How two-way translation keeps cross-border support from losing the sale

Cross-border stores lose conversations in a specific, boring way: a visitor asks in a language the agent on shift does not read, the agent pastes the message into a translator, answers slowly in uncertain English, and the visitor is gone before the second reply. Two-way translation inside the inbox fixes the mechanics of that moment. This article explains how it works in TapCub Chat, how to protect product names and prices from being mistranslated, how to route by language, and what changed in the sample numbers.

Where cross-border conversations die

Look at a cross-border store’s chat transcripts and the pattern is consistent. Conversations in the agent’s language resolve in a few messages. Conversations in another language take twice as long per reply, have more clarifying questions, and end without a resolution far more often. The visitor is not harder to help; the medium is slower, and shoppers do not wait.

The fix is not hiring a speaker of every language. Most stores see a long tail: a dozen languages a month, three of them often. What they need is for any agent to answer any visitor at normal speed.

How two-way translation works

The visitor writes in their language. The inbox detects it, shows the agent the original and a translation side by side, and the agent replies in their own language. The visitor receives the reply translated into theirs, with a one-tap control to see the agent’s original. Every message keeps both versions, so a bilingual colleague can audit a conversation later and nothing is lost in a hand-off.

Detection happens on the first message, and the agent can override it if a visitor switches languages mid-conversation. The visitor-facing widget shows a small “translated” label on each message, which customers tell us they prefer to a pretence of fluency. The sample conversation below is the view from the agent’s side; the AI chat page shows the visitor’s side.

Protecting names, SKUs and prices

Machine translation is good at sentences and bad at proper nouns. Product names get translated literally, SKUs get “corrected”, and “R$ 299” can become something else entirely. TapCub Chat has a per-workspace glossary: terms added there are passed through untouched in both directions. Add product names, plan names, SKUs, currency formats and anything legal. Review the glossary monthly from the “edited after translation” list, which shows the messages agents corrected by hand.

Tone is the other quality question. Machine translation tends to flatten a warm reply into a neutral one. Give agents two or three canned openings and closings per language, reviewed once by a native speaker, and let translation handle the middle. Visitors notice the greeting and the sign-off far more than the sentence about shipping times.

Routing by language and hours

Translation makes every agent capable; routing decides who gets the conversation first. Route by detected language when you do have a native speaker on shift, so the visitor gets the smoothest experience when it is available, and fall back to translation when not. Route by hours across regions so the 3 a.m. conversation lands with the team whose afternoon it is. Both rules are in the team settings on the chat page.

Outside all shifts, the AI assistant answers from the knowledge base in the visitor’s language and collects an email for follow-up. Knowledge base articles are written once and translated on the way out, so you maintain one source of truth.

What changed in the sample store

The table shows a sample cross-border store before and after switching on two-way translation, over two comparable months. First response time for non-native conversations fell by two thirds and resolution rate rose by a third; native-language conversations were unchanged, as expected. The conversion rate of visitors who chatted in a non-native language moved closer to the native rate, which is the number that justifies the feature.

Sample store, two comparable months. Native-language conversations are the control.

Setting it up

Translation is on from the Pro plan and up, and is enabled per workspace with one switch. Add the glossary first, then turn it on, then set routing. Agents need no training: the inbox looks the same, with one extra column. Plans and limits are on the chat pricing page, and the cross-border solution page shows how stores combine translation with analytics by country.

After the first month, read three reports: resolution rate by language, first response time by language, and conversion after chat by language. Any language that lags the others is either missing glossary terms or arriving outside the hours you route for. Both are fixable in an afternoon.