A normalised merchant name is the single, clean name an expense tracker works out from a long, messy bank narration, so that a string like UPI/4829xxxx/SWIGGY/PAYTM and another like SWIGGY BANGALORE IN both become simply Swiggy. It matters because rules, history and any AI model all work much better on one tidy name than on a different jumble of codes every time.
What does a normalised name look like?
A bank narration is written for the bank's systems, not for you. It can hold a payment channel, a reference number, a handle, a city and a date, with the merchant name buried somewhere inside. Normalising means keeping the part that identifies who you paid and dropping the rest. These are made-up examples; real narrations differ by bank and the exact result in your tracker may differ in detail.
| What the statement says | What you actually paid | Normalised name |
|---|---|---|
| UPI/402917365518/ZOMATO ORDER/okhdfc | Zomato | Zomato |
| POS 4012XXXXXXXX3391 CITY SUPERMART MUMBAI 14OCT | A supermarket | City Supermart |
| NEFT-ABCD0123456789-GREEN VALLEY SOCIETY-MAINT OCT | Green Valley Society | Green Valley Society |
Notice what disappears: the long reference numbers, the masked card digits, the city, the date and the bank code. None of those tell you who the merchant is, and all of them change from one transaction to the next.
Why does a clean name matter?
Think about what has to happen when a new transaction is imported. Something has to decide which category it belongs in. If every row of the same shop looks different, nothing can be matched reliably.
- Rules: a rule saying Swiggy means Food delivery is far easier to write and keep working when it can look at one stable name instead of forty variants.
- History: your earlier corrections are remembered against merchants. A clean name lets last month's decision be found this month.
- Reports: totals by merchant only make sense if one shop is not split across five spellings.
- AI: when AI categorisation is on, the model is given the normalised merchant name and an amount band, not your name, account numbers, the full statement text or balances.
What happens when the cleaning is not perfect?
Sometimes a name will not clean up well. Local shops are the common case, because a kirana store paid through UPI may show only a personal-looking handle. Or one chain may appear under two spellings across a bank account and a credit card. That is where merchant aliases come in: you tell the tracker that these two or three names are the same merchant, and later rows follow. Your own corrections feed history, so fixing something once usually pays off on the next import.
How does it connect to categorising in stages?
SpendixAI categorises in an order. Your own rules come first, then your history and similar merchants, and only if AI is switched on, an AI model for what is left. Every stage reads the cleaned name. That is why a messy narration is the single biggest reason an item lands in the uncategorised pile. If AI is off or a monthly AI budget has run out, unmatched items are held for review and wait for you instead of being guessed.
How can I tell if a name is wrong?
- Sort or search your transactions by merchant and look for the same shop appearing twice with different names.
- Look for a very generic name that stands for many different payees, such as only the word UPI or a bank name.
- Check that a person's name has not been treated as a business, or the other way round.
- Add an alias for any spelling that should merge.
- Re-run the check after the next import to see if the merges held.
What to remember
You do not need to clean names yourself in advance. Import, read the first screen of rows, and fix the handful that matter, which are usually the ones that recur: your delivery apps, your grocer, your rent, your fuel. Ten minutes on those repeat names saves more time over a year than fixing a hundred one-offs.
Useful ideas start with a clearer picture of your own spending.
Explore your reports


