A classic call centre has three problems you cannot solve by hiring more people: it is expensive, it speaks 1-2 languages, and it closes at 5 pm. Meanwhile customers write from a different time zone, in a fifth language, and expect an answer to a question that is already in your terms.
The solution is not a bigger support team. One AI chat that translates conversations on the fly and answers from your knowledge base, escalating the harder ones to a human, is enough. That is exactly what enuchat.com does - a chat widget that deploys with one line of JavaScript and serves customers in 60+ languages.
Five problems with a classic call centre
1. Cost per seat
A support agent in Poland costs PLN 6,000 - 9,000 gross/month (with overheads, equipment, training). A three-person shift covering 8 hours is about PLN 25,000/month - PLN 300,000 per year for a single time window.
2. Working hours
A classic call centre runs 9-5. E-commerce customers shop in the evenings and at weekends. B2B customers in Asia write while Poland sleeps. Every missed call and every unanswered email until morning is a potentially lost order.
3. Language barrier
Most Polish teams serve customers in Polish and English - sometimes German. Customers from Ukraine, Czechia, Spain or Norway get either "in English please" or nothing. Every missing language is a specific market segment you do not serve.
4. Repetitive questions eat the day
70-80% of support questions are the same dozen: order status, opening hours, returns policy, product availability, prices, location. Each takes a person 1-3 minutes to answer when the answer is already in the terms or FAQ - the customer just did not look.
5. No conversation memory
A customer writes the same thread to three channels (website form, Messenger, email). Each agent starts from zero, asks for an order number, reads the history. The customer feels the company "does not remember" and on the third attempt gives up.
AI chat with 60+ language translation - how it works
enuchat.com is a chat widget built around three things: AI from your knowledge base, real-time translation, and integration with your systems.
Two-way translation. A customer writes in Ukrainian, the operator replies in Polish - AI translates both ways on the fly. The operator sees the conversation in Polish, the customer in Ukrainian. No plug-ins, no copy-paste, no Google Translate in a second tab.
AI from the knowledge base. You upload your terms, returns policy, product descriptions, FAQs (PDF, DOCX, Markdown, or directly from URLs of your documentation). The assistant is built on Claude, uses semantic search to find the right passage and phrases the reply in the customer's language. It does not search the internet and does not hallucinate - if the answer is not in the knowledge base, it escalates to a human.
IF/THEN rules engine. Routing by language, alerts on keywords ("complaint", "cancel"), calls to external APIs. Configured in the panel, no code.
What AI handles automatically
From our experience (deployments in e-commerce, hospitality, SaaS) 60-80% of questions can be handled without a human if the knowledge base is well structured:
- opening hours, location, directions,
- returns policy, complaints, delivery,
- prices and product availability (with API integration - live from your system),
- order status (with integration - without waiting for an operator),
- technical specification questions from documentation,
- allergens, diet, ingredients (hospitality, cosmetics).
The customer gets an answer immediately and in their own language, regardless of the time of day.
What escalates to a human
The rest - judgement calls, non-standard, emotional - goes to the operator with the full conversation context. This is not "AI failed"; this is the correct split: AI handles what is repeatable, a human handles what requires judgement.
Typical escalation scenarios:
- a return request outside the policy ("I bought 2 months ago, can I still return?"),
- a quality complaint that needs a decision,
- a request for a custom B2B offer,
- any conversation in which the customer raises a legal, health or privacy issue.
The operator sees the full history of the conversation (with translation), does not start from zero, and replies in their language - the customer sees the reply in theirs.
Live data - order status without switching windows
The strongest feature for e-commerce support: live data in the chat. enuchat connects to your API (orders, invoices, availability, reservations) and pulls current data at the moment of the conversation.
A customer asks "where is my order #12345" - AI checks the status in your system, names the specific delivery stage and (if a tracking number is available) links to the courier. No window switching, no waiting, no "let me check and get back to you".
The same works for:
- current product or variant availability,
- price depending on configuration (a calculator),
- reservation status in a restaurant or hotel,
- account balance or loyalty points.
Deployment - one line of JavaScript
The technical rollout is trivial: you add one script to the page <head> and the widget appears immediately. It works on any CMS (WordPress, Shopify, PrestaShop, custom frameworks) - no plug-ins, no changes to the rest of the code, no impact on Core Web Vitals (the widget loads asynchronously, under 50 KB gzip).
End-to-end deployment takes 1-2 days operationally:
- Upload the knowledge base (terms, FAQ, policies, product descriptions).
- Configure languages for greetings and conversation routing.
- (Optional) API integrations with your shop or reservation system.
- Test with the support team in the operator panel.
The free plan works without a credit card - you can start on one page and validate the result before committing to a full rollout.
ROI - when it pays off
A simple sum for a typical e-commerce site:
- Classic call centre (3 people × 8h × 1 shift) - about PLN 300,000 / year.
- AI chat with one part-time operator - about PLN 60-80,000 / year (licence + a part-time operator).
- Savings: ~PLN 220,000 per year, plus 24/7 coverage and 60+ languages.
At smaller scale (small shops, hospitality venues, clinics) the AI chat alone is often enough - an operator only steps in for incidental cases.
GDPR for chat-based support
A baseline worth getting right. enuchat is GDPR-compliant: conversation data is encrypted, retention is configurable (30 days, 90 days, a year), export and deletion under Art. 15 and 17 GDPR are automated, hosting is in the EEA. Specific processors (Anthropic for Claude, the translation provider) are listed in the enuchat.com privacy policy - you can reference them from your own policy.
If you are rolling out AI in customer support, take the opportunity to close the privacy notice loop too - Art. 13 GDPR requires informing the customer that their conversation is processed by AI. A short sentence below the chat ("Conversations are handled with the help of an AI assistant. Details in the privacy policy.") is enough. More on this in ChatGPT, Copilot and GDPR.
Next step
If the call centre costs more than it brings in, or the language barrier cuts off a segment of your customers, take a look at enuchat.com - the AI chat widget. The free plan does not require a card - you can evaluate before you decide.
If in doubt - a free conversation takes 30 minutes. We will show a demo, estimate whether AI will cover 40, 60 or 80% of your traffic, and what that does to costs. Related: Restaurant without language barriers - menu in any language, because translating guest contact works on both sides - the menu and the chat.