Business names in Arabic and other scripts: matching, sorting and searching them
Business names in Arabic data seldom match the same shops written in Latin letters. How to join, sort and search mixed-script lists without losing customers.
Part of the guide Location intelligence automation
Business names in Arabic data rarely line up with the same names written in Latin letters, so never join two lists on the name alone. Match on something that does not change with the script, such as the phone number or the website, and only then compare names. When you must work with the names themselves, tidy them first: remove the short vowel marks and the stretching stroke, write the variant letters one way, and keep a Latin version beside the original so a person can read both.
This is for sales, operations and analyst teams in the Gulf, North Africa and the Levant who mix their own customer files with place data, and for anyone handling names in Japanese, Thai, Hindi, Greek or other non-Latin scripts. The same problems appear in all of them; Arabic simply shows them most often.
Why the same shop looks like three different shops
Take one pharmacy in Riyadh. Your ERP holds it in Arabic, as typed by a branch clerk. A distributor's sheet holds it as Al Nahdi Pharmacy. A second sheet from a marketing agency holds it as Nahdi Pharmacy, Olaya. To a computer these are three unrelated strings. Even within Arabic alone, the name can be stored in several ways that look identical on screen:
- Short vowel marks. One clerk adds them, another does not. The letters look the same to a reader but the stored text differs.
- Variant letters. The alef can be written with a hamza above, below or none; a final taa marbuta is often typed as a plain haa; a final alef maqsura is often typed as a yaa.
- The stretching stroke. A long horizontal stroke used for looks inside a word adds an invisible extra character.
- The article. Al, El, Ul or nothing at all in the Latin version, joined with a hyphen or a space.
- Order and extras. Branch names, districts and words such as Pharmacy or Est. added before or after.
Japanese has its own versions of this: the same shop can appear in kanji, in katakana and in Latin letters. Thai names can carry or skip the tone marks a romanised spelling never shows. In Delhi a shop may be listed in Devanagari on one sign and in English on another.
Match on what does not depend on the script
The safest key between two lists is a phone number. Digits are digits in every script, and the leading country code is the only thing that tends to differ. In xMap Flows, Match with my table has a same phone number option, and Remove duplicates compares phones by their last nine digits, so a number saved with +971 and the same number saved with a leading zero count as one. A website is the next best key: a domain is written in Latin letters whatever the shop's name looks like.
Names come last. Use them to break ties or to sort, not to decide that two rows are the same business.
Sorting and searching mixed names
Sorting by name works within one script but not across them. Names in Latin letters and names in Arabic will form two separate blocks in any sorted list, and two Arabic names that differ only by a variant alef may not sit next to each other. If your team needs one alphabetical list, add a Latin column and sort on that.
Searching has the same trap. A filter that looks for a word without vowel marks can miss a row stored with them. Filter ignores capitals and Latin accents, but it compares Arabic letters as they are stored, so clean the names before you filter on them.
A worked example
The figures here are made up. A food distributor in Dubai has 1,800 customers in its own system, about two thirds of them named in Arabic, and wants to know which restaurants within its delivery area are not yet customers.
- Upload the customer file to My data. Rows without coordinates are placed from their address or city columns.
- Find the restaurants in the delivery area with Nearby places. Each place comes with its name as the database stores it, plus a phone and website where known.
- Match the two with Match with my table on same phone number, set to keep only the places not found. In this made-up run 1,240 places go in and 410 come out as possible new customers.
- Send the 410 through a second Match with my table on same website, which removes another 60 that used a different phone.
- Ask an xMap LLM block to add two fields: the name in Latin letters and a one-line note on what kind of restaurant it is.
- Sort by the Latin name and send the list to the sales team.
A rep then reads the remaining 350 with both versions of each name side by side, which is where a human eye still beats any rule.
Mistakes to avoid
- Joining on the name column. It quietly misses most matches between scripts and calls a known customer a new lead.
- Throwing away the original. Keep the Arabic or Japanese name exactly as it came. The Latin version is for sorting and reading, not a replacement.
- Trusting an automatic spelling. A model can write Al Nahdi or An-Nahdi; neither is wrong, so never use its output as a matching key.
- Cleaning only one side. If you strip vowel marks from your file, strip them from the incoming data too.
- Ignoring partly blank rows. Rows with no phone and no website deserve their own small list for a person to check.
Doing it with an AI employee in xMap Flows
An AI employee can repeat the job every week so the comparison never goes stale. The order of blocks matters: the script-proof matches go first, the AI comes after, so it only works on rows that are really new. For more on what AI employees are and how they report, see the AI employees guide.
- Schedule, every Monday morning.
- Area of interest with your cities or a drawn delivery zone.
- Nearby places for the types you sell to.
- Match with my table on phone, then a second one on website, both keeping only what you do not have.
- Custom code, if you want a cleaned name field: a few lines can remove vowel marks and write the alef forms one way.
- xMap LLM to add the Latin name and a short reason the place fits.
- Email with the list, or a sheet the team works from.
Each Monday the email lists only places that are new to you, each with both spellings. The template Enrich a lead sheet and push it to HubSpot is a good base if your list lives in a sheet, and Watch my target accounts suits a fixed list of accounts you follow.
Questions
Why does my Arabic customer list not match the places I find?
Usually because the same business is written differently: with or without vowel marks, with a different alef, or in Latin letters on one side. Match on phone or website first, then use names only to check.
Can xMap Flows translate business names into English?
An xMap LLM block can add a field with each name in Latin letters or in English. Treat it as a reading aid: the spelling it chooses is one of several correct ones, so do not match on it.
Does the same problem happen with Japanese or Thai names?
Yes. Japanese shops can be listed in kanji, katakana or Latin letters, and Thai names may be romanised in several ways. The same rule applies: match on phone or website, keep the original name, add a Latin column for sorting.
What if a row has no phone and no website?
Put those rows in their own list and have a person compare them by name and address. They are usually a small share, and guessing on them causes the worst errors.
Templates that do this
Read next
More on business and places data · Every article · Templates