UPI transactions stay uncategorised because the statement narration often carries a payee handle or a person's name rather than a shop name, so there is nothing obvious to match. To auto-categorise them, create rules for the payees you pay repeatedly, correct the first few by hand so the tool learns from your history, and review what is left once a month.
Why UPI narrations are hard to read
A card statement usually shows a merchant name. A UPI line may show a code, a bank reference and a handle, and for payments to individuals it shows a person's name. A tea stall, a flat-mate and a plumber can all look like the same kind of line. Narration formats differ between banks, so check your own statement to see what yours shows.
Step by step
- Sort uncategorised rows by narration and find the payees that repeat.
- For each repeated payee decide the category: groceries, milk, auto fare, maid and so on.
- Create a rule that matches a distinctive part of the narration and assigns that category.
- Fix the remaining rows by hand, starting with the largest amounts.
- Import the next statement and see how many rows are left.
A worked example with made-up numbers
Say a month has 60 UPI rows. Twenty-four go to your regular vegetable seller, milk vendor and auto driver, and 6 go to online services. You write rules for those four payee patterns. If they cover 30 of the 60 rows, half the month sorts itself next time. The remaining 30 are one-off payments, and correcting each once adds to your history for later months.
| Payee pattern in narration | Category | Rows last month |
|---|---|---|
| Vegetable seller | Groceries | 9 |
| Milk vendor | Groceries | 9 |
| Auto driver | Transport | 6 |
| Online services | Subscriptions | 6 |
The four lines cover 9 + 9 + 6 + 6 = 30 rows. You should always check a few of the sorted rows after a rule is added, because a short pattern can match a line you did not intend.
How does SpendixAI categorise UPI rows?
Categorisation runs in stages: your own rules first, then your history and similar merchants, and only if AI is switched on, an AI model for what remains, using the normalised merchant name and an amount band rather than your name, account numbers or balances. When AI is off, unmatched items are held for review. Corrections become history, and you can add merchant aliases so different spellings map to one name. Import the statement as PDF, CSV or Excel after choosing the account, and review rows before saving.
What are the limits?
- A statement layout that a tool cannot read will need manual fixing.
- Rules that are too broad can mis-sort unrelated payments.
- One-off payees stay uncategorised until you correct them once.
- UPI app history is not the same as a bank statement, so use the bank or card statement.
Handling the odd ones
Some UPI rows will never match a rule: a one-time payment to a repair person, a gift to a relative, a deposit to a hostel. For these, pick a category and move on rather than spending ten minutes on each. If you want a record of why, add a short note on the transaction. Over a few months the history you build from these fixes means fewer rows need attention, and the ones left are the truly unusual ones that deserve a look anyway.
A monthly habit
After each import, filter for the uncategorised rows, handle the biggest ones first, and add a rule if a payee appears twice. In a few months the leftover list should be short, and the monthly totals by category become something you can compare.
Useful ideas start with a clearer picture of your own spending.
Explore your reports


