AI and MCP

Let AI find unusual transactions in your statements

How to ask an AI assistant to flag duplicate charges and odd transactions in your statements, with the checks that separate real problems from noise.

Editorial illustration for Let AI find unusual transactions in your statements

An AI assistant can find unusual transactions faster than your eyes, if you tell it what unusual means: the same amount to the same merchant twice in two days, a charge several times larger than your normal for that merchant, a merchant you have never paid before. Ask for a list, not a verdict, and verify each row yourself. SpendixAI includes a Possible duplicate charges report built on the same-amount, same-merchant, within-two-days idea.

What does unusual actually mean?

A computer cannot know your life. A ₹23,000 transfer is normal in the month you pay annual insurance and odd any other month. So give it rules you can state in a sentence, and let the numbers fall out.

Four checks worth asking for

  1. Same merchant, same amount, within two days.
  2. Any single charge more than five times your usual amount for that merchant.
  3. Merchants that appear for the first time this month.
  4. Round-figure transfers to individuals that you do not recognise.

Each one produces a short list. Short lists are what a human can actually review.

Worked example

Say your records show a streaming subscription of ₹1,499 charged on 3 November and again on 4 November. That is a possible duplicate, so the double charge sits at ₹1,499 extra. Separately, you normally pay about ₹400 for a cab, and one ride shows ₹3,900. Divide 3,900 by 400 and you get 9.75, nearly ten times your usual. Neither row proves a problem. The first may be a plan upgrade plus the old plan; the second may be an airport transfer. They simply earn a look.

FlagDateAmountFirst thing to check
Repeat charge3 and 4 Nov₹1,499 twiceDoes the app show two invoices?
Large ride9 Nov₹3,900Was it a long trip or a surge?
New merchant12 Nov₹2,750Do you remember paying it?

How do you review a flagged row?

  • Open the transaction and read the full merchant text.
  • Check any attached receipt or the merchant's own confirmation email.
  • Look at the same merchant's history for similar amounts.
  • Mark the row as reviewed in your notes, so you do not check it twice.

The app also flags possible duplicate imports, which is a different thing: the same statement row loaded twice. Fix those first, because a double import makes every other check noisy.

What if it looks like fraud?

Do not wait for an assistant to confirm it. Contact your bank through the channel printed on your card or statement, report the transaction, and follow their steps. Never share an OTP, PIN or net-banking login with anyone who calls or messages you, including someone claiming to be the bank.

Keeping the connection low-risk

If you connect Claude or ChatGPT to SpendixAI for this, start read-only. Finding odd rows needs reading, not changing. Disconnect when you are done if you do not plan to use it again, and remember that what the assistant reads passes through that provider as well.

How do you keep the noise down?

Most false alarms come from predictable events: annual renewals, rent, a festival month when spending jumps, a trip. Tell the assistant about them up front, for example that October includes Diwali shopping and that insurance renews every November, and ask it to treat those as expected. Also set a minimum amount, say nothing under ₹500, so that a list of ₹40 chai payments does not bury the two rows that matter. Tighten or loosen that floor each month until the list is short enough to read in one sitting.

A monthly routine

Spend ten minutes after each statement import: run the four checks, review the lists, resolve each row, and only then look at the budget. Odd rows caught within a month are far easier to dispute than ones found a year later.

SPENDIXAI JOURNAL

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

Explore your reports