Custom WhatsApp chatbot vs no-code platform: how to decide
No-code platforms charge per message and retain your data. A custom chatbot costs more upfront and nothing later. An honest decision framework.
"I'll use a chatbot platform, it's cheaper and faster." It is the default decision, and for many cases it is the right one. But there is a set of situations where the no-code platform is the expensive choice disguised as cheap — and it is worth understanding when you are in one of them before you sign.
What the no-code platform solves well
Let us be fair: platforms like Intercom and similar solve the simple case well. If you need an FAQ flow, basic lead qualification, or answers to predictable questions, and the volume is low, building that on a platform in an afternoon is hard to beat. There is no server to manage, and the marketing team edits the flows on its own.
For an MVP, to validate whether the channel works, or for a business with modest volume and generic context, the platform is the rational answer. Do not build what you can rent cheaply.
Where the platform becomes a trap
The calculation changes when three factors come into play.
Per-message cost that grows with success. Platforms charge per conversation or per active contact. It works well at low volume. When the channel succeeds and volume rises, the cost rises with it — and you end up punished by exactly the growth you wanted. At high volume, the platform fee can surpass, within a few months, the cost of having built something of your own.
Lock-in and data ownership. The conversation history, the learning baked into the flows, the integration — it all lives on the platform. When you want to leave, the data stays. And in regulated sectors, customer data flowing through a third party can be a compliance problem that makes adoption impossible.
Deep business context. Platform chatbots answer the generic well but get the specific wrong. When good service requires querying your real knowledge base — internal policies, catalog, customer history — you need RAG over your content, not a pre-built decision tree.
The custom case: what changes
A custom chatbot inverts the cost structure: a larger investment upfront, and then almost nothing — the operational cost becomes your own server. You host, control, and can audit it. The model can be open source (Llama, Mistral) served on your infrastructure, or Claude via your own API when it makes sense.
The central piece is usually RAG (Retrieval-Augmented Generation): the chatbot answers based on your documents, citing the source. You update the content and it uses it on the next query, with no model retraining. This anchors answers in your real content and reduces hallucination.
We built exactly this for a gastroenterology clinic: WhatsApp with self-hosted Llama 3 and LangChain, on-premise, with 100% of the data local and LGPD-friendly. Per-message SaaS platform cost: zero. The patient data never left the clinic's server — a requirement no off-the-shelf platform would meet.
The decision framework
Go with a no-code platform if: volume is low, context is generic, you have no compliance requirement on the data, and you want to validate quickly without investment. It is the rational choice for the simple case.
Go custom if: volume is high enough for per-message cost to hurt, you need RAG over your knowledge base, there is a privacy or compliance requirement (healthcare, finance), or you simply do not want to be held hostage to a subscription and lock-in.
The honest question is not "which is more modern." It is "in 18 months, which is more expensive — in money, in data, and in freedom." Answer that with numbers from your real volume, and the decision becomes obvious.