Search for what an AI chatbot costs and you get ranges that are true and useless at the same time. One agency guide says a custom RAG chatbot runs $35,000 to $150,000. Another puts an LLM chatbot at $40,000 to $90,000 for a starter build. A WhatsApp-specific guide says $5,000 to $50,000 and up. None of them is wrong. They are ranges with no scope attached, and a range that wide tells you nothing about your project.
When I built PostEngage, a WhatsApp automation product, on my own, the part that decides what to say was the easy part. Most of the work went into everything around it: webhooks that retry, templates that need approval, an inbox for humans, billing, and alerts that explain their own failures. Chatbot quotes swing by a factor of ten because some of them include that work and some quietly do not.
So this is a chatbot taken apart into line items, with hour ranges for each, published rates for who might build it, and the monthly bill once customers start using it.
Where the hours go
Here is a typical first serious chatbot: on the website and WhatsApp, answering from the company's own documents, connected to a CRM and an order system, able to hand a conversation to a person with the history attached, and with an admin panel so someone can fix a wrong answer without a developer.
Line by line
These are my own scoping estimates for a production build, meaning something real customers will use, not a demo. The low end assumes a clear brief and few surprises. The high end assumes the usual number of surprises.
| Piece | Hours | What is inside it |
|---|---|---|
| Bot core, model wiring, deploy | 60–100 | prompts, conversation state, guardrails, hosting, CI |
| Scripted flows and answers | 20–40 | menus, buttons, fixed replies for common questions |
| Answers from your documents (RAG) | 60–120 | ingestion, chunking, embeddings, search, citations, re-indexing |
| Documents plus live data | 100–180 | the above, plus reading live systems safely at answer time |
| Website chat widget | 20–40 | the widget, sessions, mobile layout |
| 40–80 | Cloud API setup, webhooks, templates and approvals, opt-in | |
| Instagram and Messenger | 30–60 | the same, for Meta's other inboxes |
| Phone calls | 120–240 | telephony, speech in and out, latency tuning |
| Each integration | 20–40 | auth, the API, error handling, tests |
| Human handoff with context | 40–80 | routing to an agent, history attached, handing back |
| Admin panel and analytics | 30–60 | edit answers, review conversations, basic reporting |
| Each extra language | 10–20 | prompts, content, testing |
| Testing and launch | +20% | evals, load test, monitoring, a staged rollout |
Two lines deserve a warning. Integrations are where estimates die, because "connect to our system" can mean a clean REST API or a 2011 SOAP service with no documentation. And testing is the line clients most often ask to cut, then pay for later when the bot confidently tells a customer the wrong refund policy.
Who builds it, and at what rate
The same hours cost very different amounts depending on who does them. These are published rate bands, not my own prices:
| Who | Hourly rate | Source |
|---|---|---|
| Development agency in India | $25–49 | Clutch, September 2026 |
| Senior back-end freelancer, global median | $61–80 | Arc.dev |
| Development agency in the US | $50–99 | Clutch |
| Development agency in Canada or Australia | $100–149 | Clutch |
| Fractional CTO in Europe | €120–280 | fractional-csuite.com |
Rate is not the only difference. An agency usually brings a project manager, a designer and a QA person, which is worth paying for on a large build and pure overhead on a small one. A freelancer is cheaper per hour but is one person, so ask what happens when they are ill or busy.
Scope your own
Tick what you need and pick who builds it. The defaults are the WhatsApp support bot above.
estimator · build hours, build cost and the monthly bill
Scope your chatbot, see what it costs
after launch
where the build hours go (upper estimate)
Hour ranges are my own scoping estimates for a production build, not a quote; testing and launch add 20% on top. Rates are Clutch's September 2026 agency bands and Arc.dev's median for back-end freelancers. Running costs cover the model, WhatsApp's per-message fees from 1 October 2026, voice minutes at about $0.045 each, and about $100 a month for servers and a managed vector database.
Three scopes, priced
| Scope | Hours | Weeks, one developer | Agency in India | Senior freelancer | Agency in the US |
|---|---|---|---|---|---|
| Lean website bot, scripted answers | 120–216 | 4–8 | $3,000–$10,600 | $7,300–$17,300 | $6,000–$21,400 |
| WhatsApp support bot, as above | 348–672 | 12–23 | $8,700–$32,900 | $21,200–$53,800 | $17,400–$66,500 |
| Every channel plus phone calls, live data, 4 integrations, 2 languages | 636–1,224 | 22–41 | $15,900–$60,000 | $38,800–$97,900 | $31,800–$121,200 |
Put those next to the ranges agencies publish and the reason for the spread becomes visible. Each published range covers several of these scopes at once.
quick check
An agency quotes $5,000 for "a WhatsApp chatbot with AI". At $25 to $49 an hour, what is that most likely to leave out?
$5,000 buys roughly 100 to 200 hours at those rates. That covers a bot core and WhatsApp, not the 110 to 220 hours that two integrations, a handoff to humans and an admin panel usually take. Ask for the line items before comparing quotes.
The monthly bill after launch
Build cost is paid once. Running cost is paid every month, and it is smaller than most people expect. For the WhatsApp support bot above, at 5,000 conversations a month with six bot replies each:
| Line | India | United Kingdom |
|---|---|---|
| Model (Claude Haiku 4.5) | $103.50 | $103.50 |
| WhatsApp service messages, from 1 October 2026 | $42.00 | $660.00 |
| Servers and a managed vector database | $100.00 | $100.00 |
| A month | $245.50 | $863.50 |
| Per conversation | $0.049 | $0.173 |
The model is rarely the line that hurts. In India the whole bot runs at about five cents a conversation. In the UK, WhatsApp itself is more than six times the model cost, because Meta starts charging for every bot reply on 1 October and the UK rate is sixteen times India's. A bigger model roughly doubles the model line; the arithmetic behind that is here. Phone calls add about $0.045 a minute, taken apart here.
Where chatbot projects go over budget
- The integration nobody looked at. Budget a spike to read the target system's API before you fix a price for connecting to it.
- Nobody owns the knowledge. A bot that answers from your documents is only as good as the documents. If no one on your side owns keeping them current, the bot will be confidently out of date within a quarter.
- Handoff designed last. How a conversation moves to a person, and back, shapes the whole data model. Bolting it on after launch is where the rework comes from.
- WhatsApp paperwork. Business verification, template approvals and opt-in rules take calendar time even when they take few hours. Start them in week one.
- No evals. Without a set of real questions and expected answers to test against, every prompt change is a guess, and every model upgrade is a risk.
How to buy one without overpaying
- Ask for line items. A quote without them cannot be compared with another quote. The table above is a reasonable checklist of what to ask about.
- Buy in phases, each with a fixed price. Phase one is one channel and one knowledge source. Phase two is whatever the first month of real conversations says it should be. This is how I price my own work, because it keeps both sides honest.
- Ask for the monthly number. Anyone who builds these should be able to tell you roughly what it will cost to run at your volume. If they cannot, they have not thought about it.
- Own your accounts and your code. The WhatsApp Business account, the model provider keys, the repository and the knowledge base should be in your name from day one.
- Consider not building. If your needs are menus and FAQs, a hosted chatbot builder may be enough. Indian providers such as AiSensy list chatbot builders at around ₹2,500 a month, and WATI's plans start at $29. Build when you need your own data, your own systems, or your own margins.
questions people ask
How much does it cost to build an AI chatbot?
Roughly $3,000 to $20,000 for a lean website bot with scripted answers, $9,000 to $65,000 for a WhatsApp support bot that answers from your documents with integrations and human handoff, and more for every channel plus phone calls. The spread comes mostly from who builds it and how many systems it has to connect to.
How long does it take to build a chatbot?
About 4 to 8 weeks for one developer on a lean bot, and 3 to 6 months for a support bot with documents, integrations and handoff. Several people working in parallel can shorten that, but not linearly.
How much does a WhatsApp chatbot cost per month to run?
For 5,000 conversations a month, about $250 in India and about $860 in the UK, covering the model, WhatsApp's per-message fees from October 2026 and hosting. WhatsApp's fees depend heavily on the country your customers are in.
Should I hire a freelancer or an agency to build a chatbot?
A freelancer is usually cheaper and faster for a focused first phase. An agency makes more sense for a large, multi-channel build that needs design, project management and QA in parallel. Either way, ask for line items and a fixed price per phase.
Is it cheaper to use a no-code chatbot platform?
For menus, FAQs and simple flows, yes. Once the bot needs to answer from your own documents, act in your own systems, or you want control over running costs, a custom build usually pays for itself.
The short version
A chatbot is a small bot inside a lot of plumbing, and the plumbing sets the price. List the channels, the knowledge sources, the systems it must touch and whether humans take over, and the estimate stops being a range and becomes a number.
Then buy it in phases. Launch one channel, read a month of real conversations, and let them decide what phase two is. The most expensive chatbot is the one built to a guess.