Summarise this article with:
On this page8 sections
Use a rule-based chatbot on WhatsApp when the journey has fixed steps and one right answer: booking, payments, order status, document collection. Use an AI agent when customers ask open questions in their own words about a large catalogue or policy set. Most businesses need both on one number, with a person behind them.
The choice is not old technology versus new. Each approach fails in a different way, costs a different amount per reply, and sits under different policy limits. This guide compares them by job, so you can decide which conversations go where.
The difference in one table
A rule-based chatbot follows paths you draw in advance: buttons, list menus, keyword triggers and API calls in a fixed order. An AI agent uses a language model to read the message, decide what it means, answer from your content and call your systems to finish a task.
| Rule-based chatbot | AI agent | |
|---|---|---|
| How it decides | Paths and conditions you define | A model interprets each message |
| Same input, same output | Yes | Not guaranteed |
| What you build | A flow: steps, buttons, branches | Knowledge, instructions, tool connections |
| Typical failure | Customer types something off-menu and hits a dead end | Confident wrong answer, or an answer outside your policy |
| Languages and free text | Only what you anticipated | Handles varied wording and mixed languages |
| Changing it | Edit and republish the flow | Edit knowledge or instructions, then re-test broadly |
| Auditing a reply | Trace the path taken | Inspect the model and tool steps behind it |
Neither column is better overall. The question is which failure mode a given job can tolerate.
Jobs where a rule-based bot wins
Pick rules when the journey has a correct sequence and a wrong answer costs money, trust or compliance.
- Bookings and reschedules. A Kochi dental chain needs date, time and branch, in that order, validated against real slots.
- Payments and refunds. Amounts and status should come from your system, never be phrased by a model.
- Order status and tracking. One API call, one templated answer.
- Lead qualification. A fixed set of questions with validated answers feeds your CRM cleanly.
- Document collection and KYC steps. A fixed checklist with format checks.
- Opt-outs and consent. These must work the same way every time.
WhatsApp’s interactive messages are built for this. Meta’s reply buttons allow up to three predefined replies, and list messages up to 10 rows across all sections. For longer structured input, Meta’s own WhatsApp Flows (in-chat forms, not a bot builder) collect several fields on one screen; InfiQ’s WhatsApp Flows guide covers them.
Regulated sectors lean this way for a further reason. Meta’s Business Agent overview excludes the Finance, Government, Health, Alcohol, Gambling, over-the-counter drug and matrimony verticals from Meta Business Agent. An NBFC or a hospital can still use AI in a custom build, but scripted paths with a human behind them are the safer default for anything transactional or advisory.
Jobs where an AI agent wins
Pick an AI agent when the questions are open-ended, worded in many ways, and answerable from content you control.
- Product discovery across a large catalogue. “Something for my mother’s 60th, under ₹3,000, not too shiny” has no button.
- The long tail of FAQs. Your top questions fit a menu; the next few hundred do not.
- Mixed-language free text. Customers write in Hinglish, Tamil or English with typos, often in the same message.
- After-hours first response. Understanding the question and gathering details before a person picks it up next morning.
- Pre-sales questions. Comparing two variants, checking a policy, explaining delivery options.
An agent’s answers are only as good as the knowledge it retrieves from, and preparing that content is most of the work. Design for the failure mode: when the agent cannot find an answer, it should say so and hand over, not improvise.
Hybrids: how the two share one number
Most working setups combine both. Three patterns cover most cases.
Buttons first, AI for everything else. The greeting offers three or four buttons for the frequent transactions, plus “something else”. Free text and “something else” go to the agent. This keeps payments and bookings predictable while open questions still get an answer.
AI front door, scripted back office. The agent greets, understands the request, and hands transactional intents to a scripted flow: “reschedule” starts the booking flow, “refund” starts the refund flow. The agent never handles money directly.
Either way, a person behind both. The WhatsApp Business Messaging Policy allows automation inside the 24-hour window only with prompt, clear and direct escalation paths, such as an in-chat transfer to a human agent. That applies to rule-based bots and AI agents alike.
Routing matters on a shared number. On Meta Business Agent, control of a conversation passes between the agent and your app, and Meta’s testing guide warns that any message your app sends takes control and stops the agent replying until control is released. Meta’s overview also says an active authorised-agent integration on a number blocks Meta Business Agent, so plan one AI agent per number.
Cost per reply since 1 October 2026
Meta has charged for service messages, the free-form replies inside the customer service window, since 1 October 2026, according to its pricing update page. They were free until 30 September 2026. On InfiQ’s India rate card a service message costs ₹0.145 per delivered message, ex-GST. Every bot reply counts, whichever kind of bot sends it.
The three set-ups bill differently:
| Set-up | What Meta charges per reply | What else you pay |
|---|---|---|
| Rule-based bot on the Cloud API | One service message | Your bot platform’s fee |
| Custom AI agent on the Cloud API | One service message | Your model provider’s token bill |
| Meta Business Agent | Tokens at USD 2.00 per million (since 1 August 2026), instead of a service message | Your platform’s fee |
Meta’s page says a non-template message is charged either as a Meta Business Agent message or as a service message, never both, and estimates Business Agent messages at approximately 4–5 US cents each. Agent messages are charged for tokens even inside free entry point windows.
A hypothetical Surat saree retailer handling 4,000 chats a month at five bot replies each sends 20,000 replies. As a rule-based bot or custom agent on the Cloud API, that is 20,000 × ₹0.145 from Meta since 1 October 2026, plus model costs for the custom agent. On Meta Business Agent at Meta’s 4–5 cent estimate, it is roughly USD 800 to 1,000 in token charges instead.
Risk and policy limits
Accuracy risk. A rule-based bot’s worst case is a dead end; an AI agent’s is a fluent wrong answer about price, eligibility or refunds. Keep anything with a legal or financial consequence on scripted paths or with a person.
The AI providers clause. Section 4.7 of the Meta Terms for WhatsApp Business Platform bars providers of AI from the platform where AI is the “primary (rather than incidental or ancillary) functionality”, as Meta decides. An agent that serves your own customers about your business is the incidental case; an “ask me anything” assistant is not. The same section bars using platform data to train AI models, except to fine-tune one for your exclusive use. InfiQ’s glossary entry on the WhatsApp Business Solution Terms covers the wider contract.
The 24-hour window. Neither kind of bot can send free-form messages outside the window. Proactive messages need approved templates, whatever generates them.
A quick decision checklist
Answer these for each conversation type, not for the whole business:
- Is there one correct answer or sequence? If yes, use rules.
- Would a wrong answer cost money or break a rule? If yes, use rules or a person.
- Do customers phrase it in dozens of ways? If yes, AI helps.
- Do you have clean, current content to answer from? If not, AI will guess; fix the content first.
- Is the volume high enough to matter? Automate the frequent jobs first, whichever approach they need.
InfiQ’s chatbot design guide covers how to design the rule-based side once you know which jobs it owns.
Where InfiQ fits
InfiQ Flows, InfiQ’s drag-and-drop chatbot builder, is available and priced per flow, with 27 building blocks, sandbox failure drills and versioned publishing with one-click rollback. The InfiQ AI Agent is built on Meta Business Agent; its page marks the Business Agent integration as live and other Meta-announced capabilities as upcoming.
