SMALL BUSINESSES NEED AN AI TIME-SAVINGS LEDGER THAT LEADS TO RESULTS
5 min read
An AI chatbot costs anywhere from about $25 to $100 a month for an off-the-shelf tool you set up yourself, to roughly $8,000 to $80,000 for a custom chatbot that reads your own documents and works inside your systems, plus a monthly running cost. The right answer depends less on the technology than on how many conversations you handle and what the bot has to do once the conversation ends.
Here is what each option costs, what WhatsApp adds to the bill, and the simple calculation that tells you which one makes sense for your volume.
Customer service platforms now include an AI agent that answers from your help pages and FAQ. You pay a subscription, and usually an extra fee per AI conversation or per resolved question. Setup takes days, not months, and there is nothing to maintain. The limits: the bot only knows what you feed it, it connects to a fixed list of tools, and the per-conversation fees grow with your volume.
No-code builders and automation platforms let you assemble a bot yourself or with a freelancer: you choose the AI model, upload your documents and connect a few tools. It is cheaper per conversation than most SaaS plans, but someone has to build, test and look after it.
A developer or agency builds a bot around your data and processes: it can check stock or availability, create a booking, qualify a lead and log it in your CRM, or hand over to a person with the full context. Most of these bots use retrieval-augmented generation (RAG): before answering, the bot looks up the relevant passages in your own documents, which keeps answers grounded in your information rather than the model's general knowledge.
The figures below come from vendors' public pricing pages and from agencies' published guides. They are market reference points, not quotes.
| Option | Typical published price | What it covers |
|---|---|---|
| SaaS, small business plan | Tidio: Starter from about $24/month billed annually; its Lyro AI agent from $32.50/month for 50 AI conversations | Website chat widget, AI answers from your content, a shared inbox |
| SaaS, per-resolution pricing | Intercom: $0.99 per outcome for its Fin AI agent, plus seats from $29 per agent per month | AI agent across chat, email and messaging, with human handover |
| Custom support bot (RAG) | about $8,000 to $25,000 at the low end of agency guides | A bot that answers from your documents on one or two channels |
| Custom multichannel bot with CRM integration | about $30,000 to $60,000 in several agency guides | Website and WhatsApp, actions in your systems, reporting |
Published ranges for custom chatbots vary enormously, from a few thousand dollars for a rule-based bot to six figures for enterprise systems with compliance requirements. The spread comes from scope, not from the AI: the integrations, the quality of your documentation and the testing are what take time.
Custom bots also have a running cost: hosting, AI usage and monitoring. Agency guides put it anywhere from a few hundred dollars a month for a small bot to several thousand for high-volume systems.
The AI itself is cheap. AI providers charge per token, roughly a fragment of a word. In mid-2026, OpenAI's GPT-5.4 mini, a small model suited to customer service, was listed at $0.75 per million input tokens and $4.50 per million output tokens. A support conversation where the bot reads about 10,000 tokens (the question, the history and the passages pulled from your documents) and writes about 500 tokens of answers costs around one cent in AI usage.
Compare that with per-resolution pricing at $0.99. At 1,000 resolved conversations a month, that is about $990 a month in AI fees on a SaaS plan, against roughly $10 of raw AI usage for a custom bot. That difference is what you pay for the platform, its reliability and the fact that you have nothing to build or maintain.
That gives a simple rule of thumb:
Where your customers write to you on WhatsApp, as is common in Morocco and across much of Africa, Europe and Latin America, a WhatsApp chatbot often matters more than a website widget. Three things to know.
The free WhatsApp Business app only offers basic automated replies such as greeting and away messages. An AI chatbot needs the WhatsApp Business Platform (the Cloud API), accessed directly through Meta or through a business solution provider, which may add its own monthly fee.
Since 1 July 2025, Meta charges per delivered template message. When a customer writes to you, a 24-hour customer service window opens, and inside it your replies are free, as are utility templates such as order or booking confirmations. When a customer reaches you through a click-to-WhatsApp ad or a Facebook Page button, all messages are free for 72 hours. You pay mainly for messages you start yourself outside that window, especially marketing templates, at rates that vary by country.
So a WhatsApp bot that answers incoming questions can cost very little in Meta fees. A bot that sends promotional broadcasts will not.
Since 15 January 2026, Meta's WhatsApp Business terms bar general-purpose AI assistants offered by AI providers from the platform. Bots that serve a business's own customers, for support, bookings, order tracking or sales, remain allowed. In practice: build a bot that answers about your products and services, not a general assistant that happens to carry your logo.
They work well when three conditions are met, and disappoint when one is missing:
Before any build, list the twenty questions your team answers most often and how long they take each week. That list tells you whether a chatbot is worth it, and it becomes the bot's first training material.
A chatbot is one way to save your team's time or to answer prospects faster; it is not always the first one to pull. Our free Growth Scan shows in three minutes which lever matters most for your business. For a precise plan, the Growth Diagnosis maps your conversations, volumes and tools in two weeks and tells you whether an off-the-shelf tool is enough. See how we approach automation and AI assistants, automated WhatsApp follow-up, or a working example with our sample AI agent. For the wider picture of costs, read what AI automation really costs a business.
Already know what you need? Tell us about it and we will tell you which option fits your volume.
Published small business plans start at around $25 to $100 a month, with AI conversations often billed on top. A custom chatbot has running costs instead: hosting, AI usage and monitoring, typically from a few hundred dollars a month for a small bot.
The build costs the same as any chatbot. On top, Meta charges per template message you send; replies inside the 24-hour window after a customer writes to you are free. A provider's platform fee may apply if you do not connect to Meta's Cloud API directly.
The one that connects to the tools you already use and answers from your own content. For standard questions at modest volume, an off-the-shelf tool is usually the best choice. A custom bot makes sense when volume is high or the bot has to take actions in your systems.
Several tools have free plans with a small number of conversations, enough to test the idea on your website. A chatbot your customers rely on needs paid usage, testing and someone to keep its information up to date.
A chatbot that searches your own documents before answering (retrieval-augmented generation). It keeps answers grounded in your prices, policies and products, and is the usual approach for custom business chatbots.
They can be, if you choose where data is stored, limit what the bot can access, and tell customers in your privacy policy. Avoid letting a bot read data it does not need to answer the question.
Comments
No comments yet. Be the first to share your thoughts.
Leave a comment
Your comment will be published after moderation. Your email is never displayed.