the two month columns. For every category, give the change in rupees and as a percentage of last month. List only those with an absolute change above ₹500, ordered by size. Show the subtraction."
Expect Shopping +₹5,500, Health +₹2,200, Fuel +₹1,400, Eating out +₹1,300 and Groceries +₹600. Utilities, down ₹200, drops out. The five changes less ₹200 equal ₹10,800, which ties back to the totals.
Which changes were one-offs and which will repeat?
Prompt: "Using my Note column, label each change One-off, Recurring or Unknown. For Unknown, ask me a question. Then recalculate what this month's total would have been without the One-off items."
Then ask the assistant to list, in one line, the categories that moved by less than your threshold. Learning that Utilities and Groceries stayed calm tells you the extra spending came from a few lines, not everywhere.
By your notes, Health, Shopping and Fuel were one-offs. Taking out their increases (₹2,200, ₹5,500 and ₹1,400, together ₹9,100) leaves ₹1,700 from the rest: Eating out +₹1,300, Groceries +₹600 and Utilities −₹200. That ₹1,700 is the underlying drift worth watching, and a good assistant should say so.
Ask also for the largest single transactions in the categories that moved most. A rise in Shopping is easier to understand when you see that one entry of ₹3,000 caused half of it.
How do I stop the assistant over-explaining?
Say how long the answer should be: "Reply in a table plus at most three lines of comment. No advice unless I ask." Without a limit, assistants tend to produce a lecture and bury the table.
What if the months differ in length or events?
Ask for a per-day figure: divide each category by the days in its month. February against March, or a month with a long trip, deserves the extra column. Also tell the assistant about events: "This month included a festival, last month did not." Context cuts down false alarms.
If you have the data, run the same comparison against the same month last year too. Festivals and school terms fall in similar months each year, so a month-on-month view alone can mislead when the calendar shifts the pattern.
A repeatable routine
- Put both months and a Note column in one table.
- Set a rupee threshold before you paste.
- Ask for one-off versus recurring, then recalculate without the one-offs.
- Keep the recurring drift figure in your notes from month to month.
If your records live in SpendixAI, its reports let you pull category totals for two periods to paste. Claude or ChatGPT can also be connected through MCP, where you pick the scope and may begin with read-only access.
Useful ideas start with a clearer picture of your own spending.
Explore your reports

