AI and MCP

Search merchants by chat: finding every payment to one shop

How to search merchants by chat with an AI assistant when statements show messy names, and how to add up every payment to one shop correctly.

Editorial illustration for Search merchants by chat: finding every payment to one shop

To find every payment to one merchant, search for the part of the name that stays the same, read the variants that come back, and ask the assistant to total only the rows you confirm. Statements rarely print a shop's name the same way twice, so a single search term often misses half the payments. SpendixAI normalises merchant names and lets you add aliases, which narrows the gap.

Why is one shop spelled five ways?

Statements are written by payment systems, not by people. The same online store can appear as its brand name, a shortened code, a payment-gateway prefix or a long reference string with digits. A neighbourhood kirana paid by UPI usually appears as a person's name plus a handle. So searching for the shop you remember can return nothing, while searching for a fragment finds everything.

  • Search for the shortest distinctive fragment, such as AMAZ rather than the full brand.
  • Ask the assistant to list the exact merchant text of every match, not just the total.
  • Ask for the date range explicitly.
  • Ask for matches grouped by exact text, so you can see the variants.

Worked example: three spellings of one store

Say a search for AMAZ over a quarter returns three groups of text, with totals from your made-up records.

Merchant text on statementRowsAmount
AMAZON PAY INDIA6₹1,240
AMZN MKTP3₹899
AMAZON PRIME3₹2,310
All three12₹4,449

Adding: 1,240 + 899 = 2,139, plus 2,310 = ₹4,449. Whether you want all three in one total depends on your question. For 'how much did I spend with this company', yes. For 'how much did I spend shopping', no, because the third line is a subscription, and belongs in a different category. Search finds the rows; you still decide what the rows mean.

What should you do with the variants?

  1. Pick one clean name for the shop, for example the brand name.
  2. Add the other texts as aliases, so future statements match automatically.
  3. Check the category on each variant and align them.
  4. Re-run the search a month later to see whether new variants appeared.

Corrections you make become history, and your own rules are applied before AI categorisation, so this tidy-up pays off every time you import.

What about person-to-person UPI payments?

A payment to a person's name is not a merchant, even if that person is the vegetable seller. Search by the name only if you recognise it, and avoid putting full UPI IDs or phone numbers into a chat; the name or a short fragment is enough for the search. If you pay one person for several reasons, split them in your notes or use sub-categories, since a name alone cannot say whether it was rent, a loan repayment or dinner.

Keeping the chat connection tight

A merchant search only reads data, so read-only access is enough. If you connected Claude or ChatGPT to SpendixAI, choose the smallest scope, and disconnect when finished. What you type and what the assistant reads pass through that provider, so keep identifiers out of your messages.

A small habit that helps

Once a month, sort the uncategorised list by merchant text and give each new variant a home. Five minutes of alias work keeps next month's searches complete, and keeps totals for a single shop honest without any extra effort at search time.

SPENDIXAI JOURNAL

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

Explore your reports