Three tools, one honest comparison
Zapier vs Make
vs n8n
Almost every page comparing these three is written by someone who sells one of them. I sell none of them. I built a lot of n8n, including a self-hosted production pipeline, before moving my entire practice to custom code. That is a worse position for selling and a better one for telling you the truth, which is that all three are good and all three have a ceiling.
The short version
Zapier wins on speed to first result and on the sheer number of things it connects to. Make wins on price per operation and on handling branching logic without falling over. n8n wins on ownership: self-hosted, no per-task fee, your data on your infrastructure, and code nodes as an escape hatch. If you are choosing between them today with no other constraint, that is the order to think in: how fast, how complex, how much do you want to own.
Where each one actually stops
Zapier stops at cost. It is priced per task, so success is the thing that breaks it: the more your business grows, the more the bill grows, and none of that spend buys you an asset. Make stops at comprehension. It handles far more logic than Zapier, and the failure mode is a scenario with forty modules that technically works and that nobody dares to change. n8n stops at everything that is not a workflow: persistent state across runs, automated test suites, per-client data isolation, user interfaces, and volume beyond event-by-event processing.
The thing all three share
None of them can be tested the way software is tested. You verify a workflow by running it, usually against live data. That is entirely fine when a failed run is annoying. It stops being fine when a failed run sends the wrong invoice, misses an order, or silently stops without telling anyone. The gap is not features, it is the absence of a way to know something broke before your customer does.
The comparison, straight
| Zapier / Make / n8n | Custom system | |
|---|---|---|
| Time to first result | Minutes to hours, all three | 2 to 4 weeks for a first scope |
| Cost model | Per task, per operation, or your server bill | Built once, then nothing to rent |
| Branching logic | Zapier limited, Make good, n8n good | Native, and readable a year later |
| Testing | Run it and watch, on live data | Automated suites, staged rollout |
| State across runs | Workarounds via data stores | Queues, retries, state machines |
| Multi-tenant isolation | Not what any of them is for | Row-level security in the database |
| User interfaces | None of the three | Portals, approvals, dashboards |
| Failure visibility | You find out when someone complains | Monitoring, alerts, logged every run |
| Ownership | Zapier and Make: their account. n8n: genuinely yours | Repo, database and cloud on your accounts |
The four signals you have outgrown the tier
One: the monthly bill grows faster than the value. Two: exactly one person understands what runs where, and nobody wants to change it. Three: the workarounds have quietly become the system, whether that is a data store standing in for a database or code nodes holding the actual logic. Four: something failed and you found out from a customer. The fourth one is not a tool problem anymore, it is operational risk, and no amount of tier-shopping between these three fixes it.
What I would actually tell you
If the flow is linear, low volume and nobody needs a screen, use the tool. Pick Zapier for speed, Make for logic, n8n if ownership matters and you have someone who can run a server. If the process needs state, tests, an interface, per-client isolation or real data volume, no amount of picking between the three will help, because they all stop in the same place. And be suspicious of anyone whose answer never changes, including me: I have shipped both answers in production and both were right at the time.
Common questions, answered straight
Which one is cheapest?
At low volume, Make usually. At high volume, self-hosted n8n, because you pay for a server instead of per task. At any volume where the automation carries real revenue, the cheapest option over three years is usually the one you own outright, which is none of the three. One client was paying 10,000 euros a month for a rented setup; the replacement is theirs and rents nothing.
Can we start with one and migrate later?
Yes, and it is a reasonable plan. Migrating a workflow into custom code is a contained project, not a rescue mission, especially from self-hosted n8n where the logic and data are already yours. What is genuinely painful is migrating years of undocumented workarounds, so keep notes as you go.
Do you build in any of these?
Not anymore. That is exactly why my recommendation is worth something: if your problem fits one of the three, my teardown says so, and a specialist will build it far cheaper than I could.
What does the custom option cost?
Published, not negotiated in a call: the teardown is free, the audit is $490 and is credited toward the build, build sprints start at $6,900 fixed scope with 50% upfront, and operator retainers start at $2,900 a month, cancel monthly.
Related reading
Not sure which tier your problem belongs to?
Free systems teardown: your three biggest time sinks, what each costs per month, and an honest tool recommendation for each one. If the answer is Zapier, it will say Zapier. In writing, within 72 hours.
Get your free systems teardown