A fuzzy merge in Power Query matches rows whose text is similar rather than identical. Load your transactions and a clean merchant list as two queries, choose Merge Queries, select the name columns, tick Use fuzzy matching and set a similarity threshold. Then review the borderline matches, since lowering the threshold finds more but also wrongly joins different shops.
What do you need before you start?
- A transactions table with a merchant or description column, loaded as a query.
- A clean merchants table with one row per shop and a clean name, also loaded as a query. A few dozen rows is a good start.
- Excel 2016 or later, or Microsoft 365. Menu names vary a little, so check your version.
Fuzzy matching works on text columns and can be slow on very large tables, so try it on a sample first.
How do you set up the merge?
- In the Power Query editor, open the transactions query and choose Home, then Merge Queries.
- Pick the clean merchants query as the second table.
- Click the transaction name column in the top table, and the raw-name column in the bottom one.
- Set the join kind to Left Outer, so every transaction is kept even when nothing matches.
- Tick Use fuzzy matching to perform the merge, and open Fuzzy matching options.
- Click OK, then use the expand button on the new column to bring in the clean name.
What do the fuzzy options mean?
| Option | What it does | Advice |
|---|---|---|
| Similarity threshold | A number from 0 to 1. Higher means stricter, and the usual default is around 0.8 | Start at the default, then try 0.7 and 0.9 and compare |
| Ignore case | Treats upper and lower case as the same | Leave it on for narrations |
| Match by combining text parts | Can join words split across a gap, such as ANAND DAI RY | Try it when names are broken by spaces |
| Maximum number of matches | How many rows of the second table may match one row | Set it to 1 so a transaction gets one clean name |
| Transformation table | A two-column table of text swaps applied before matching | Use it for known equivalents such as PVT and PRIVATE |
How do you tune the threshold?
Think of the threshold as a dial between missing matches and making false ones. An invented example: the clean list has ANAND DAIRY and ANAND SWEETS. A narration of ANAND DAIRYY clearly belongs to the first, but at a low threshold a narration of ANAND STORES might be matched to either. Try three thresholds, add the clean name next to the raw one and look at the rows that change between them. Those are the borderline rows that deserve your eyes.
What goes in a transformation table?
It has two columns, From and To. Each row says that, before comparing, the text in From should be treated as the text in To. For narrations you might list common abbreviations or noise words. This is a good way to teach the matcher something you know, without loosening the threshold for everything else.
How do you review the borderline matches?
- Filter for transactions where no clean name was found, and add the missing merchants to your list.
- Spot-check one raw name from each clean name, especially the ones with many rows.
- Compare the total amount before and after the merge. It must be the same, and it will not be if a transaction matched two rows. That is why you set the maximum matches to 1.
- Keep the raw name column in your output.
Does the next month's file clean itself?
Yes, in the sense that the merge is recorded as a step. Paste new rows into the source table, press Refresh All, and the same threshold applies. Still glance at the Unmatched rows each month. New merchants will appear.
If maintaining a clean list and a query is more than you want, SpendixAI is a different approach: it normalises merchant names, lets you add aliases and learns from your corrections. Power Query remains the better tool if you want every rule visible in your own workbook.
Useful ideas start with a clearer picture of your own spending.
Explore your reports

