n8n, Zapier or Custom: Picking an Automation Layer You Won't Outgrow
Broadleaf Organics, a spice brand in Nagpur, started with 14 automations on a no-code plan and a bill of about ₹1,800 a month. Eighteen months later they were pushing every order, every courier webhook and every review request through the same platform, and the bill had passed ₹47,000. Nothing had gone wrong. They had simply grown into a pricing model designed to grow with them.
This is the n8n vs zapier pricing conversation in one paragraph. The question is never which tool is best. It is which cost curve matches the volume you are heading toward, and when you should switch.
- Hosted per-task tools are the cheapest way to start and the most expensive way to run high volume.
- Self-hosted n8n trades a per-task bill for a server bill plus someone who maintains it.
- Custom code wins on cost at scale and on complex logic, and loses badly on speed to first version.
- The realistic migration triggers are volume, conditional complexity, latency and data residency.
- Most Indian SMBs should start hosted, move the two heaviest workflows to n8n or code, and keep the rest where they are.
The three layers, honestly described
Hosted no-code (Zapier and similar). Enormous connector library, genuinely good reliability, a non-developer can build something useful on a Tuesday afternoon. You pay per task or per operation, and every step in a multi-step workflow counts.
Self-hosted or cloud n8n. Fair-code, node-based, and crucially it lets you drop into JavaScript when the visual builder runs out. Self-host it and your marginal cost per execution approaches zero, which is the entire reason n8n vs zapier pricing comparisons keep landing on n8n for high-volume teams. The cost moves to infrastructure and upkeep.
Custom code. Queue workers in Laravel, Node or Python, running on your own infrastructure. Total control, no execution ceiling, proper version control and tests. Slowest to build and it needs developers who will still be around next year.
- Monthly executions are in the low thousands
- Nobody on the team writes code
- You need obscure SaaS connectors that already exist
- The workflows are linear and rarely change
- Downtime for an hour is an inconvenience, not a crisis
- Executions are in the tens of thousands and climbing
- Workflows branch, loop, and retry with real logic
- You need data to stay on infrastructure you control
- You want the workflow in git with a staging copy
- The per-task bill has become a line item people argue about
How the pricing actually works
The trap in per-task pricing is that a task is a step, not a workflow. A single order-confirmation flow that checks stock, writes to a sheet, sends WhatsApp, tags the customer and posts to Slack is five tasks. Broadleaf Organics were billing 60,000 orders-worth of steps and thinking of it as one automation.
| Layer | What you pay for | Marginal cost of run 100,001 | Who maintains it |
|---|---|---|---|
| Hosted no-code | Tasks or operations per month | Full per-task rate | Vendor |
| n8n cloud | Workflow executions, not steps | Low per execution | Vendor, mostly |
| n8n self-hosted | Server, storage, your time | Effectively zero | You |
| Custom code | Build hours, then hosting | Effectively zero | You or your agency |
That step-versus-execution distinction is the single biggest driver of the bill. A five-step workflow on a per-task plan costs five times what the same workflow costs on a per-execution plan, before you argue about the rate.
The cost curve and where it crosses
Illustratively, for a five-step workflow, here is the shape. Numbers are indicative typical market ranges rather than any vendor's current rate card, and you should price your own volume before deciding.
Broadleaf Organics crossed the first threshold quietly and the second one loudly. The tell was not the invoice. It was a Diwali week where they hit their task cap on a Saturday morning and order confirmations stopped going out for four hours. Volume ceilings hurt most on your best day.
Four migration triggers
Pricing your own crossover in ten minutes
Do this on paper before you talk to anyone. Broadleaf Organics' order-confirmation flow has five steps and fires on every order. At 6,000 orders a month that is 30,000 billable tasks on a per-task plan, and 6,000 executions on an execution-priced plan. Same work, five times the meter. Multiply by three more flows of similar shape and you can see how a ₹1,800 bill became ₹47,000 without anyone approving a change.
Now cost the other side honestly. A self-hosted n8n instance needs a VPS with enough memory to run queue mode rather than the default single process, a managed Postgres or a database you back up yourself, and roughly half a day a month of someone's attention for upgrades and the occasional stuck execution. Put a real rupee figure on that half-day. For most Indian SMBs the all-in n8n number lands somewhere in the region of ₹8,000 to ₹15,000 a month once you count the human, which is why the crossover sits in the tens of thousands of runs rather than the thousands.
One objection we hear often: "we tried n8n and it kept falling over." Nine times out of ten that was the cheapest available droplet, running in main mode, with no restart policy and no alerting. That is not a verdict on the tool. It is the same server you would have given any other service and expected to fail.
The hybrid answer most teams land on
Broadleaf Organics did not migrate everything, and they were right not to. They moved order confirmation, courier tracking and review requests, the three highest-volume flows, onto self-hosted n8n with a small worker service handling the parts that needed real logic. Everything else, the finance alerts and the monthly reports and the six automations that run twice a week, stayed on the hosted plan at the bottom tier.
Their combined bill landed at roughly a third of the peak, and the maintenance burden was one afternoon a month. That is the usual shape. Three workflows generate 90% of your executions; move those and leave the long tail alone.
Choosing without regretting it in a year
- Estimate your monthly executions at today's volume and at three times today's volume.
- Count steps per workflow, not workflows. Per-task plans bill steps.
- Ask whether anyone on the team can restart a stalled queue at midnight.
- Check whether the connectors you need exist, especially Indian payment gateways, couriers and the WhatsApp Business API.
- Confirm you can export your workflow definitions. Portability is the only real bargaining power you have.
- Decide where personal data may live before you build, not after.
The trade-off nobody enjoys admitting: self-hosting saves money and costs attention. If your team has no appetite for maintaining a server, a slightly higher hosted bill is genuinely the cheaper option, because a stalled automation nobody notices for two days costs more than the difference. We have watched that happen. It is not a fun invoice to explain.
Broadleaf Organics is a composite illustration; all volumes and costs above are examples.
- Start hosted. Speed to first working automation beats theoretical efficiency.
- Understand n8n vs zapier pricing as steps versus executions before you compare rates.
- Move only your top three workflows when volume, complexity, latency or residency forces it.
- Keep the long tail on no-code forever. Migrating it wastes money.
- Never pick a layer you have nobody to maintain.
DL Minds Team
Digital marketing and web development expert at DL Minds. Passionate about helping businesses grow through innovative technology solutions and strategic digital marketing.
