Modern AI chat changes the math: 50+ languages handled natively, enabled with a toggle rather than a translation project. Here is the five-step playbook for turning that toggle into revenue.
Step 1: Find your top non-English geos in analytics
Don’t guess. Pull 90 days of sessions by country and browser language, and pick the two or three non-English geos already sending meaningful traffic — for a US or UK store, typically Germany, France, Spain, or the Netherlands. Three is the sweet spot: enough to move the needle, few enough to execute well.
Note the conversion gap per geo while you’re there. That number is your baseline for Step 5 — and the figure that justifies the project to anyone who asks.
Step 2: Enable languages with a toggle, not a project
You don’t translate your store. You switch the languages on in your chat agent.
Oscar Chat, for example, ships with 50+ languages out of the box: it detects the visitor’s language and responds in it, mid-conversation, with no string files to sync and no per-language QA pass. The model does the understanding and the answering in real time.
The mental shift: you’re localizing conversation, not content. Product pages can stay in English — shoppers tolerate that far better than unanswered questions. Chat is where doubt gets resolved, and that’s the moment that needs their language.
Step 3: Translate the 20 intents that cover most pre-sale questions
The toggle gets you 80% there. The remaining 20% deserves real care: the answers that must be exactly right.
Pre-sale chat clusters around the same twenty intents everywhere: shipping cost and time, returns policy and return shipping, sizing and fit, stock and restocks, discount codes, payment methods and security, order tracking, compatibility, materials and care, warranty, gift options, taxes and duties, and escalation to a human.
Get these twenty answers professionally reviewed in each target language — by a native speaker who shops online in that market, not just a bilingual colleague. Twenty intents times three languages is sixty short answers: a freelancer finishes in days, for a fraction of one agent’s monthly salary. The bot handles the long tail of unusual questions from your knowledge base; you polish the vital few.
Step 4: Localize the money answers
Translation gets the words right; localization gets the facts right — and the facts that kill international sales are almost always about money.
A German shopper wants shipping terms for Germany: days to Munich, cost, carrier, customs handling. A French shopper wants to know whether the return address is in France. A Spanish shopper wants to know if duties are included or a surprise at the door. No model can guess these from your English FAQ.
For each target geo, pin down four facts and make the bot state them unprompted:
- Shipping time, cost, and carrier for that country.
- Duties and taxes — included, or paid by the buyer on delivery.
- Return address and return shipping cost for that country.
- Delivery problem handling — who to contact, what the remedy is.
Write them per geo, feed them to the bot as ground truth, and you’ve removed the four most common reasons international shoppers abandon.
Step 5: Measure conversion by geo, before and after
After 30 days, compare each target geo against its Step 1 baseline — and against a geo you didn’t enable, to rule out seasonality. Watch four numbers:
- Conversion rate by geo, before vs. after — the headline metric.
- Chat engagement by geo — if non-English visitors aren’t opening chat, the problem is discovery (reposition the prompt), not language.
- Resolution without escalation — rising in-chat resolution means your twenty intents are working.
- Average order value by geo — guided selling in the shopper’s language often lifts basket size too.
For stores with meaningful EU traffic, this play often pays for the entire chatbot on its own — and unlike ad spend, the improvement compounds with every future visitor from that geo.
Why this matters beyond commerce: a founder’s note
Oscar Chat wasn’t built for stores first. It was built so patients could get answers — I started it while undergoing cancer treatment, after seeing how hard it is to get a straight answer from a clinic when you’re scared and the stakes are high.
Many clinics serve international patients who don’t speak English, Spanish, or Dutch. A patient arriving for treatment in a country whose language she doesn’t speak faces the same barrier as that German shopper — except her unanswered question isn’t about shipping times. We presented the project to Dutch clinics to explore how a chat agent speaking 50+ languages could bridge that distance: a patient asking about her appointment or medication, in her own language, at midnight, when no human is there to answer.
Commerce is where most of our customers live today, and this playbook is written for them. But the idea underneath is the same: no one should be left without answers because of the language they happen to speak.
Frequently asked questions
Do I need to translate my whole website for this to work?
No. English product pages plus multilingual chat is enough — shoppers buy from English pages once their specific questions get answered in their language. Full-site translation is a separate, far larger project; do it later if the geo data justifies it.
How many languages should I enable at the start?
Two or three — your top non-English geos. Each language multiplies the care work in Steps 3 and 4, so nail three geos, measure the lift, then expand.
Will the AI make mistakes in languages I can’t verify?
It can — which is why Step 3 has every high-stakes answer professionally reviewed. Set the same guardrails as in English: the bot escalates gracefully when unsure instead of guessing. Review a sample of non-English conversations weekly at first; you’ll quickly see whether quality holds.
What about different scripts or right-to-left languages?
The widget renders Unicode natively, so Arabic, Hebrew, Japanese, and other scripts display correctly with no special setup. For RTL markets, verify the chat bubble mirrors properly with one test conversation before launch.
How is this different from adding Google Translate to my site?
Translated pages don’t answer questions — they just render static text. The conversion problem isn’t “I can’t read the page,” it’s “I have a specific doubt and no one will answer it.” Chat resolves the individual worry at the moment of hesitation, in the visitor’s language. That’s a different job than translation.
What does it cost versus hiring native-speaking agents?
One agent per language per shift is a full salary each. The chat route is a platform subscription plus a one-time freelance review of twenty intents per language — days of work, not months of salary. And the bot covers every time zone from day one.
Enable 50+ languages in your chat today — no translators, no staffing project, no per-language QA.
Try Oscar Chat free
