Table of contents
A simple scripted FAQ chatbot typically costs ₹3L–₹8L. Add a real LLM behind it and the range moves to ₹5L–₹15L. Ground it in your own documentation with retrieval (RAG) and the range moves to ₹8L–₹25L. A multi-channel, multilingual enterprise chatbot with live-agent handoff and analytics can run ₹15L–₹35L or more. The gap between these bands has almost nothing to do with how the chat bubble looks and everything to do with what the chatbot actually needs to know and do.
This article breaks the cost down by chatbot type, covers what actually drives the price within each band, and gives a build-vs-buy framework so you're not paying custom-development rates for a problem an off-the-shelf platform already solves well.
- Scripted / rule-based bot
- ₹3L – ₹8L
- LLM-powered conversational bot
- ₹5L – ₹15L
- RAG-grounded chatbot (your own data)
- ₹8L – ₹25L
- Multi-channel enterprise chatbot
- ₹15L – ₹35L+
Indicative ranges — the specific number depends on channel count, data readiness and how much of the conversation needs to be grounded in your own content.
What kind of chatbot are you actually building?
"Chatbot" covers a wide spectrum, and most cost confusion comes from comparing quotes for genuinely different things.
| Chatbot type | What it does | Approximate cost |
|---|---|---|
| Scripted / rule-based | Decision-tree flows, fixed responses, no real language understanding | ₹3L – ₹8L |
| LLM-powered (no retrieval) | Natural conversation, but no grounding in your own data | ₹5L – ₹15L |
| RAG-grounded chatbot | Answers from your documentation, product data or knowledge base | ₹8L – ₹25L |
| Multi-channel enterprise chatbot | Web, WhatsApp, and/or voice, with live-agent handoff and analytics | ₹15L – ₹35L+ |
| Chatbot with transactional capability | Can take actions (bookings, order status, refunds) — effectively a scoped AI agent | ₹15L – ₹35L |
What actually drives the cost within each band
Conversation and data readiness
An LLM-powered bot with no grounding is fast to build because it doesn't need a data pipeline. The moment you want it to answer from your own documentation, pricing, or product catalog, you're building the same ingestion and retrieval pipeline covered in our RAG development guide — chunking, embeddings, vector search — which is where most of the cost in a "smart" chatbot actually goes.
Channels
A chatbot confined to your website is simpler than one that also needs to work over WhatsApp, SMS, or voice — each channel has its own API, message-format constraints, and testing surface. WhatsApp specifically requires working within Meta's Business API constraints (template messages, session windows), which is a real integration effort, not a checkbox.
Escalation and live-agent handoff
A chatbot that can recognize when it's out of its depth and hand off to a human agent — with the conversation context intact — is significantly more useful (and more expensive to build well) than one that just fails silently or loops.
Multilingual support
Supporting multiple languages well means more than translating static strings — retrieval, response generation and any transactional logic all need to work correctly across languages, which adds real testing surface, particularly for languages with less LLM training data.
Analytics and continuous improvement
Understanding what users actually ask, where the bot fails, and how to improve it over time requires logging and analysis infrastructure — the same observability discipline covered in our LLM observability guide, scaled to a chatbot's specific needs (conversation-level analytics, drop-off tracking, escalation-rate monitoring).
Timeline
| Type | Typical timeline |
|---|---|
| Scripted / rule-based | 3–6 weeks |
| LLM-powered (no retrieval) | 5–8 weeks |
| RAG-grounded chatbot | 8–14 weeks |
| Multi-channel enterprise chatbot | 12–20 weeks |
WhatsApp is usually the highest-ROI channel to start with
For most Indian and UAE businesses, customers are already on WhatsApp — meeting them there tends to produce higher engagement than a website-only widget. If you're specifically automating WhatsApp conversations, campaigns or customer support, see BotBridge, built around WhatsApp Business API workflows directly rather than as an afterthought channel bolted onto a generic chatbot.
Build vs. buy
Not every chatbot need justifies a custom build.
| Situation | Recommended approach |
|---|---|
| Generic FAQ deflection with well-known question patterns | Buy — an off-the-shelf chatbot platform |
| Needs to answer from your specific, changing documentation | Custom RAG-grounded build, or a platform with a genuine RAG feature |
| Needs deep integration with your CRM, order system, or internal data | Custom build |
| Needs to take real actions (bookings, refunds, order changes) | Custom build — effectively an AI agent, not a chatbot |
| Standard e-commerce or support use case with mainstream tooling available | Buy, unless differentiation genuinely requires custom logic |
Planning a chatbot or conversational AI project?
Talk to our engineering team about scope, channels and realistic cost before you commit to a build.
Common mistakes
- Treating launch as the finish line. A chatbot's accuracy and usefulness depend on ongoing tuning against real conversations — budget for iteration, not just a one-time build.
- No defined escalation path. A bot that can't recognize its own limits and hand off to a human frustrates users faster than having no bot at all.
- Skipping analytics. Without visibility into what users actually ask and where the bot fails, you're optimizing blind.
- Underestimating channel-specific constraints, especially WhatsApp's messaging-window and template rules, which materially affect what conversational flows are even possible.
For a broader view of where chatbots fit against other AI investment options, see our AI development cost guide.
Frequently asked questions
How much does it cost to build an AI chatbot?+
A simple scripted bot costs ₹3L–₹8L. An LLM-powered conversational bot runs ₹5L–₹15L. A RAG-grounded chatbot answering from your own data runs ₹8L–₹25L. A multi-channel enterprise chatbot can run ₹15L–₹35L or more.
What's the difference between a scripted chatbot and an AI chatbot?+
A scripted chatbot follows fixed decision-tree flows with no real language understanding. An AI (LLM-powered) chatbot understands natural language and can hold a genuine conversation, optionally grounded in your own data via retrieval.
Why does a chatbot that answers from my own documents cost more?+
Because it requires a retrieval pipeline — document ingestion, chunking, embeddings and vector search — not just a conversational interface. This is the same RAG architecture used in AI knowledge assistants generally.
Should I build a custom chatbot or use an off-the-shelf platform?+
Off-the-shelf platforms work well for generic FAQ deflection and mainstream use cases. Custom builds are worth it when you need deep integration with your own systems, grounding in your specific data, or the ability to take real actions rather than just answer questions.
How much does a WhatsApp chatbot cost specifically?+
WhatsApp adds integration effort on top of the base chatbot type — working within Meta's Business API constraints (templates, session windows). Costs generally track the same bands above, with WhatsApp-specific integration as an added line item.
How long does it take to build a chatbot?+
A scripted bot typically takes 3–6 weeks. An LLM-powered bot takes 5–8 weeks. A RAG-grounded chatbot takes 8–14 weeks. Multi-channel enterprise chatbots typically take 12–20 weeks.
When does a chatbot actually become an AI agent?+
When it stops only answering questions and starts taking real actions — processing a refund, making a booking, updating a record. At that point it's architecturally an agent, with the corresponding cost and guardrail requirements covered in our AI agent guides.
Written by
CodeSurge AI Engineering Team
The CodeSurge AI team designs and builds AI systems, SaaS products and enterprise integrations for clients in India, the UAE and beyond — this section shares the architecture patterns, cost drivers and implementation tradeoffs we work through on real projects.