AI Chatbot Development Cost in India: 2026 Pricing Guide

"Chatbot" now spans everything from a scripted FAQ widget to an AI agent that resolves tickets end to end — which is why quotes for "the same thing" range from ₹50,000 to ₹8 lakh. In India in 2026, a scripted bot costs ₹0.5–1.5 lakh, an LLM chatbot grounded in your own documents costs ₹1.5–4 lakh, and an agentic bot that takes actions costs ₹3–8 lakh or more. Running costs add ₹5,000–35,000 a month depending on volume and channels.

What moves you inside a band is rarely the AI. It is integrations, accuracy requirements, and how much of your business knowledge has to be cleaned up before a model can use it reliably. Here is the full breakdown.

The three kinds of chatbot, and why the price roughly doubles each step

1. Scripted bot with LLM fallback — ₹0.5–1.5 lakh. Menu-driven flows covering the twenty questions that make up most of your queries, with a language model handling free text politely instead of dead-ending. Ships in 3–5 weeks and costs very little to run. The right choice for lead capture, store hours, pricing FAQs, and appointment requests that get routed to a human.

2. LLM chatbot with retrieval (RAG) — ₹1.5–4 lakh. The bot answers from your content: product catalogues, policy documents, past support tickets. The engineering here is mostly unglamorous — chunking documents sensibly, building an evaluation set, and suppressing hallucinations. Expect 6–10 weeks. This is the tier most Indian SMBs actually need.

3. Agentic bot — ₹3–8 lakh and up. The bot does things: looks up an order in your ERP, reschedules a delivery, creates a lead in your CRM, sends a payment link. Every action needs an API, guardrails, and failure handling. 10–16 weeks. Worth it when the bot replaces measurable staff hours — not before. Our AI agents guide covers this tier in depth.

AI chatbot cost breakdown (India, 2026)

These reflect competent Indian agency rates in 2026, roughly $25–60 per hour blended. Metro agencies quote materially higher. At the other end, a ₹50,000 "AI chatbot" from a freelancer is usually a thin wrapper around a model API with no evaluation, no guardrails and no admin panel — the hidden costs of cheap software explains how that tends to end.

What moves the price

• Knowledge base readiness. Clean help-centre articles make retrieval straightforward. Answers buried in PDFs, old tickets and the owner's head mean data work before the bot can be trusted.
• Channels. A web widget alone is simplest. Adding WhatsApp, Instagram and voice multiplies integration and testing work.
• Actions. Read-only bots answer questions. Bots that check order status or book appointments need secure API integrations, and each one adds scope.
• Evaluation and guardrails. This is the difference between a demo and something you put in front of customers: test sets, hallucination controls, monitored handoff paths.
• Languages. Hindi, Tamil and other Indian languages are very achievable with modern models, but each adds evaluation work.

What are the monthly running costs of an AI chatbot?

This is the question buyers skip and later regret. In 2026:

• Model inference. A support bot handling around 3,000 conversations a month on a mid-tier model runs roughly ₹3,000–12,000. Heavy retrieval with long context can reach ₹30,000. Model prices keep falling, so an architecture that lets you swap models matters more than today's rate card.
• WhatsApp Business API. Meta bills per conversation in India, with utility conversations costing a fraction of a rupee and marketing conversations costing more; a few thousand conversations typically lands in the low thousands of rupees per month, plus provider platform fees. Check current rates before budgeting — Meta revises them.
• Hosting and vector storage. ₹2,000–15,000 a month. A modest stack (app server plus managed Postgres with pgvector) sits at the low end.
• Maintenance. Budget 15–25% of build cost per year. Bots decay as products, prices and policies change, and someone has to re-evaluate them. See our AMC and maintenance guide.

Realistic all-in for an SMB RAG bot: ₹8,000–30,000 a month. Set against one support agent's salary in tier-2 India, the arithmetic works out once the bot handles even a third of routine queries.

Which channels should your chatbot live on?

For Indian businesses: WhatsApp first, web widget second, Instagram DM third.

WhatsApp is where your customers already are, and a WhatsApp bot gets engagement a web widget never will. It also brings Meta's template approval process, per-conversation billing, and a 24-hour customer-service window your flows have to respect. We cover those specifics separately in the WhatsApp chatbot cost guide.

Voice bots replacing regional-language IVR are the current frontier. They add meaningful cost and per-minute speech charges, and are worth attempting only after a text bot is working well.

Language support matters more than most vendors admit. A bot serving Tamil Nadu retail or Tiruppur's export trade should handle Tanglish — Tamil typed in English script. Modern models do this surprisingly well, but your evaluation set has to include it, or you will find out from annoyed customers instead.

The hidden cost: your knowledge base

The most common budget overrun in chatbot projects is not code. It is content.

Retrieval bots are only as good as the documents behind them, and most businesses discover their "documentation" is several contradictory PDFs, a price list from last year, and knowledge that lives in one person's head.

Budget two to four weeks of internal effort, or ₹50,000–1.5 lakh of vendor effort, for knowledge base cleanup: writing canonical answers, removing contradictions, structuring product data. Skipping this is the single biggest reason cheap chatbots hallucinate.

It is also why an evaluation set matters — 100–300 real questions with verified answers, assembled before any tuning. Accuracy you cannot measure is accuracy you do not have.

How do you calculate chatbot ROI?

A simple model that holds up in practice:

1. Count monthly inbound queries across phone, WhatsApp and email. Say 4,000.
2. Estimate the share a bot can fully resolve. An honest number for a good retrieval bot is 40–60%.
3. Multiply by handling time (3–5 minutes) and loaded staff cost.

Four thousand queries at 50% deflection, four minutes each, is around 130 staff hours a month — plus after-hours coverage and faster lead response, which matters because leads answered within minutes convert far better than leads answered the next morning.

Against a ₹3 lakh build and ₹20,000 a month running cost, payback typically lands inside a year on deflection alone, sooner if the bot also captures enquiries you currently lose overnight. Model your own numbers in the project cost calculator.

Build, buy, or SaaS?

There is a legitimate middle path between doing nothing and a custom build: SaaS chatbot platforms. Yellow.ai, Haptik and Verloop sit at the enterprise end; Wati and AiSensy serve SMBs. Indian SMB tiers typically run ₹5,000–50,000 a month.

They are genuinely good for standard flows — order updates, appointment reminders, catalogue browsing on WhatsApp.

Where they run out of road: answering from proprietary documents with measurable accuracy, deep write access into your ERP or CRM, custom guardrails for regulated industries, and per-conversation economics at high volume.

The crossover we see: once you are paying ₹30,000 or more a month to a platform and still routing complex queries to humans, a custom build starts paying for itself within one to two years — and you own the asset. The same arithmetic as our custom CRM cost guide: subscriptions are cheap to start and expensive to stay.

Compliance: the DPDP Act and your chatbot

Chatbots collect personal data — names, phone numbers, sometimes health or financial details. Under the Digital Personal Data Protection Act, that makes you a data fiduciary.

Practical implications: state clearly what the bot collects, do not send personal data to third-party model APIs without a data processing agreement, log consent, and honour deletion requests.

Ask any chatbot vendor a simple question before signing: where do the transcripts live? For regulated clients — clinics on our Clinic CRM, lenders on LoanKard — we keep transcripts in the client's own database rather than a model vendor's. Our AI data privacy playbook covers the implementation detail.

Five chatbot mistakes that waste money

1. Launching without an evaluation set. If you cannot state the bot's accuracy as a number, you shipped a demo.
2. Letting the bot pretend to be human. Users forgive a bot that admits it is a bot. They screenshot one that lies.
3. No human handoff. Every flow needs an exit to a person during business hours, and a clear promise of when after hours.
4. Automating before documenting. A retrieval bot over contradictory documents amplifies the contradictions at scale.
5. Ignoring the transcripts. Real conversations are a free, continuously updated product roadmap. Reading a sample weekly is what separates bots that improve from bots that plateau.

What a realistic project timeline looks like

• Weeks 1–2. Query analysis against your actual WhatsApp and call logs, scope the top intents, define the evaluation set.
• Weeks 3–6. Build the flows, retrieval pipeline and integrations; wire up channels.
• Weeks 7–8. Evaluation loop — measure accuracy against the test set, fix retrieval, tighten prompts.
• Weeks 9–10. Pilot with a slice of traffic, tune the handoff, then launch fully.

Vendors who skip the evaluation weeks ship demos rather than products.

Is a chatbot worth it for your business?

Above roughly 300 support conversations a month, usually yes. Below that, a well-structured FAQ page is the better first investment, and we will tell you honestly which side of the line you are on.

Send us twenty real customer questions from your WhatsApp and we will tell you which tier you need, what it will cost, and what it will not be able to do. Talk to us, or read the broader AI MVP cost guide for how chatbots fit alongside other AI builds.

For how this sits against every other kind of build — websites, apps, SaaS, CRM, ERP and AI, priced side by side — see our software development cost guide for India.