WhatsApp AI chatbot development

What a WhatsApp AI chatbot does

A WhatsApp AI chatbot is an assistant on your business WhatsApp number that understands what customers write and replies from your own information. Unlike a button menu, it handles free-text questions. Unlike a general chatbot, it only answers from content and systems you approve, and it hands the conversation to a person when it should.

Azimuthforge builds these chatbots on the official WhatsApp Business Platform, using the same retrieval and evaluation practices as our RAG and LLM applications.

Official platform vs unofficial tools

	Official WhatsApp Business Platform	Unofficial automation tools
Allowed by WhatsApp	Yes	No, it breaks WhatsApp's terms
Risk to your number	Low, when messaging rules are followed	High risk of the number being banned
Verified business profile	Available	Not available
Business-initiated messages	Through Meta-approved templates	Unreliable and against policy
Scale	Built for high volumes and multiple agents	Limited to one phone or session

Rule-based, AI or hybrid?

	Rule-based (menus and buttons)	AI chatbot	Hybrid
Understands free text	No	Yes	Yes
Predictable flows (payments, bookings)	Strong	Needs guardrails	Strong
Answers open questions	Only pre-written ones	From your documents	From your documents
Setup effort	Low	Medium	Medium
Best for	Simple, fixed journeys	Support and information	Most businesses

We usually recommend a hybrid: structured flows for things that must be exact, such as booking a slot or confirming an order, and AI for open questions.

What a WhatsApp chatbot can handle

• Questions from your documentation: policies, pricing pages, product details, opening hours, FAQs.
• Order, booking and delivery status, looked up live in your systems.
• Lead capture and qualification, with details pushed to your CRM.
• Appointment and table booking, with confirmations and reminders.
• Notifications and one-time passwords through approved templates.
• Hand-off to a person, with the full conversation history for your team.
• Hindi, Hinglish and other languages, replying in the language the customer uses.

Example conversations

These examples show how a hybrid WhatsApp chatbot behaves. Details change with each business.

Restaurant or cloud kitchen. A customer asks whether a dish contains nuts. The chatbot answers from the menu and allergen sheet, then offers to start an order. If the customer asks about a large party booking, it collects the date and headcount and hands the conversation to the manager.

Clinic or service business. A patient asks for the next available appointment. The chatbot checks live availability, offers slots, books the chosen one and sends a reminder template the day before. Medical questions are always handed to staff.

D2C brand. A customer asks where their order is. The chatbot looks up the order by phone number, shares the tracking status, and, if a return is requested, explains the policy and opens a return request for the support team.

B2B sales. A prospect asks about pricing for a service. The chatbot answers from approved material, asks qualifying questions such as company size and timeline, and books a call with the sales team, sending the details to the CRM.

Messaging rules we design around

• 24-hour customer service window. After a customer messages you, free-form replies are allowed for 24 hours.
• Templates for everything else. Business-initiated messages outside that window must use templates approved by Meta.
• Opt-in. Customers must agree to receive messages from your business, and we record that consent.
• Quality rating. Meta monitors how customers respond to your messages; relevant, expected messages keep your number in good standing.

Proof from our own platform

Restrofi, the restaurant and hotel operations platform we operate, sends WhatsApp order updates and one-time passwords through the WhatsApp Business Platform. The template, consent and delivery-tracking practices on this page come from running that in production.

Integrations

We connect the chatbot to the systems that hold the answers: CRMs, order and booking systems, point-of-sale and inventory systems, payment links, help desks and internal databases, through their APIs. See custom AI development for AI agents that can also take actions in these systems with human approval.

How a WhatsApp chatbot project works

1. Discovery. We list the conversations to automate, the systems involved and when to hand off to a person.
2. WhatsApp setup. Business verification, phone number and message templates on the WhatsApp Business Platform.
3. Knowledge and flows. Your documents are indexed for AI answers, and structured flows are built for exact tasks.
4. Evaluation. The chatbot is tested against real customer questions, in each language you need.
5. Launch and hand-off setup. Your team gets an inbox to take over conversations.
6. Monitoring. Conversations are reviewed to improve answers and add missing content.

What drives the cost

• The number of conversation types and the systems they connect to.
• Languages and the volume of content the chatbot must answer from.
• Hand-off and inbox requirements for your team.
• Running costs: Meta's charges for template messages, billed directly by Meta, plus model and hosting costs that grow with message volume.

The build is priced from a written scope after discovery, so you approve a fixed price before work starts.

Privacy and consent

Conversations contain personal data, so we design for India's Digital Personal Data Protection Act, 2023: consent is recorded, data is kept in your cloud account, access is limited to the people who need it, and conversation logs follow a retention period you set.