Customers want answers immediately, while support teams answer the same questions every day. An AI chatbot for customer service connects those two needs. Unlike menu-based bots, chatbots built on large language models understand free-form questions, answer in natural language from your company knowledge and hand the conversation to a person when needed. This guide walks through setting one up on your website and on WhatsApp.
AI chatbot vs. rule-based chatbot
| Rule-based bot | AI chatbot | |
|---|---|---|
| Understanding | Keywords and button menus | Free text, typos, varied phrasing |
| Knowledge | Hand-written answer trees | FAQs, policies, product data (RAG) |
| Maintenance | New flow for every question | Often just update the documents |
| Risk | Low but rigid | Flexible; needs guardrails and testing |
A hybrid usually works best: fixed flows for critical actions such as starting a return or booking, AI for informational questions.
Which questions should the chatbot handle?
Start by reviewing your support history; requests tend to cluster around a few recurring topics:
- Order and delivery status
- Returns, exchanges and warranty
- Product features, stock and compatibility
- Appointments and reservations
- Opening hours, locations and contact details
Complaints, legal matters and emotionally sensitive cases should go to a human from the start.
Technical architecture
Website and WhatsApp channels should share the same brain. A typical setup includes a channel layer (web widget, WhatsApp webhook, in-app chat), a conversation manager for sessions, identity and handoff rules, an AI layer with system instructions and RAG search, secure integrations with order, CRM or booking systems, and an admin panel for monitoring, handoff and knowledge base updates. One knowledge base update then applies to every channel.
Off-the-shelf chatbot platforms can be a quick start, while a custom build gives full control over integrations, data retention and brand experience.
Website and WhatsApp chatbot setup
Both channels share one backend, but each has its own setup steps and rules.
Website AI chatbot
- Knowledge base: FAQs, policies, catalog and help articles are loaded into a RAG pipeline.
- Integrations: secure API connections to e-commerce or ERP for personal queries, with customer verification before sharing any details.
- Performance: load the widget lazily so it does not slow the page.
- Brand voice: tone, language and boundaries are defined in system instructions.
On Next.js sites we build the chat UI and server side together; see our web development services.
WhatsApp AI chatbot
For business automation you need Meta's WhatsApp Business Platform (Cloud API), not the free WhatsApp Business app. Main steps:
- Verify your Meta Business account and register a phone number.
- Deploy a webhook server to receive incoming messages.
- Connect the webhook to the AI layer and knowledge base.
- Prepare approved message templates for notifications outside the 24-hour window.
- Add handoff to a live agent panel and store conversation history.
Meta's messaging policies and conversation-based pricing apply and change from time to time, so check the current terms at the start of the project.
Why human handoff matters
A good chatbot knows when to stop. Hand off when the customer asks for a person, when the bot fails twice, when a complaint or cancellation is detected, or when no reliable source is found. Pass a conversation summary to the agent so the customer does not repeat themselves.
Outside business hours the bot should say so, log the request and tell the customer when to expect a reply, so your team starts the next morning with a summarized, prioritized queue instead of lost messages.
Privacy and security
Conversations often include names, phone numbers and order numbers. Make the privacy notice accessible, define retention periods, minimize and mask data sent to the model provider, and never reveal account details without verification. GDPR or KVKK requirements apply depending on where you operate.
Guard against users trying to push the bot outside company policy: combine system instructions, content filters and hard business rules so the bot cannot, for example, promise discounts or approve refunds on its own.
Measuring success
- Resolution rate without handoff
- Answer accuracy on sampled conversations
- Customer satisfaction ratings
- First response time and change in agent workload
Review unanswered questions regularly and add them to the knowledge base. The same backend can also power a chat inside your mobile app.
BernSoftware builds AI chatbots for websites, WhatsApp and mobile apps end to end. Visit our AI solutions page or contact us to discuss your support needs.
Frequently asked questions
Is the free WhatsApp Business app enough for an AI chatbot?
No. Automated AI replies and system integrations require Meta's WhatsApp Business Platform (Cloud API).
What if the chatbot gives wrong information?
The bot is restricted to approved sources, hands off when unsure, and conversations are sampled and reviewed regularly to catch errors.
Can one chatbot handle several languages?
Yes. Current language models work across languages, but the knowledge base should contain accurate content in each language and testing should cover all of them.
Planning a project like this?
Plan it in 10 steps