Decipher
Book a Call Get Started
Home / AI Solutions / Chatbots / Cost Guide
Guide · Pricing

AI chatbot development cost in India: 2026 pricing guide

If you're getting quotes ranging from ₹50,000 to ₹50 lakh for what sounds like the same chatbot, you're not alone. Vendors are quoting different products under the same name. This guide walks through the six cost drivers, four honest pricing tiers, ongoing operational costs, and the hidden expenses nobody warns you about. Written by the team that shipped VakeelSaathi's production RAG on 50,000+ Indian court judgments.

Published 24 July 2026 12 min read By Decipher Consultancy Services
₹1.5-3 L
Prototype
₹6-15 L
Production build
₹40k/mo
Managed ops
6-9 wks
Total timeline

The real reason prices vary from ₹50k to ₹50 lakh

Type "AI chatbot development cost in India" into Google and you'll get answers ranging from ₹50,000 to ₹50 lakh. That's a 100x spread. Buyers, understandably, get confused. Are the ₹50k vendors lying about capability, or are the ₹50L vendors overcharging? Neither, mostly. They're quoting genuinely different products.

At the low end, you're buying an off-the-shelf ChatGPT wrapper. A no-code platform (Botpress, Chatbase, Landbot, or an in-house PHP script) that answers questions using OpenAI on a small FAQ file. It works, in the same way a bicycle works for going to the airport. If your use case is single-channel web chat on 50 FAQ items with no compliance requirements, this is fine.

At the high end, you're buying a production-grade Retrieval Augmented Generation system on your own private data, integrated with your CRM and helpdesk, deployed on WhatsApp and web, with an evaluation harness, cost caps, monitoring, DPDP-aligned data handling, and a team on retainer to keep it working. This is what an enterprise or regulated business actually needs, and it costs 30-50x more because the work is 30-50x more.

Most of the confusion in quotes you receive is vendors comparing apples to oranges. This guide fixes that. Below is the honest map: what drives cost, what tier fits your use case, and what the ongoing monthly cost actually looks like once you're live.

The 6 cost drivers, explained

Six variables move your quote up or down. Understand these before you talk to any vendor and you'll instantly spot which ones are being honest and which are lowballing to win the contract.

1

Chatbot type

An FAQ bot that answers scripted questions is a different piece of software from a RAG chatbot that reasons over 10,000 of your documents, which is a different beast from a multi-turn AI agent that can book meetings, update your CRM, and complete transactions. FAQ bots start at ₹75,000 because they're mostly configuration. Product-level Q&A chatbots begin around ₹1.5 lakh. Full RAG chatbots on private data are ₹6-15 lakh. Multi-turn agents with tool use, memory, and workflow orchestration cross ₹15 lakh easily. Vendors quoting ₹50k for "an AI agent that handles everything" are either building something much simpler than they're describing, or they're stacking hidden costs post-signature.

Impact: 5x-10x
2

Data volume and complexity

A chatbot on 100 clean policy documents is a two-week job. A chatbot on 10,000 mixed PDFs, Word files, spreadsheets, and scanned images with OCR requirements is a two-month job. A chatbot on 100,000 documents (like VakeelSaathi's 50,000+ court judgments) needs data pipelines, structured extraction, chunking strategy tuning, and infrastructure that survives real query load. Each 10x jump in data volume is roughly a 2-3x jump in engineering cost. Complexity matters more than raw count. 1,000 legal contracts with tables and clauses are harder to index well than 5,000 blog posts.

Impact: 2x-4x
3

LLM choice

OpenAI GPT-4o is the default. Anthropic Claude 3.5 wins on nuanced reasoning and long-context work. Google Gemini 1.5 Pro is competitive on price at high context lengths. Self-hosted Llama 3 or Mistral runs on your own infrastructure with no per-token cost but needs GPU. The cost implications are large. GPT-4o at $2.50/1M input tokens works out to roughly ₹0.20-0.35 per user message. Self-hosted Llama 3 8B on a single A100 costs ₹2.5 lakh/month regardless of volume. Below 5 million tokens per day, cloud APIs win. Above that, or when data-residency requires on-prem, open-source wins. Vendors who default to one model without asking about your volume, latency, and privacy needs are optimizing for their comfort, not your cost.

Impact: 3x-8x on monthly ops
4

Deployment complexity

A web widget you paste onto one page is roughly one engineering week. WhatsApp Business API adds session management, template approval, and BSP integration (Gupshup, Interakt, MSG91). Add another two weeks and roughly ₹1-2 lakh. Multichannel deployment (web plus WhatsApp plus Slack plus in-app) needs a unified conversation state, cross-channel handoff, and consistent tone. Add another 2-3 weeks. On-premises deployment inside a client VPC needs GPU procurement, VPN setup, air-gapped model deployment, and security review. Add ₹4-8 lakh and 3-4 weeks. Each channel roughly doubles the QA surface area.

Impact: 2x-3x on build
5

Integrations required

A standalone chatbot that only answers questions is the cheap end. Once the bot needs to read from Shopify, write to Salesforce, check inventory in your ERP, look up tickets in Zendesk, or trigger a payment in Razorpay, you're building integrations. Each integration is roughly 3-8 days of work depending on the API quality. Salesforce is easy because Salesforce has documented everything. A homegrown PHP CRM built by a previous vendor with no docs is 3-5x the work. Ask vendors to list every integration in scope with a line-item day estimate. Vendors who wave this away as "we'll figure it out during build" are guaranteeing scope creep.

Impact: ₹50k-3L per integration
6

Compliance and security

Public data with no PII is easy. Any customer name, phone, email, or address triggers India's DPDP Act obligations. Health data triggers HIPAA-equivalent handling. Financial data triggers RBI rules. Regulated verticals (banking, insurance, healthcare, legal) usually require on-prem or private-cloud deployment, encryption at rest and in transit, audit logs, data residency in India, and SOC 2 Type 2 posture from your vendor. Each layer adds 15-30% to build cost and roughly 20% to monthly ops. Vendors who ignore compliance in scoping are pricing an unrealistic project. When you go live, the security review will kill the timeline.

Impact: +25%-60% total

Detailed pricing tiers

Four honest tiers. Pick the row that matches your data volume, integration count, and compliance needs. If a vendor is quoting inside Tier 1 pricing but describing Tier 3 scope, you have a problem.

Tier What you get Ideal for Prototype Full build Monthly ops
Starter
FAQ Bot
Rule-based flows plus basic LLM fallback on ~100 FAQs. Web widget only. Basic analytics. Small business, single channel, low query volume ₹75k-1.5L ₹2-5 L ₹15k-25k/mo
Standard
RAG Chatbot
Full RAG on 500-5,000 documents. WhatsApp plus web. Basic evals. Handoff to human. Analytics dashboard. Mid-market SaaS, D2C brands, growing services businesses ₹1.5-3 L ₹6-15 L ₹40k-60k/mo
Enterprise
Production RAG
Full-scale RAG on 10k+ documents. Multichannel. Evaluation harness. Monitoring. DPDP compliance. SSO. Audit logs. Enterprise, regulated verticals, high query volume ₹4-6 L ₹15-30 L ₹80k-1.5L/mo
Custom
Multi-agent + On-prem
AI agents with tool use, workflow orchestration, on-prem or private-cloud LLM, custom integrations, dedicated infra. Banks, insurance, healthcare, government, defence ₹6-10 L ₹30-60 L ₹1.5-3 L/mo

Not sure which tier fits your use case?

15-minute call. We'll ask five questions about your data, channel, and compliance needs, then tell you honestly which tier you actually need. No upsell. If Tier 1 is right for you, we'll say so.

Book Free 15-min Call

What drives ongoing monthly costs

The build cost is the small number. The interesting number is what you spend month after month for the next three years. Most buyers focus on quote comparisons and skip this. That's how a "cheap" ₹4 lakh build becomes ₹1.5 lakh/month in surprise API bills six months in.

For a mid-market production chatbot handling roughly 50,000 conversations a month, here's the typical breakdown:

LLM API costs (OpenAI, Anthropic, Google)40-55%
Vector database (Pinecone, Weaviate, pgvector)5-15%
Compute and hosting (AWS, Azure, GCP)10-15%
Monitoring and observability (Langfuse, Datadog)3-7%
Human time (evals, incident response, retraining)15-25%
Channel fees (WhatsApp BSP, SMS gateway)5-10%

The LLM line is the one that surprises people. GPT-4o at scale, without prompt caching and without context pruning, easily runs ₹25,000-70,000/month for a moderately busy bot. A well-engineered RAG system with tight prompts, response caching for repeated queries, and cheaper models on easy questions can cut this 40-60%. That optimization work is what a managed-ops retainer buys you.

Hidden costs vendors don't mention in the quote

These items rarely appear on the first quote you receive. They're not always malicious, they're just what happens when a vendor is trying to win the bid. Ask upfront and you'll see who's being honest with you.

Data preparation and cleaning
Often 30-40% of prototype effort and rarely quoted. Extracting text from scanned PDFs, deduplication, chunking strategy, metadata tagging. If your data is messy (and it always is), this is where budgets die.
Evaluation harness and testing
A curated set of 500+ real user questions with expected answers, run nightly, so you catch quality regressions before your customers do. Skip this and your bot silently degrades over time. Budget ₹1-3 lakh in build plus ~10% of monthly ops.
Post-launch model drift monitoring
Model providers update or deprecate versions. Your prompts that worked on GPT-4-turbo may misbehave on GPT-4o. Someone has to catch this. Budget 4-6 hours of engineering per month for a mid-market bot.
LLM API cost overruns from unbounded usage
A single misconfigured prompt loop or a jailbreak attempt can burn ₹50,000 in OpenAI credits overnight. Rate limits, per-user caps, and cost alerts must be built in from day one. They rarely are.
Compliance audit trail requirements
DPDP requires you to log what data was accessed, by whom, when, and why. If you're regulated, add SEBI, RBI, or IRDAI on top. This is engineering work, not paperwork. Budget ₹1.5-3 lakh in build if you're in a regulated vertical.
WhatsApp Business API template approval delays
Every message template needs Meta's approval, which takes 2-24 hours per template and can bounce for cosmetic reasons. Plan for 2-3 rounds of template revision before launch. This can push a 6-week timeline to 8 weeks if you didn't budget for it.

Cost comparison: build vs buy vs off-the-shelf

The three real choices in front of you. Each makes sense for a different profile of business.

Option Setup cost Monthly cost Ownership Break-even point
Off-the-shelf
Intercom Fin, Zendesk AI, Freshchat AI
₹0-50k ₹5-15 L/mo at scale Vendor lock-in Under 500 resolutions/month
Build in-house
Your own team operates it
₹15-30 L ₹10-40k/mo infra Full ownership Requires ML plus DevOps team
Build with Decipher
We build it, you own it, we operate it
₹6-15 L (fixed) ₹40-80k/mo Full ownership, managed by us Above 500 monthly resolutions
When off-the-shelf wins

If you're at fewer than 500 support resolutions a month and your product doesn't have proprietary data the bot needs to know, Intercom Fin at $0.99 per resolution is cheaper than any custom build. We'll tell you this on a discovery call. Building custom below that threshold is bad math.

Regional cost comparison: India vs US vs Eastern Europe

Some readers are foreign buyers considering an offshore build. The Indian AI dev market is competitive on price without being competitive on quality, provided you pick the right partner. Here's the honest comparison. USD figures use a rough ₹83 conversion.

Region Prototype cost Production cost Monthly retainer
India
Boutique agency like Decipher
$2k-4k $8k-20k $500-2k/mo
Eastern Europe
Poland, Ukraine, Romania
$5k-10k $20k-40k $1.5k-4k/mo
US boutique
Small AI agencies
$15k-30k $60k-150k $5k-15k/mo
Enterprise firm
Accenture, TCS, Deloitte AI
$50k+ $200k+ $15k+/mo

The Eastern European ecosystem was cheaper than India in 2018. It isn't anymore. Prices there have crept toward US rates while India has held steady. For English-language, RAG-focused, cloud-deployed chatbot work, India is the current pricing sweet spot.

How to get better quotes

Six habits that will save you weeks of back-and-forth and a lot of money.

  • 1
    Know your data volume and integration count before you ask. Count your source documents, list every system the bot must read from or write to, and estimate expected daily query volume. This one-page brief will get you honest quotes instead of vague ranges.
  • 2
    Ask for a prototype phase separately. Don't skip straight to production. A ₹1.5-3 lakh prototype tests the retrieval quality on your actual data before you commit to a ₹10-20 lakh full build. Vendors who refuse to sell you a prototype are trying to lock you in.
  • 3
    Require explainability if you're in a regulated vertical. Ask the vendor how the bot cites sources and how you'd defend an answer to a regulator. If they don't have a clear answer, they haven't built for compliance before.
  • 4
    Ask about model API cost caps. What happens if the bot suddenly gets 10x its normal traffic? Is there a hard rate limit? A per-user cap? A daily budget alert? Uncontrolled API spend is the biggest post-launch surprise.
  • 5
    Get the evaluation harness in scope. Ask the vendor to include 500+ curated Q&A pairs, nightly regression testing, and a quality dashboard. Vendors who scope this out of the initial build are setting up a future upsell.
  • 6
    Ask about on-prem deployment as an option. Even if you don't need it now, knowing that the vendor can deploy inside your VPC or on-prem later without a rewrite tells you they've built the architecture properly. If they can't do it, they've probably hardcoded OpenAI dependencies everywhere.
Final honest note

An AI chatbot is not a one-off software project. It's a system that needs to be operated, tuned, and re-evaluated as your data changes and as model providers push updates. Whichever vendor you pick, budget for the operations. The ones who quote a build price without discussing ongoing ops are the ones you'll be replacing 12 months in.

FAQ

Common pricing questions

Questions we hear on nearly every discovery call about chatbot cost.

A basic FAQ chatbot using a low-code platform (Botpress, Chatbase, Landbot) with ChatGPT wired in can start at ₹50,000 to ₹1.5 lakh for a small business. This is a scripted flow with LLM fallback on 50-100 FAQ items. It works for single-channel web chat, low query volume, and non-critical use cases. If you need RAG on your own documents, multichannel deployment, or regulated data, budget from ₹1.5 lakh for prototype and ₹6 lakh upward for production.

Because vendors are quoting different products under the same name. A no-code FAQ bot with a ChatGPT wrapper is fundamentally different work from a production RAG system on 10,000 documents with WhatsApp integration, evaluation harness, monitoring, and DPDP compliance. Data volume, integration count, compliance requirements, LLM choice, deployment channel, and ongoing operations each move the number. Ask vendors for line-item breakdowns and you'll see where the numbers diverge.

It depends on the vendor. Some quote build cost only and pass API costs through at runtime. Others bundle a monthly cap into a retainer. At Decipher, our managed operations retainer (from ₹40,000/month) includes model API costs up to defined query volumes. Beyond the cap, we pass costs through with an operational margin. Ask for cost caps and overage rates upfront. Uncontrolled API spend is the number one hidden cost in production chatbots.

WhatsApp Business API has three cost layers. First, BSP fees (Gupshup, Interakt, MSG91, or a direct provider), typically ₹0.30-0.60 per conversation. Second, Meta's conversation charges (marketing, utility, authentication, service categories) which range from ₹0.35 to ₹0.80 per conversation in India. Third, one-time setup and template approval, roughly ₹15,000-40,000. For a bot handling 10,000 conversations a month, budget ₹8,000-15,000 in WhatsApp costs on top of your LLM and infrastructure spend.

Yes, if you build custom. When you use OpenAI, Anthropic, or Google APIs, you don't own the model but you own the prompt engineering, retrieval logic, integration code, and evaluation harness. If you deploy an open-source LLM (Llama 3, Mistral, Qwen) on your own infrastructure, you own everything end-to-end. At Decipher, all code and model artifacts we build are transferred to you at project close. No source-code escrow needed.

A rule-based chatbot uses if-then decision trees. Setup is cheap (₹50k-2L) because there's no AI involved. But you rewrite it every time your product changes. An LLM chatbot understands intent, handles paraphrasing, and can be updated by changing source documents instead of rewriting flows. It costs 3-5x more upfront but scales without linear engineering effort. Most 2026 deployments blend both. Rules for deterministic paths (payments, authentication), LLM for open-ended queries.

Six line items. (1) LLM API costs, 40-55% of ongoing spend for most bots. (2) Vector database, 5-15% depending on whether you're on managed Pinecone or self-hosted pgvector. (3) Compute and hosting, 10-15%. (4) Monitoring and observability tools (Langfuse, Datadog), 5%. (5) Human time for evals, incident response, content re-indexing, 15-25%. (6) WhatsApp Business API and other channel fees where applicable, variable. For a mid-market production chatbot, expect ₹40,000-60,000/month all-in.

Not always. Open-source LLMs eliminate per-token API costs but require GPU infrastructure that runs 24/7. A single A100 GPU on AWS Mumbai costs ₹1.8-2.5 lakh/month. That break-even threshold is roughly 5 million tokens per day. Below that, OpenAI or Anthropic APIs are cheaper. Above that, and especially when data-residency or fine-tuning requirements enter the picture, self-hosted open-source wins. We help you model this before recommending.
Related Reading

Keep going

Get a real quote in one 15-minute call

Tell us your data volume, top 3 use cases, and compliance context. We come back within 48 hours with a fixed prototype scope, a tier recommendation, and a monthly ops estimate you can plan against.

Decipher Assistant
Typically replies instantly