Turns Out · · 15 min read
n8n vs Make vs Zapier vs custom code: how to choose an automation stack in 2026
No-code vs low-code vs custom automation: cost at scale, EU data residency, AI steps, error handling, and where n8n, Make and Zapier each break down.
Every pricing page in this category shows a monthly price next to a volume figure, and the volume figures are not the same thing. Zapier sells tasks, Make sells credits, n8n sells executions, and a workflow that runs 10,000 times a month consumes a different count of each. We have watched teams pick a no-code automation tool from the entry prices, build a few dozen workflows on it, and then meet the invoice where the volume tier, not the plan, set the bill.
So we stopped comparing plans and started comparing one workflow. Below is the arithmetic for a single six-step lead-intake flow on the two no-code platforms (Zapier, Make), the low-code one (n8n) and plain code, then the rows that decide the choice once the price is understood: EU data residency, AI steps, error handling, versioning and lock-in. Plan prices, units and limits were read on the vendors' own pages on 16 November 2026, on annual billing; n8n prices in euros and Make and Zapier in US dollars, and we quote each as listed without converting. Make prints its plan prices only in a region-aware selector, so its figures come from dated price-tracker snapshots, labeled where they appear. Verify every figure at the time of reading.
The totals first, for the reference flow described in the next section:
Reference flow, annual billing, list prices read 16 Nov 2026
10,000 runs/month 100,000 runs/month
n8n Cloud Pro, 10K executions: €50 (at cap) above Pro: Enterprise quote or self-host
Make 50K credits → 80K tier: selector 500K credits → Pro/Teams ladder: selector
Zapier 40K tasks → 50K tier: $289 400K tasks → 400K tier: $1,269
Custom code VM + build + upkeep (example below) same VM; build pays back in monthsWhat one run costs on each platform
The reference workflow is the one almost every client asks for first: a form submission arrives by webhook, test entries are filtered out, the CRM is checked for an existing contact, an LLM classifies the lead, the CRM record is updated, and Slack gets a message. One trigger, one filter, four actions.
Each vendor meters it differently:
- n8n counts workflow executions — one run of the whole workflow is one execution "regardless of complexity", per the pricing page. Six steps or sixty, the meter moves once.
- Make counts credits — one credit per module operation, including the trigger module's check. Filters, routers and error handlers consume nothing. Our flow has five modules (the filter is a setting between modules, not a module, so it costs nothing), so five credits per run.
- Zapier counts tasks — a task is counted "whenever Zapier successfully completes a unit of work", and failed actions are not counted. Triggers, polling, filters, formatters and paths are free. Four actions, four tasks per run, counting the AI step at one task: a provider app step, a bring-your-own-key step or a Standard-tier AI by Zapier model. Zapier's model-tier rate card charges 3 tasks for an Advanced model and 5 for Premium, the default for a new AI by Zapier step on a paid plan, so the same flow is 6 or 8 tasks per run on those tiers; 40,000 is the floor. One MCP tool call costs two tasks.
- Custom code counts nothing — a queue worker on your own server bills by the month, not by the run; you pay in engineering hours.
Make has one more meter that catches people: a polling trigger spends a credit every check, whether or not it finds data. The reference flow uses an instant webhook; a scheduled trigger does not:
15-minute polling: 4 checks/hour × 24 × 30 = 2,880 credits/month before any work is done
1-minute polling: 60 checks/hour × 24 × 30 = 43,200 credits/monthZapier polls too (every 15 minutes on Free, 2 minutes on Professional, 1 minute on Team), but its pricing page is explicit that polling checks are not tasks. On n8n the meter is the execution, so test whether your trigger node produces one on an empty check before budgeting.
No-code automation pricing at 10,000 and 100,000 runs a month
At 10,000 runs of the reference flow:
- n8n Cloud — Pro includes 10,000 executions at €50 a month on annual billing (Starter is 2,500 at €20; the page prices annual billing 17% below monthly). Unlimited active workflows on every paid plan.
- Make — 50,000 credits needs the 80,000-credit tier, which only the selector prices. The entry tier is 10,000 credits a month; Make prices by region, and two trackers that read the selector in September 2026 report $9 / $16 / $29 (Jet Admin) and $12 / $21 / $38 (Coworker) a month on annual billing for Core, Pro and Teams. Read the selector from your own country.
- Zapier — 40,000 tasks lands on the 50,000-task tier: $289 a month on Professional with annual billing, per Zapier's own tier tables (10,000 tasks is $129; the 750-task entry tier is $19.99). With the AI step on an Advanced or Premium model the flow needs the 100,000-task tier at $489.
- Custom code — a Laravel queue or a Node worker on a small VM, plus the hours to build and watch it. Per-run cost is close to zero; the first month is not.
Two footnotes on the n8n line. 10,000 runs on a 10,000-execution plan leaves no headroom, and an error workflow is a workflow of its own, so its runs are executions too. The pricing page lists overage fees as “capped” on Starter and Pro without printing a figure, and for Business and Enterprise says workflows keep running past the quota with overage charges. There is no hosted tier priced between Pro and Enterprise, so at this volume you either accept the overage, ask Sales, or self-host.
At 100,000 runs the picture changes shape, not scale:
- n8n — the hosted plans priced on the page stop at Pro's 10,000 executions; Business (40,000 executions, €667 a month) is sold as a self-hosted license. 100,000 hosted executions is an Enterprise conversation, or you self-host the community edition and pay only for the server and the hours, costed below.
- Make — 500,000 credits a month, which is deep into the Pro or Teams ladder; at this volume Make automation is priced by tier, not plan.
- Zapier — 400,000 tasks: $1,269 a month on Professional with annual billing per Zapier's tables (100,000 tasks is $489, 200,000 is $769). Past the plan's tasks, Zapier switches to pay-per-task at 1.25 times the base rate on annual plans and pauses at three times the subscription.
- Custom code — the same VM, probably. This is the volume where a build pays for itself; the worked example below puts a number on it.
To make “depends on your hourly rate” concrete, one worked example with every assumption visible. Substitute your own numbers; the shape of the result is what matters:
Assumptions (illustrative): VM €20/month · build 20 h · upkeep 2 h/month · developer €80/h
(euro and dollar treated at par to keep the arithmetic readable)
Custom worker build 20 h × €80 = €1,600 once · running €20 + 2 h × €80 = €180/month
Self-hosted n8n no build, the flow is clicked together · running €180/month · no execution cap
vs n8n Cloud Pro €50/month custom saves −€130/month → never pays back at 10,000 runs
vs Zapier 50K $289/month custom saves ≈ €109/month → break-even around month 15
vs Zapier 400K $1,269/month custom saves ≈ €1,089/month → break-even in month 2Read it as a shape. At 10,000 runs the platform wins unless residency or retry semantics force the build; at 100,000 runs the build is paid for before the second invoice. Self-hosted n8n costs about what the custom worker costs to run, with no build and no execution cap, which is why it is the default in the second archetype below.
The honest caveat: below roughly 1,000 runs a month all four options cost less than a lunch, and the right choice is whichever your team will maintain. The arithmetic only decides anything past a few thousand runs.
EU data residency and self-hosting
"Do you keep our data in the EU?" gets three different answers, and only one of them is a plain yes.
- n8n — the pricing FAQ states that for hosted plans "data is stored within the EU, on servers located in Frankfurt, Germany". Self-hosting puts it wherever you put it.
- n8n self-hosted license — the self-hosted edition ships under the Sustainable Use License 1.0: use and modify it "for your own internal business purposes" without paying, but files marked
.ee.need an Enterprise license, and you may pass it on to others only free of charge and for non-commercial purposes. - Make — an organization sits in a data zone: US1, US2, EU1 or EU2. That covers scenario execution. It does not cover AI: Make's own help page says selecting EU1 or EU2 "does not automatically mean every request gets processed and stays in the EU", several built-in models are served from US or UK regions, and bringing your own AI provider key "removes Make's EU guarantee entirely".
- Zapier — the data privacy page is unambiguous: "Zapier hosts data in AWS servers located in the United States", and to the question of EU-only storage, "Zapier does not support this option." Transfers rest on the EU–US Data Privacy Framework and standard contractual clauses; Zap history is kept roughly 29 to 69 days, or a custom 7 to 30 days on Company and Enterprise plans.
- Custom code — residency is whatever your infrastructure already promised, which is usually the point.
If a DPA names an EU region, Zapier is out regardless of price, Make needs its AI steps reviewed one by one, and n8n is the default. "Self-hosting is free" needs its footnote here: someone has to patch it and answer the pager, which the worked example above budgets as a small VM plus a couple of hours a month.
AI steps: who bills the tokens
All three platforms put an LLM step a click away, and bill it three ways.
- Zapier — built-in AI steps draw from the same task allocation as everything else, and the pricing page notes that the AI model tier affects how many tasks an action uses; the OpenAI or Anthropic app steps cost tasks like any other action while the provider bills you for tokens. An MCP tool call is two tasks. Zapier Agents are metered separately, in activities.
- Make — a third-party AI app (OpenAI, Anthropic, Gemini) is one credit per operation and you pay the provider for tokens. Make's built-in AI provider instead charges credits by token: on the Small tier, 18,080 input tokens or 2,260 output tokens per credit at the time of writing.
- n8n — the AI Agent and model nodes run with your own API keys inside one execution; a classification step adds only the provider's token price.
- Custom code — a direct API call: cheapest per request at volume, and the only option where you control retries, caching and model routing.
The deciding question is rarely price but whether you can evaluate the AI step: log prompt and answer, sample weekly, measure drift. That is one line of code in a custom worker, two extra nodes in the same n8n workflow, and a design exercise in Make or Zapier.
Error handling, retries and versioning
This is where price per run turns into price per incident.
- n8n — each node has its own retry settings and an "on error" branch; a workflow-level error workflow starts from the Error Trigger and can alert Slack or open a ticket. Version history keeps 24 hours of changes for everyone and 5 days on Cloud Pro; full history is an Enterprise feature, and Git-based source control with environments is Business and Enterprise.
- Make — connection and timeout errors are retried with exponential backoff, eight reruns on one of two schedules: with incomplete executions stored, the gaps run from 1 minute up to 3 hours (1, 10, 10, 30, 30, 30 minutes, then 3 hours twice); without them, from 1 minute up to 24 hours (1, 2, 5, 10 minutes, then 1, 3, 12 and 24 hours). Failed bundles can be parked as incomplete executions to fix by hand, and error handler routes (break, resume, ignore, rollback, commit) attach to any module.
- Zapier — Autoreplay on Professional and above re-runs any errored Zap run across the account; the Free plan allows manual replay of errored runs only. Importing and exporting Zaps as JSON is a Team and Enterprise feature, so on lower plans the editor is the only copy of the logic.
- Custom code — nothing until you write it: retries with jitter, a dead-letter queue, idempotency keys, alerting, tests. The platforms give you most of this on day one; the remaining part (a retry that must not double-charge a customer) is the part you cannot configure.
Observability follows the same ranking: n8n and Make show per-step inputs and outputs for every run within the retention window, Zapier's Zap history does the same for tasks, and custom code shows what you logged. Much of a platform's price is this debugger.
Lock-in: what you can take with you
- n8n — an n8n workflow exports as JSON and runs on any n8n instance, including one on your own server. Lowest lock-in of the three platforms.
- Make — scenarios export as blueprint JSON, but nothing else runs a blueprint. Portable as documentation, not as software.
- Zapier — export requires a Team or Enterprise plan; on Free and Professional a Zap lives in the editor. There is no runtime outside Zapier either way.
- Custom code — yours entirely, including the maintenance and the person who understands it.
Three archetypes, three answers
The matrix collapses to three situations.
A solo marketer or a two-person ops team with under 2,000 runs a month and no personal data beyond names and emails: Zapier or Make. The app catalog, the instant webhooks and the fact that the next person can read the flow matter more than $20 a month. Catalog size is the row where Zapier wins outright: its pricing page claims 9,000+ apps and its directory counts over 10,000, against Make's 3,500+ and n8n's 2,000+ (the vendors' own pages, with n8n's HTTP Request node covering the rest). Make suits scenarios that branch and loop; Zapier automation suits flows where every step is "when X, do Y". Migrating a working Zap to n8n to save $30 a month is a loss.
An ops team holding EU customer data, 5 to 50 workflows, a developer available part-time: self-hosted n8n, or n8n Cloud in Frankfurt if nobody wants the server. Budget the operations honestly (updates, backups, staging; the worked example above lands near €180 a month at a placeholder rate) and use the Business tier's Git environments once more than two people edit workflows. The other half of this archetype is data: if the CRM and the ERP disagree about who a customer is, no platform fixes that, and a free AI readiness assessment finds that out before the first workflow does.
A product-critical flow (orders, payments, onboarding, anything a customer sees fail): custom code with a queue, tests and alerting. Laravel queues, BullMQ or SQS, a worker you can deploy and roll back, with n8n or Make for the glue: notifications, enrichment, reporting. Our decision rule: if the steps are known in advance and a failure costs money, write it; if the steps change monthly and a failure costs an email, click it.
What we advise against in all three cases: moving a flow that works to save a small monthly fee; self-hosting anything without a named owner; and putting a payment or a legal-notice step in a no-code tool where the retry semantics are the vendor's, not yours. For the longer list of workflows that pay back first, and the ones that do not, see 12 workflows worth automating first.
Frequently asked questions
Is n8n cheaper than Zapier?
Per run, usually yes, because n8n automation bills a workflow execution and Zapier bills every action inside it: 10,000 runs of a four-action flow is €50 a month on n8n Cloud Pro against $289 on Zapier's 50,000-task tier, or $489 if the AI step runs on an Advanced or Premium model. At 500 runs a month the difference is a rounding error, and self-hosted n8n adds server and maintenance cost.
Which no-code automation tool keeps data in the EU?
n8n Cloud states Frankfurt for hosted plans and can be self-hosted anywhere. Make offers EU1 and EU2 data zones for scenario execution, with the caveat that built-in and bring-your-own AI steps may process outside the EU. Zapier hosts in the United States and does not offer EU-only storage. Confirm the current pages before signing a DPA.
Can I move from Zapier to n8n or Make later?
Zapier exports Zaps as JSON only on Team and Enterprise plans, and nothing else runs that JSON, so a migration is a rebuild. Keep a one-page runbook per Zap (trigger, steps, field mappings, known failures); it turns the rebuild from archaeology into typing.
When is custom code the right Zapier alternative?
When the steps are known in advance, the volume is in the tens of thousands a month, a failure costs money, and you need tests, idempotency and your own retry logic. If any of those is missing, a platform is cheaper for the first year, and probably the second.
If you want a second opinion on which of the four fits a specific set of workflows, including the ones already running, our AI automation and integration service starts with exactly this arithmetic on your own volumes.