Written and maintained by the Icon Web Solution team in Ahmedabad. Last reviewed November 2026.
“AI chatbot” now covers everything from a scripted FAQ widget to an assistant that reads your documents, checks your CRM and books appointments on WhatsApp. The price range is correspondingly wide. This guide explains the decisions that set the cost, the running costs people forget, and where our own chatbot projects start, which is USD 2,000.
Rule-based or AI? The first fork
A rule-based bot follows a script: buttons, menus and fixed answers. It is cheap, predictable and fine for simple routing (“Sales or Support?”). An AI bot uses a language model to understand free-text questions and answer from your content. It handles far more situations, costs more to build well, and has a running cost per conversation. Many good deployments combine both: AI for open questions, rules for sensitive steps like payments or cancellations.
What decides the build cost
- Channels. Website only is simplest. WhatsApp Business API adds approval, templates and a provider; Instagram, Facebook Messenger and in-app chat each add integration and testing.
- Knowledge sources. A bot that answers from twenty FAQ entries is quick. One that must read PDFs, policy documents, product catalogues or a help centre needs a retrieval setup, document cleaning and testing for wrong answers.
- Integrations. Looking up an order status, checking stock, creating a lead in your CRM, booking a slot or sending an invoice each means connecting to another system securely.
- Languages. English, Hindi and Gujarati, and the mixed “Hinglish” people actually type, need testing in each.
- Human handover. Passing a conversation to a person on WhatsApp or a helpdesk, with context, is a feature in itself and one of the most valuable.
- Guardrails and tone. Keeping the bot on your topics, refusing what it should refuse and sounding like your brand takes prompt design and test conversations.
- Analytics. Knowing what people asked, what the bot could not answer and which conversations turned into leads is how the bot improves.
Running costs: the part most quotes leave out
- Model usage. AI providers charge per token (roughly per word) processed. A busy bot costs more per month than a quiet one. Ask for an estimate at your expected conversation volume.
- WhatsApp fees. Meta charges per conversation, and your API provider adds its own fee.
- Hosting for the bot’s back end and vector store.
- Maintenance. Content changes, new products, model updates and reviewing wrong answers. A bot that nobody maintains gets worse as your business changes.
A fair quote separates the one-time build from the monthly running cost and tells you what drives each.
What a well-built chatbot includes
- A written list of the questions and tasks the bot will handle in version one, and the ones it will hand to a human.
- Your content prepared and loaded, with a process for updating it.
- Conversation design: greeting, fallbacks, handover, and how it asks for a name and phone number without being annoying.
- Testing with real questions from your sales and support inbox, in the languages your customers use.
- Lead capture into your CRM or at least your email and a sheet.
- A dashboard or monthly report of conversations, unanswered questions and leads.
- Privacy basics: what is stored, for how long, and a way to delete it.
Signs a quote is too thin
- No question about your content or your systems before a price is given.
- No mention of running costs.
- No human handover.
- No testing phase with your real questions.
- “It learns by itself.” It does not; someone has to review and update it.
Start small, then expand
The projects that succeed usually start with one channel and one job: answer the twenty questions your team answers every day, capture the lead, hand over when stuck. Once that works and the numbers show it, add WhatsApp, integrations and more content. A pilot keeps the first invoice small and gives you real data to decide the next step.
Where our pricing starts
Icon Web Solution AI chatbot projects start from USD 2,000. That covers a website or WhatsApp assistant that answers from your prepared content, captures leads, hands over to a person, and comes with a month of tuning from real conversations. Additional channels, document-heavy knowledge bases, CRM or order-system integrations and multi-language support are priced as separate lines in the written scope, and we give you a running-cost estimate at your expected volume before you commit.
How to get an accurate quote
- Which channel first: website, WhatsApp or both.
- The ten to twenty questions your team answers most often.
- Where the answers live today: website pages, PDFs, a help centre, people’s heads.
- Which systems the bot must talk to, if any.
- Languages your customers use.
- Rough monthly conversation volume and your deadline.
Send that through the contact page and we will reply within one working day with a scoped pilot. See AI chatbot development for examples of what we build, or AI and machine learning development for custom models beyond chat.