Make, n8n, Zapier or custom: which fits your small business?
When an assembly tool is enough, when it is not, and how each option bills. A decision table, five signals and three questions to settle it.

Make, n8n, Zapier or custom development: which one should a small business pick?
Direct answer: start with an assembly tool (Zapier, Make or n8n) as long as the process is simple, low-volume and harmless if it fails. Move to custom development when the process becomes critical, when it outgrows what the tool bills reasonably, or when the business logic no longer fits in visual blocks. In most small companies both coexist: the tool for the simple, code for the core.
Almost every comparison on this topic is written by a provider who sells one of the options. This one is no exception: the studio builds custom tools. So we also say, in detail, when custom is not the right answer. The plan: what each option does well, how each one bills, a decision table, then the signals that show a tier is no longer enough.
What each option does, in one sentence
- Zapier: the fastest to learn and the widest catalogue (over 9,000 apps according to its pricing page). It suits short chains between two or three tools.
- Make: a visual editor for building scenarios with branches, loops and data transformations. It takes longer to learn than Zapier, in exchange for more flexibility.
- n8n: a tool of the same kind, with a self-hosted version available on GitHub and billing per complete workflow execution. It needs someone able to administer it.
- Custom development: a program written for your process, hosted by you or by a provider of your choice, that reads and writes in your tools through their interfaces.
How each one bills: the rule that changes everything
The sticker price matters less than the billed unit. The unit decides what happens when volume grows, based on the official pricing pages consulted in November 2026.
- Zapier bills per task. A task is one successful step of a Zap. Triggers, built-in tools such as Formatter, Filter or Delay, and failed actions do not count.
- Make bills per credit. Most actions use one credit, reading data included. A scenario run can use anywhere from two to several thousand credits. Error-handling modules and the router use none.
- n8n bills per execution. An execution is one complete run of the workflow, whatever its number of steps. Users and workflows are unlimited on the cloud plans.
- Custom does not bill per use. The cost is an upfront investment, then hosting and maintenance. Volume barely changes it.
A worked example, as a hypothesis. A quote follow-up has six steps: read the CRM, test the date, look up the history, draft the message, queue it for approval, update the record. With 1,000 quotes followed up per month, this process weighs about 5,000 tasks at Zapier (the trigger is free), around 6,000 credits at Make, and 1,000 executions at n8n. Same process, three bills of a different nature. At small volume the difference is negligible. After a year of growth, it is not.
Decision table
| Criterion | Zapier | Make | n8n | Custom |
|---|---|---|---|---|
| Billed unit | Task (successful step) | Credit (action) | Complete execution | None: upfront cost, then maintenance |
| Time to start | A few hours | A few days | A few days, plus installation | Several weeks |
| Complex logic (branches, loops, edge cases) | Quickly cramped | Good, readable up to a certain size | Good, code allowed in nodes | No ceiling |
| Where your data goes | Through the vendor | Through the vendor | Stays with you if self-hosted | Stays with you |
| Who fixes it when it breaks | You | You | You (self-hosted) or the vendor (cloud) | The provider, per the support contract |
| Best ground | Simple chains, testing an idea | Multi-step scenarios run by an autonomous team | A team with a technical profile, sensitive data | Critical process, proprietary business logic |
This table compares categories of tools, not frozen offers: plans and prices change, and they must be checked on the vendors’ pages before any commitment.
When an assembly tool is enough
An assembly tool is enough when the process meets four conditions: it links applications that already have a connector, it has few steps, its monthly volume stays modest, and a mistake can be recovered without damage (a duplicate in a list, a reminder sent late). Typical cases: a notification to the team chat, a form copied to a spreadsheet, attachments filed into folders.
In that case, custom would be a luxury. A tool you configure in an afternoon, understand, and can switch off with one click is a better investment than a program written for the same thing. With no technical profile on the team, Zapier or Make are the safest entry points. With one, and with data that should not pass through a third party, self-hosted n8n deserves a trial.
Five signals that a tool is no longer enough
- The bill follows volume, not value. The same process costs more and more while earning no more. Compute the cost per successful execution: if it exceeds the cost of the hour of work avoided per batch, it is time to compare.
- The scenario looks like spaghetti. More than twenty blocks, nested branches, exceptions added over the months. Nobody but its author can change it without breaking it.
- A mistake is expensive. A quote sent to the wrong client, a wrong accounting entry, a payment triggered twice. As soon as failure has a consequence, you need checks, approvals and logs that standard blocks handle poorly.
- You work around the tool. Hacked-together code steps, intermediate spreadsheets, calls to interfaces the connector does not support. The tool is no longer a gain, it has become a disguised obstacle.
- The process is your trade. A calculation rule specific to your firm, a risk weighting, an internal scale. That logic is an asset: leaving it in a third party’s interface means losing control of it.
Two or more signals is a serious reason to price a move to custom. A single signal is often fixed by reorganising the scenario. To measure what the process really costs before deciding, the express diagnostic in five questions gives a first order of magnitude.
Three typical situations
- A three-person firm that copies each new engagement into its shared calendar. Volume: about ten a month, no consequence if it runs late. A Zap or a Make scenario settles it, and custom would be out of proportion.
- An agency that follows up its quotes and has seen its tool bill double in a year. The process grew, it touches customers, and the follow-up logic became more complex. This is the moment to price rewriting the core in code, while keeping the rest in the tool.
- An independent manager whose weighting calculation is specific to their method. The rule is a confidential asset and it must be tested. It belongs in code you own, not in a scenario hosted by a third party.
The hybrid strategy, the most common one
The real answer is rarely an exclusive choice. A company that automates seriously ends up with two layers. At the bottom, assembly tools for simple links with no stakes. On top, a small number of processes written custom: the ones that touch money, customers or compliance. The two talk through webhooks or interfaces, and neither needs to replace the other.
It is the same logic as automating without replacing your tools: plug the new piece into what exists rather than rebuild everything. A typical case: the quote follow-up runs custom, with a message drafted by the AI and approved by a person (see the follow-up after a quote), while copying new contacts to the newsletter stays in an assembly tool.
When custom is not the right choice
- The process does not exist yet. If you do not know how the task should be done, automate it by hand first, then in a light tool. Coding an unstable process freezes a mistake.
- Volume is low and so are the stakes. A few dozen executions a month on a harmless task: the tool is cheaper and can be undone in five minutes.
- Nobody will be responsible. A custom program needs maintenance. Without a support contract or an internal owner, it ages badly, like an orphaned scenario, only more expensive.
- A standard connector already does exactly the job. Writing what comes in a kit adds nothing.
How to decide in three questions
- What happens if it fails silently for a week? If the answer is “nothing serious”, stay on a tool. If it is “a lost client” or “an entry to correct”, move to a solution with checks and logs.
- How many executions per month in twelve months? Multiply the current volume by your growth rate, then recompute the bill with each tool’s unit.
- Who maintains it in three years? A name, not a department. With no answer, the default is the option easiest for a third party to take over.
If the three answers point to custom, that is exactly the kind of work the studio takes on, within its limits: the studio’s three missions are described on the home page. If they point to a tool, take the tool, and keep this comparison for the day volume has changed.
Frequently asked questions
Is Make cheaper than Zapier?
Often, at equal volume, because the billed units differ: Zapier counts successful steps, Make counts actions as credits. The comparison depends on the shape of your scenarios. Compute the cost of your real process with each vendor’s pricing page rather than relying on an average.
Is n8n free?
A self-hosted version is available on GitHub, but hosting, security, updates and monitoring remain your responsibility. The cloud plans are paid and billed per complete workflow execution. The real cost therefore includes administration time.
When should I move to custom development?
When at least two signals are present: a bill that follows volume rather than value, a scenario nobody but its author can change, an error with a real consequence, repeated workarounds, or business logic that is your own edge.
Can I combine an assembly tool with custom work?
Yes, and it is the most common case. Simple links stay in the tool, processes touching money, customers or compliance are written custom, and the two exchange through webhooks or interfaces.
What happens to my automation if the vendor changes its prices?
It keeps running, but its bill changes. Vendors alter their plans and sometimes their billing unit: Make states, for instance, that credits have replaced operations. This dependency risk is one reason to keep critical processes in code you own.


