A custom AI chatbot typically costs US$3k–10k for a FAQ or rules-based bot, US$10k–40k for a bot that answers from your own documents, and US$40k or more for an AI agent that takes actions in your systems. In India the same tiers run roughly ₹50k–2 lakh, ₹2–8 lakh and ₹8–25 lakh. Running costs add 15–25% of the build per year.
Those are typical market ranges GRWW sees, not fixed prices. Below is what puts a project at the bottom or top of each range, the costs people forget, and how to avoid paying for capability you won't use.
What are the main types of custom chatbot, and what does each cost?
Chatbot cost is driven mostly by what the bot is allowed to know and do, not by how clever the conversation feels. A bot that only answers from a fixed FAQ is cheap; one that reads your documents and updates your CRM is a software project.
| Chatbot type | What it does | India (₹) | Australia (AUD) | Global (USD) |
|---|---|---|---|---|
| FAQ / rules chatbot | Answers set questions, collects lead details, hands over to a human | ₹50k–2 lakh | $5k–15k | $3k–10k |
| AI chatbot on your own documents (RAG) | Answers from your policies, catalogues, manuals, with sources | ₹2–8 lakh | $15k–60k | $10k–40k |
| AI agent with integrations | Books appointments, checks orders, updates CRM, sends follow-ups | ₹8–25 lakh | $60k+ | $40k+ |
"RAG" stands for retrieval-augmented generation. In plain terms, the bot searches your documents for relevant passages and uses them to write its answer, rather than relying on what the model learned in general training. It's what lets a bot answer "what's your cancellation policy?" correctly for your business.
What makes a chatbot project cost more?
Integrations, messy source content and the level of accuracy you need are the three biggest cost multipliers. The model itself is a small part of the bill.
| Cost driver | Lower end | Higher end |
|---|---|---|
| Channels | Website widget only | Website, WhatsApp, Instagram and email |
| Knowledge source | 20–30 clean FAQ answers | Hundreds of PDFs, scanned forms, spreadsheets |
| Actions | Collects a name and number | Reads and writes to CRM, calendar, order system |
| Languages | One | Several, including mixed-language chat (e.g. Hinglish) |
| User access | Public, anonymous visitors | Logged-in customers with different permissions |
| Accuracy needs | Wrong answer is mildly annoying | Wrong answer has legal or financial impact |
| Admin tools | None; developer updates content | Staff dashboard to edit answers, review chats, see analytics |
| Testing | Quick checks | Structured test set of real questions, rerun after every change |
If your quote seems high, ask which of these rows is pushing it up. Often one requirement, such as handling scanned documents or supporting four channels, accounts for a large share and can be phased in later.
What ongoing costs should you budget for?
Expect four recurring costs: model usage, hosting, maintenance and content upkeep. Together they're often smaller than the build, but they never reach zero.
- Model usage. Providers such as OpenAI, Anthropic and Google bill by the amount of text processed. Short FAQ answers cost very little per conversation; bots that pull long documents into every answer cost more. Set a monthly cap and alerts.
- Hosting and storage. Servers, a database for conversations, and a vector database (a search index for document chunks) if you use RAG.
- Maintenance. Roughly 15–25% of the build cost per year, covering model version changes, bug fixes, security updates and integrations that change on the other side.
- Content upkeep. Someone on your team must update prices, policies and FAQs. An out-of-date bot gives confident wrong answers, which is worse than no bot.
- Channel fees. If the bot runs on WhatsApp, Meta charges for certain message types under the WhatsApp Business Platform rate card, which varies by country and message category.
Is a SaaS chatbot subscription cheaper than a custom build?
For simple use cases, yes. If you need a website widget that answers common questions and captures leads, a monthly SaaS chatbot tool is usually cheaper and faster than a custom build.
Custom starts to make sense when:
- You need the bot to work with your own systems (a custom CRM, booking engine or internal database) that off-the-shelf tools don't connect to.
- Customer data is sensitive and you need control over where it is stored and processed.
- Per-seat or per-conversation subscription fees would exceed the cost of owning it within two or three years.
- The conversation flow is specific to your business and generic tools force workarounds.
A sensible path for many businesses is to start with a subscription tool, learn what customers actually ask, then build custom once the requirements are clear. That way the custom build is based on real questions, not guesses. Our AI automation guide for small and mid-size businesses covers this build-versus-buy decision more broadly.
What does a typical quote break down into?
Discovery, content preparation, build, integrations and testing are the usual line items; testing is the one most often under-budgeted. Here is an illustrative split for a mid-range document-based bot (not a real client quote):
| Line item | Share of budget (illustrative) | What it covers |
|---|---|---|
| Discovery and conversation design | 10–15% | Use cases, tone, handover rules, success metrics |
| Content preparation | 10–20% | Collecting, cleaning and structuring your documents |
| Core build | 30–40% | Chat interface, retrieval, prompts, conversation storage |
| Integrations | 15–25% | CRM, WhatsApp, calendar, email |
| Testing and launch | 10–20% | Test set, fixes, supervised launch period |
If a quote has no testing line at all, ask how they will know the bot answers correctly before your customers find out.
How long does it take to build?
As a rough estimate, a FAQ bot takes a few weeks, a document-based bot one to three months, and an agent with several integrations longer. Timelines depend heavily on how quickly your content and system access are ready.
The most common delay isn't development. It's waiting for someone in the business to write down the real answers to the questions customers ask.
How can you reduce the cost of a custom chatbot?
Narrow the first version to one channel, one job and one source of truth. You can add more once it's working.
- Pick the single highest-value job. Usually that's answering pre-sale questions and capturing qualified leads.
- Start with one channel. Choose where your customers already message you. For many Indian businesses that's WhatsApp; for others it's the website. Our comparison of AI chatbots, live chat and WhatsApp bots helps you decide.
- Prepare your content yourself. A clean, current FAQ document saves real money in the content-preparation phase.
- Hand over to a human early. A bot that confidently passes tricky questions to your team is cheaper and safer than one engineered to handle everything.
- Avoid agents until you need actions. If the bot only needs to answer and collect details, you don't need the extra cost of an agent. Read what AI agents are and when businesses need them before you commit.
- Plan maintenance from day one. A bot nobody updates decays quickly; budgeting for upkeep protects the original investment, just as with website maintenance costs.
When is a custom chatbot not worth it?
If you get only a handful of enquiries a week, a chatbot rarely pays back. A well-written FAQ page and a fast human reply on WhatsApp will do more.
It's also not worth it if your answers depend on case-by-case judgement (complex legal, medical or immigration advice), or if nobody can own the content. Fix those first.
Frequently asked questions
What is the cheapest way to get an AI chatbot?
A SaaS chatbot tool that you load with your FAQs is the cheapest starting point, often a monthly subscription. A custom FAQ or rules bot typically starts around US$3k (₹50k in India). Choose custom only when you need integrations or control that subscription tools can't offer.
Why do AI chatbot quotes vary so much?
Because "chatbot" covers everything from a scripted FAQ widget to an agent connected to five business systems. Compare quotes on the same scope: channels, knowledge sources, integrations, languages, admin tools and testing.
Do I pay the AI provider separately?
Usually yes. Model usage is billed by the provider based on volume, either to your own account or passed through by your developer. Ask for an estimate based on your expected number of conversations and set a spending cap.
Can I train the chatbot on my own documents?
Yes. Most businesses do this with retrieval (RAG) rather than retraining a model. The bot searches your documents for each question and answers from what it finds, which is cheaper to maintain and easier to update.
Who owns the chatbot after it's built?
With a custom build, you should own the code, prompts, configuration and conversation data, subject to your contract. Confirm this in writing, along with where the data is hosted, before work starts.
Get a clear number for your chatbot
Tell us which channel, what the bot should know and what it should do, and we'll give you a scoped, fixed-price estimate with running costs spelled out. Examples of what we've delivered are on our work page, or request a chatbot quote here.