Short answer. Pick the simplest tool your team can own. Zapier or Make fit low-risk workflows between SaaS apps, run by operations staff. n8n fits when you have a technical owner, higher volume or rules about where data lives. Custom code fits processes where a silent failure costs money, misses a deadline or fails an audit. Most companies end up with two layers.
Features rarely decide it. Three questions do. Who gets the alert at 03:00? What does a run cost at ten times today's volume? Where does the data sit?
The prices below come from the vendors' public pages, read on 23 October 2026. Prices change. Where a vendor does not publish a number, we say so and do not guess.
What are we comparing?
Four options for the same job:
- Zapier: hosted, with the largest app catalogue. Built for people who do not write code. You pay per task.
- Make: hosted, with a visual scenario builder. It gives you more control over branches, loops and error routes than Zapier. You pay per credit.
- n8n: a workflow tool you can rent as a hosted service or run on your own server. Code nodes are a normal part of it. Cloud plans count executions. The self-hosted Business plan has its own licence and quota.
- Custom code: a small service or a set of functions in your own repository. You choose the tests, version control and monitoring.
So the real choice is about two things: who maintains the workflow, and what happens when it breaks.
How much do they cost at 10k, 100k and 1M runs a month?
The vendors count different things. Zapier bills tasks, roughly one per action step that runs. Make bills credits: a module run normally uses one credit, and some AI modules (those using Make's own AI provider) use more (Make Help Center). n8n Cloud bills one execution per workflow run, however many nodes it has. The same workflow looks very different on each invoice.
To compare them, we assume a workflow with five steps: four billable on Zapier and five on Make. That is our assumption. Put in your own numbers.
| Runs per month | Zapier (tasks) | Make (credits) | n8n | Custom code |
|---|---|---|---|---|
| 10,000 | ~40,000 tasks. Pro 50,000-task tier: $289/mo billed annually, $433.50 billed monthly | ~50,000 credits. Entry plans start at 10,000 credits for roughly $9 (Core) to $16 (Pro) a month billed annually; the price of a 50,000-credit tier is not shown on the static page | Cloud Pro, 10K executions: €50/mo billed annually. Self-hosted Community: no licence fee, you pay for the server | Hosting plus build and maintenance time. No list price |
| 100,000 | ~400,000 tasks. Published tiers go up to 2,000,000 tasks (Pro $3,389/mo billed annually); the 100,000-task tier is $489/mo Pro, $599/mo Team, annual | ~500,000 credits. Not published on the static page; use the plan selector | Above the Pro limit on Cloud: contact sales. Self-hosted Business: €667/mo billed annually includes 40K executions; overage is listed at €4,000 per 300,000-execution bucket (billing of partial buckets is not stated) | Hosting grows with load. Most of the cost is engineering time |
| 1,000,000 | ~4,000,000 tasks. Not published. Enterprise, custom | ~5,000,000 credits. Not published | Self-hosted Business overage tiers or Enterprise. Self-hosted Community runs on your own infrastructure, sized by your queue and database | As above. Infrastructure cost grows slowly. Build cost stays the same |
Sources: Zapier pricing and n8n pricing, both read 23 October 2026. Make's 10,000-credit entry range comes from a third-party summary dated 26 September 2026, because Make's pricing page shows prices only in its interactive selector. Treat the Make number as a guide.
Three things stand out.
1. Per-step pricing makes long workflows expensive. On Zapier and Make, a twelve-step enrichment flow costs three times as much as a four-step one. On n8n Cloud both cost the same. If your workflows are long, the billing model matters more than the headline price.
2. Published prices stop before the volume where they hurt. Past the largest self-serve tier, Zapier and n8n Cloud send you to sales. At 1M runs a month you will negotiate, so plan a budget for that and not for the list price.
3. Self-hosting moves the cost to your side. You pay for the server, backups, upgrades, security patches and the person on call. Once it is set up, that is often a fraction of a day a month. Setup and incidents take much more. We do not quote an hourly figure because it depends on your team. Estimate it before you compare it with a SaaS invoice.
Zapier also documents what happens past your limit. With pay-per-task turned on, extra tasks cost more than the base rate (the page states 1.25× on annual plans and 2.5× on monthly plans), and usage pauses at three times your plan limit (Zapier pricing). Make states that scenarios stop when credits run out unless you buy more (Make pricing). Both rules are fine if you know them before a month-end spike.
What happens when a run fails?
The tools differ most here, and vendor comparison pages say the least about it. Ask three questions of any platform. Does it retry? Does it keep the failed data so you can replay the run? Who gets told?
- Zapier lets you replay failed runs and sends error emails. Anything more needs extra Zaps: opening a ticket, checking that a retry will not post twice, or alerting a channel. Check Zapier's current help pages for which plans include replay and whether replayed steps use tasks.
- Make has error handlers on each module. It stores incomplete executions so you can resume them and keep the data. For multi-step flows this is a real advantage over Zapier.
- n8n lets you set an error workflow. It starts with an Error Trigger node and receives the execution ID, the error message and the last node that ran (n8n docs). The Stop and Error node lets you fail a run on purpose. You decide what counts as a failure, so you also have more to design.
- Custom code can do all of it: retries with backoff, dead-letter queues, idempotency keys and alerts in your own on-call tool. You also have to build all of it.
Idempotency matters more than the retry button. If the same event arrives twice, or a step retries after a timeout, do you send two invoices? No platform answers that for you. In a no-code tool you check before every write. In code you put a unique key on the write. If a duplicate costs money or creates a legal problem, use code or a very disciplined workflow.
Our own rule is to check the result after every action runs. In one client project, an alert about missing order lines turned out to be false: a data source was 15 to 40 minutes behind. The fix became a standing rule: re-check before you alert. The B2B operations automation write-up shows ten processes on n8n built this way. Slack is the interface, a person approves anything that involves money with one click, and every failure reaches a named owner.
Some processes need more than a chain of triggers. In solar and heat-pump installation, a grid-connection application passes through a fixed sequence of approvals. Commissioning must stop when a required approval is missing. That is a state machine with an audit trail. A workflow engine can run it, but the states and rules should still be written down and testable.
What about GDPR, EU hosting and self-hosting?
A tool cannot be "GDPR compliant" by itself. Compliance comes from where data is processed, which processors touch it, what you keep and why. The tools differ in how much of that you control.
- Hosted tools (Zapier, Make, n8n Cloud): sign a data processing agreement with each vendor and keep a record of their sub-processors. Check the data region options on your plan before you commit. If personal data of people in the EU goes to a vendor in a third country, you also need a valid transfer mechanism. Confirm the details on each vendor's trust page. This article is not legal advice.
- Execution logs hold data. Every tool above stores run inputs and outputs for debugging. A workflow that handles CVs, patient data or payroll leaves personal data in the log. Keep logs only as long as debugging needs, and mask fields where the tool allows it.
- Self-hosted n8n: you choose the region and the retention. You also own patching, backups, access control and the safety of the credentials stored in the instance. One licence detail matters. The self-hosted Business licence key contacts n8n's licence server daily and reports production execution counts (n8n pricing). If that rules out an air-gapped setup, talk to n8n before you buy. Read the Community edition licence against your use case too.
- Custom code: the same control as self-hosting, with fewer parts written by someone else.
A practical rule. If the workflow moves personal data about people outside your clients' staff, such as patients, job candidates or tenants, default to self-hosted or custom code in an EU region with short log retention. If it moves business contacts between a CRM and a mailing tool, a hosted plan in an EU region with a signed DPA is usually enough. Your data protection officer makes the final call.
How do you see what your automations are doing?
You should be able to answer "did yesterday's run for customer X succeed?" in under a minute. If you cannot, the setup is not ready for production. Compare the tools on four points:
- Run history: how long it is kept, and whether you can search it by a business ID such as an order or invoice number, and not only by time.
- Alerts: whether a failure reaches a person in a channel they read. An email to someone who left last year does not count.
- Metrics: whether you can chart failure rate and run time over weeks. Most no-code tools give you run lists. Few show trends.
- Ownership: a named person for each workflow. No tool gives you this. Missing owners are the most common reason automations decay.
n8n lets you stream logs and reuse one Error Trigger workflow across many workflows (n8n docs). Make's execution history is good for debugging one scenario at a time. Zapier's run history is clear for a few simple Zaps and harder to use with dozens. With custom code, logs go where the rest of your monitoring already lives. For us, one place to look is the main practical reason to move critical flows out of a workflow tool.
When is Zapier fine?
More often than engineers like to admit. Stay on Zapier when:
- the workflow connects two or three mainstream SaaS apps that Zapier supports well;
- a failure is annoying but cheap, like a missed Slack notification or a lead not copied to a spreadsheet;
- the volume is far below the point where the invoice matters;
- the person who built it will also fix it, and they are not a developer;
- building it yourself would take a week, and a Zap takes an hour.
Fast setup has real value. A workflow that runs next Tuesday beats a better one planned for next quarter. Moving a cheap, working Zap to a more fashionable tool is a bad trade. The next section covers when Zapier stops being fine.
When should you move from no-code to code?
Use these exit signs. If two apply to a workflow, plan the move for that workflow only.
- A failure has a direct cost. Missed payments, duplicate orders, a regulatory deadline, a customer-facing outage. You need tests, alerts and rollback, and the tool does not give you those.
- The logic has state. Several approvals, timeouts, exceptions by country or customer type. The canvas becomes unreadable and nobody wants to edit it.
- The scenario has more than about 20 to 30 steps or many nested branches. This is a judgement call. The sign is a new colleague who cannot explain the flow in ten minutes.
- Changes need review and history. If every change needs code review, a staging run and a clear rollback, you have outgrown a visual editor, however good its version history is.
- The pricing works against you. Long workflows on per-step pricing, or volume that pushes you into sales-only tiers.
- Data must stay on your infrastructure, and even a self-hosted workflow tool leaves too much to secure.
- You hit the platform's limits on run time, payload size, concurrency, or an API the tool does not support well.
It works the other way too. A simple, stable, cheap workflow gains nothing from a move to code. It only adds maintenance. Keep a short list of what runs where, with an owner and a review date for each.
A common setup is the core process (orders, jobs, approvals, billing) as a small service with tests and logging, plus a workflow tool at the edges for notifications, CRM updates and report delivery. Approval steps can live in either layer.
So which one should you choose?
The decision path we use with clients, in order:
- Who owns it? No developer: Zapier or Make. A technical owner or a partner: n8n or code.
- What does a failure cost? Little: any tool. A lot, or the process is regulated: code, or n8n with strict error workflows, tests and alerts.
- Where must the data sit? Anywhere in the EU under a DPA: hosted is fine. Restricted: self-hosted n8n or code.
- How long and how frequent are the runs? Short and rare: Zapier. Long and frequent: n8n or code, because per-step pricing adds up.
- Will this exist in three years? If yes and it is core, code. If it bridges the gap until you replace a system, a workflow tool.
Make usually beats Zapier once you need branches, loops or error routes. Zapier wins on catalogue size and on how fast a non-technical person can start. Between Make and n8n, the difference is mostly hosting and maintenance. n8n gives you self-hosting and code nodes. Make gives you a managed service and a friendlier canvas.
A warning about switching tools. Moving a workflow to another platform means rebuilding it. Do not migrate everything to save a few hundred euros a month. The rebuild time costs more than you save.
How do you check the numbers for your company?
Take three real workflows. For each, write down runs per month, steps per run, failure cost, data sensitivity and owner. Price them on each tool from the vendors' current pages. Then add the hidden costs: hosting, maintenance hours and on-call time. If you want the numbers worked out for one real process, the free automation blueprint shows where the hours go and what automating them is worth.
If the answer is still unclear, that is the work we do in business process automation. We map your workflows and pick a tool for each one. We build what should be code, hand over what should stay no-code, and name an owner for every workflow.
FAQ
Is n8n cheaper than Zapier? Per run, usually yes. n8n Cloud counts one execution per workflow run. Zapier counts a task for each action step. At low volume the gap is small. At a few hundred thousand steps a month it is large. Self-hosting removes the per-run fee but adds a server and maintenance work, so compare the total cost.
Is n8n GDPR compliant? A tool cannot be compliant by itself. Compliance depends on where data is processed, your data processing agreements, how long logs are kept and what personal data goes through the workflow. Self-hosting n8n in an EU region lets you choose the location and the retention. You still need a lawful basis and a DPA with every service the workflow touches.
When should we stop using Zapier? Look at three signs: the bill grows faster than the value, failures need handling the tool cannot express, or data must stay on your own infrastructure. If two of the three are true for a workflow, move that workflow. Leave the rest where they are.
Can Make or n8n replace custom code? For routing, syncing and notifications between SaaS tools, yes. For logic with many states, strict audit needs, large data volumes, or changes that need tests and code review, custom code is safer. Many teams run both: a workflow tool at the edges and code at the core.
Which one should we choose? It depends on four things: who owns the workflow, what a failure costs, where the data must sit and how long and frequent the runs are. Make or Zapier suit teams without developers and low-risk workflows. n8n suits teams with a technical owner or a partner. Custom code suits core processes that must not fail silently.
Prices and plan limits read from vendor pages on 23 October 2026. Vendors change them, so check the linked pages before you decide.
