Multilingual support without a multilingual team
Published 16 September 2026 · Replio
Adding a language to your support is usually treated as a hiring problem. It is really a coverage problem, and coverage is what makes the hiring so expensive. Here is the arithmetic most teams do too late, and what can honestly be automated before you post the job.
What "we support three languages" really costs
A language is not a person. It is a rota. If you promise a same-day reply in English, Arabic and French while your desk is open, you need someone who speaks each of those languages present for every hour the desk is open, every working day, including the days people are sick, on leave, or already on another call.
Put your own number to it. Call the fully loaded monthly cost of one agent C: salary, tools, management, the recruiting you did to find them. One language during office hours, with nobody covering the gaps, is C. Three languages is 3C, and every one of those three is a single point of failure the first time somebody takes a week off. Give any of it night and weekend cover and you are multiplying by shifts, not adding a person.
The part that stings is that the volume rarely justifies it. Pull your last 30 days and count messages by language before anything else. Most businesses that "need" a third language find it is a small share of the inbox, heavily weighted toward the same handful of questions: opening hours, price, delivery, is this in stock, where is my order. That is not a job. That is a rota you are paying for to cover a trickle.
Translation tools help the agent, not the clock
The usual first fix is to put translation in front of the human desk, so an English-speaking agent can read a Portuguese message and answer it. That is a genuine gain and worth having. It means one team can serve everyone instead of routing by language, and it kills the "sorry, our Arabic colleague is back Monday" reply.
What it does not do is change when the reply happens. A translated inbox still needs an agent awake, logged in and free. The queue is the same length, the night is still unstaffed, and the customer who wrote at 11pm still waits until morning. Translation solves comprehension. It does not solve availability, and availability is what the second and third language actually cost you.
This is worth being clear about because both things get sold under the same word. Replio's own Live Translate is the agent-side kind: a shared inbox your team reads and replies in, in any language. The AI answering customers in their own language is the other kind, and it is the one that covers the hours nobody is at the desk.
Three ways to cover three languages
| Hire per language | Translate a human desk | AI answers in each language | |
|---|---|---|---|
| Hours covered | Whenever that person works | Whenever anyone works | Every hour, no rota |
| Time to first reply | Queue plus availability | Queue | Seconds |
| Adding a fourth language | Another hire | No new hire | Nothing to do |
| What you maintain | A team per language | One team | One knowledge base |
| Cost shape | Grows with languages and shifts | Flat per seat | Flat per plan |
| Single point of failure | One speaker per language | Spread across the team | None |
| Handles judgement calls | Yes | Yes | Hands off to a person |
The three are not alternatives so much as layers. The AI takes the repeat volume in every language at any hour, the translated inbox lets whoever is on shift pick up what the AI escalates, and the hire is reserved for the language where the conversations genuinely need a native speaker.
Write the knowledge once, not once per language
The hidden cost of multilingual support is content, not headcount. Teams that localise properly end up with a help centre per language, and every price change, policy tweak and new product has to be pushed through all of them. The versions drift, and the language with the least traffic gets the stalest answers.
An AI agent inverts that. You write your business knowledge once, in the language you run the business in, and it answers from that in whichever language the customer wrote. Adding a fourth language is not a fourth content project. In practice this is what makes the whole thing affordable:
- One knowledge base, every channel. The same answers serve WhatsApp, Instagram DMs, Messenger, Telegram and your website widget, so a customer gets the same answer wherever they wrote.
- Detection from the message, not the country. The language is read from what the person actually typed, which matters in any market where the phone number and the language do not line up.
- The widget can speak the language too. The website widget's own interface can be switched, including full right-to-left Arabic, so the buttons match the replies instead of framing them in English.
- Escalation stays multilingual. When the AI hands over, the conversation lands in a shared inbox your team can read and reply in, whatever language it is in.
The one thing to do by hand: anything you need worded exactly, such as a refund policy or a regulated line, should be written the way you want it read and checked by someone who speaks the language. That is a short list, and it is a one-off.
What still needs a native speaker
Automating the trickle is not the same as claiming you do not need people. The conversations that should still reach a human who speaks the language properly are fairly consistent:
- Complaints where tone decides the outcome. A technically correct reply in a language you do not control can read as cold or dismissive.
- Negotiation and anything bespoke. Pricing conversations, custom scopes, and a customer talking themselves out of leaving.
- Regulated or legal wording. If a sentence has to be exact in that market, a person signs it off.
- The language that turns out to carry real volume. If one language is a third of the inbox and half the revenue, hire for it. The point of the arithmetic is to hire where it pays, not to avoid hiring.
There is also an honest limit worth naming: machine quality varies by language. Where the output in a language is clearly worse than a language both sides read comfortably, the better answer is the one the customer can actually use, not the one in the flag they expected. More on the mechanics of that on the multilingual AI support page.
What this costs with Replio
Languages are not metered, and there is no per-language fee. Every plan answers in 50+ languages, detected automatically, with no setup. The plans differ by which channels they cover, how many team seats you get, and how many AI replies are included.
The free plan is $0/month (AI on Telegram and the website widget). Starter is $49/month for one channel of your choice plus Telegram and the widget, Scale is $149/month for WhatsApp, Instagram and Messenger together, and Pro is $349/month for every channel with 15 seats. Messages are unlimited on every plan. Live Translate, the multilingual team inbox, is included free on Scale and Pro and is a $19/month add-on on Starter. There is a free 7-day trial on Starter, no card required.
Set that against the rota. A second language covered by a person is another C every month, before you have covered a single night.
A checklist before you hire
- Count the inbox by language for the last 30 days. Decide on the number, not the impression.
- Split each language into repeat questions and judgement calls. The first pile is what automation is for.
- Decide the hours you are actually promising in each language, and price the rota honestly, cover included.
- Automate the repeat pile everywhere first and let it run for a few weeks before you decide anything else.
- Read what escalated. The languages that keep producing real conversations are the ones worth a hire.
- Hire for the gap that is left, which is usually one language rather than three.
Answer every language, at every hour
Connect WhatsApp, Instagram, Messenger or Telegram and let an AI agent answer in each customer's own language, 24/7, from one knowledge base. No card required.
Start free trial