To read customer complaints in Arabic and English fairly, sort both languages with one scheme, so that a complaint about waiting in Arabic and the same complaint in English land in the same category and count together. If you run separate processes for each language, or leave one language to be read by hand, branches whose customers write in different languages cannot be compared. The test for any tool is simple: give it a mix of your real messages in both languages and see whether it files the same problem in the same place.
This guide explains why bilingual comparison goes wrong, what to look for in a tool and how to test one.
Why this matters for multi-branch businesses
In many markets, the customers of one branch write mostly in Arabic and the customers of another mostly in English, and many write in both. If the two are handled differently, a branch's apparent performance depends partly on the language of its customers.
Three things go wrong.
- Split piles. Complaints in each language are sorted into different categories, so the same problem looks like two small ones.
- Uneven attention. One language gets read by a person, the other by software, or one gets read late.
- Hidden branches. A branch whose customers write in the language your team reads less well looks quieter than it is.
The aim is that a complaint at branch 40 is measured like one at branch 3, whichever language it is written in.
What makes Arabic complaints harder to read
Arabic is not harder in principle, but several features need to be handled.
- Dialects. Customers often write in a regional dialect, not Modern Standard Arabic, and the same word can differ between regions.
- Mixed language. Messages commonly switch between Arabic and English in one sentence.
- Spelling and short forms. Typed messages are often short, informal and inconsistent.
- Arabic written in Latin letters. Some customers type Arabic with English letters and numbers, which many tools handle poorly.
- Voice. Spoken complaints depend on transcription quality, and accents matter.
- Layout. Right-to-left text must display properly alongside left-to-right text, including numbers, names and mixed lines.
None of these is an argument against using software. They are the things to test.
What to look for in a tool
One scheme, both languages
Ask whether the same categories and the same logic apply to Arabic and English, and whether a complaint in either language lands in the same issue.
The original message behind every finding
Whoever reads the result should be able to open the original text or recording. You need this to check the tool's reading, and your Arabic-speaking managers will want it.
Full right-to-left support
Check the whole interface, not only the customer page. A dashboard that is left-to-right with Arabic pasted in is harder to use and easier to misread.
Handling of mixed and informal text
Try messages with dialect, with English words inside Arabic sentences and with typos. See what happens.
Voice and video
If customers can speak, check how well speech in each language is transcribed and whether it is then read by the same scheme as typed messages.
A language setting for people and customers
The language of the customer's page and the language of the staff dashboard are different choices. Ask whether each user can pick their own, and whether changing one affects the other.
Honesty about limits
Be wary of a vendor who claims perfect accuracy. Ask where it is weakest, and ask for a trial.
How to test a tool with your own messages
- Collect a sample of real complaints from several branches in both languages. Include short ones, dialect, mixed language and, if you have them, Arabic in Latin letters.
- Have a bilingual person on your team sort them by hand into your categories. This is your reference.
- Run the same messages through the tool.
- Compare. Look at where the tool agrees with your reference, and especially where it does not.
- Check pairs. Find an Arabic and an English complaint about the same problem and see whether the tool groups them.
- Repeat with a fresh sample, so that you are not testing only the examples you have already tuned for.
A small sample is enough to see obvious problems, but it will not tell you accuracy in general. Treat the result as a reason to continue or stop, not as a measured rate.
What not to do
- Do not assume that a tool that handles English well handles Arabic equally.
- Do not translate everything to English first without checking what is lost.
- Do not rely on a vendor's demo data in place of your own messages.
- Do not let each branch categorise its own messages.
- Do not treat a branch with mostly Arabic messages as lower priority because they are read less easily.
Where Outhentik fits
Outhentik reads Arabic and English on one reader, with one way of sorting what customers say, so that a complaint in either language lands in the same issue. The dashboard and the customer's message page work in both languages with full right-to-left layout, and each person picks their own dashboard language. The original message stays behind every finding.
Short typed messages and Arabic written in Latin letters are the hardest cases for any tool, and you should test them with your own messages before relying on the result.
Frequently asked questions
Can one tool read both Arabic and English complaints?
Yes, some can. What matters is whether it uses one scheme for both languages and files the same problem in the same place. Test it with your own messages.
Why not translate everything into English first?
Translation can lose meaning, especially with dialect, idiom and mixed language. If you do translate, keep the original beside it and check a sample.
What about customers who write Arabic in Latin letters?
It is among the hardest cases for any tool. Include examples in your test and see how the tool handles them.
Does right-to-left support matter for managers?
Yes. A dashboard that does not display Arabic properly is harder to read and trust. Check the whole interface.
Can customers speak instead of typing?
In Outhentik, customers can type, speak or record up to 60 seconds. Check how well any tool transcribes speech in the languages and accents your customers use.
How do I compare a branch with mostly Arabic messages to one with mostly English?
Use one scheme and one reader for both, and apply the same minimum evidence rule to every branch. Then the comparison does not depend on language.
How large should my test sample be?
Large enough to include the range of messages you receive: short and long, formal and informal, single-language and mixed, from several branches. A small sample shows obvious failures, but does not measure accuracy in general.
Try Outhentik free for 7 days, no credit card required
Outhentik gives head office a direct line from customers at every branch: a case with an owner for each customer who asks to hear back, every message read the same way, and a daily view of which problems repeat and where. For a business with Arabic and English customers, that means one reader and one scheme, so a complaint in either language lands in the same issue.