WhatsApp automation: n8n or a custom backend?

The question underneath the question

Almost every guide compares tools. Very few start where the risk actually sits: how you connect to WhatsApp at all. There are two routes, and they are not equivalent.

Meta's WhatsApp Business Cloud API is the sanctioned route. It needs a verified business and a developer app, which is friction, and it is stable. Automating on it will not get your number banned, because that is the point of it existing.

Unofficial gateways log in to WhatsApp Web as though they were a person, running a headless browser or socket client, and expose that session as an API. Free, no business verification, and what a large share of automation tutorials quietly use. They also reverse-engineer a service whose terms forbid exactly this, and Meta bans numbers it detects behaving like a bot.

Weigh that properly. If WhatsApp is where customers reach you, the number is not a technical detail. It is on your invoices, your signage, and in the phone of every customer you have served. Losing it is not an outage you restore from backup. Our honest position: if the number matters, do not build on an unofficial gateway, whatever it saves this month. That decision is independent of everything below, because you can run the official API through n8n and an unofficial one from custom code. Settle the API first, then pick the tool.

What n8n is genuinely good at

n8n deserves its reputation, and not only because it is quick.

• Time to first working version. A flow that receives a message, checks a sheet and replies is an afternoon, not a sprint.
• Breadth of connections. Several hundred integrations already exist. If the job is moving information between WhatsApp and the eight other systems a business runs, that library is the product.
• Readable by someone who did not build it. A visual flow is legible to an operations manager in a way a code repository is not.
• Self-hosting removes per-operation charges. Past roughly ten thousand executions a month, a small server is meaningfully cheaper.

For notifications, lead routing, order updates, and connecting WhatsApp to a CRM or spreadsheet, n8n is very often the right answer, and reaching for custom code instead is overbuilding.

Where n8n starts to hurt

The limits are real, and they arrive in a predictable order.

Conversations have memory and workflows do not. This is the big one. A workflow tool runs a trigger through steps to completion. A conversation is a state machine: the customer replied "2", and what that means depends on what you asked three messages ago and what they already confirmed. You can model state in n8n, but it gets awkward fast, and the awkwardness compounds with every branch.

Complex flows resist review. A workflow is stored as a large JSON document. Fine until two people maintain it, at which point there is no meaningful code review, no readable history, and no way to test a branch in isolation.

Self-hosting is not free, it is shifted. You own updates, scaling, monitoring and the pager: a fair trade at volume, a poor one at ten messages a day.

The licence matters if you are building a product. n8n ships under a sustainable use licence rather than a standard open source one. Irrelevant for internal automation. If you intend to embed it in something you sell, read the terms before designing around it.

Where a custom backend earns its cost

Writing your own service against the Cloud API is more work up front, and buys specific things.

• Conversation state modelled properly. Sessions, timeouts, resumption and context become things you designed rather than worked around.
• Business logic that is actually yours. GST rules, Tally reconciliation, channel-partner hierarchies, approval chains. The same reason businesses outgrow a horizontal CRM, covered in building a CRM around your process.
• Testing and version control. Unit tests on the decision logic, reviewable history, safe rollback.
• Data stays where you put it. Under India's data protection rules, being able to say exactly where customer messages are processed is a real advantage.
• No ceiling. Throughput becomes a function of your infrastructure, not a workflow engine.

The honest cost: you rebuild what n8n gave you free. Retries, queueing, delivery receipts, template management and scheduling are yours now, and they are where the time goes.

What it actually costs

A point that gets lost: your build choice does not change what Meta charges. Both routes use the same API at the same rates.

For India as of January 2026, Meta's marketing template rate is around ₹0.86 per message, having risen roughly 10% from about ₹0.78, while utility and authentication templates sit near ₹0.115. On top of Meta's charge you pay your provider's platform fee and any per-message markup, commonly 10% to 30%, and 18% GST applies to both. India remains one of the cheaper markets for this globally.

So the real comparison is engineering cost, not message cost. n8n is cheaper to start and dearer to maintain as complexity grows. A custom service is the reverse. Where those lines cross depends on how complicated your conversations are, not how many messages you send.

The hybrid most businesses actually land on

Presenting this as a binary is tidy and slightly false. Most businesses end up with both, each doing what it is good at. A small custom service owns the conversation: state, replies, and the rules that decide what happens next. n8n sits around it as connective tissue, moving results into the CRM, the accounting system and the alerting your operations lead wants. Proper control where the logic is hard, a large integration library where it is not. Same principle as any sensible integration work, which we covered in why your systems don't talk to each other.

How to decide

• Under a thousand conversations a month, simple flows, no memory. Use n8n. Building a service here is overbuilding, and we would tell you so.
• WhatsApp is a revenue channel, conversations are multi-turn, it touches your ERP or accounting. Build the conversation layer properly. You will otherwise spend a year fighting a workflow engine.
• You intend to sell it. Custom, for control and for licensing reasons.
• You do not know yet. Start on n8n against the official API and move the conversation layer out when it hurts. The API choice is the expensive one to reverse, not the tool.

That last point matters. Choosing n8n costs little if you outgrow it, provided you built on the sanctioned API. Choosing an unofficial gateway can cost you the number your customers use.

The short version

Pick the official Cloud API unless the number is genuinely disposable. Use n8n when the work is connecting systems and the conversation is simple. Build a service when the conversation has memory and the rules are yours. Expect to end up with both. Whichever you choose, Meta's bill is the same, so decide on engineering effort rather than message price. Our workflow automation work maps the process first, which is often how we find a business needed a smaller fix than it expected. The fixed-scope version is one workflow automated in two to three weeks, more short answers are in the FAQ, and it starts with a thirty-minute discovery call.

Common questions

Can Meta ban my number for WhatsApp automation?

Not for using the official Cloud API, which exists to be automated. Numbers get banned when businesses automate through unofficial gateways that log in to WhatsApp Web and imitate a human session, because Meta detects that and its terms forbid it. If the number matters, use the sanctioned API.

Is n8n good for WhatsApp automation?

Yes, for notifications, order updates, lead routing, and moving information between WhatsApp and other systems. It fits poorly for long multi-turn conversations carrying state, because a workflow engine runs steps to completion while a conversation remembers what came before.

How much does the WhatsApp Business API cost in India?

As of January 2026, Meta charges roughly ₹0.86 per marketing template message and about ₹0.115 for utility and authentication. Your provider adds a platform fee and typically 10% to 30% markup, plus 18% GST on both. Identical whether you build with n8n or your own service.

Should I self-host n8n or use the cloud version?

Cloud is simpler and costs per operation. Self-hosting removes those charges and becomes cheaper past roughly ten thousand executions a month, at the price of owning updates, scaling and monitoring. Below that, it usually costs more in attention than it saves in fees.