n8n, Make, Zapier or custom code: which one your automation should run on.

Three platforms dominate UK small business automation, and the difference between them that matters most is not the number of integrations or the look of the editor. It is the unit each one bills. Zapier counts tasks, one per successful action step. Make counts credits, one per module action including the trigger that reads the data. n8n counts executions, one per run of the whole workflow however many steps it has. The same five-step workflow therefore costs four units on Zapier, five on Make and one on n8n every time it runs, and that ratio compounds with volume until the cheapest tool at 500 runs is the most expensive at 50,000.

This guide reads the three pricing pages as they stood on 14 September 2026, works the per-run arithmetic at 1,000, 10,000 and 100,000 runs a month with the sums shown, prices self-hosting n8n from the vendor’s own documentation and a cloud provider’s price list, and compares what each publishes about data residency, lock-in and error handling. It ends with a decision rule and with where we stand, since we build on n8n where it fits and in code where it does not, and the quote says which.

Currency: n8n bills in euros, Make and Zapier in dollars. Conversions use the Bank of England spot rates for 11 September 2026, €1.1649 and $1.3530 to the pound. UK buyers add VAT. One caution before the tables: platform prices change more often than agency prices, and two of these three pages changed their units or tiers in the last year, so the date at the top of this page is part of the data. If a figure here disagrees with the page you are looking at, the page wins, and we would like to hear about it.

What each vendor shows on its own pricing page. Zapier and n8n list their tiers as text; Make lists base prices and puts its larger credit volumes on a slider whose prices are not published as text, which is recorded as such rather than guessed. The free tiers are worth a line of their own: Zapier’s allows two-step Zaps only and polls every fifteen minutes, Make’s allows 1,000 credits with a fifteen-minute minimum interval, and n8n has no free cloud tier but a free self-hosted edition, which is a different kind of free with a different kind of cost.

Read on 14 September 2026 from n8n.io/pricing, make.com/en/pricing, zapier.com/pricing, help.zapier.com, docs.n8n.io and digitalocean.com. Annual-billing prices unless stated. Zapier’s page shows monthly billing at 1.5 times the annual rate; n8n’s page shows annual prices only.

Take a workflow with one trigger and four actions: a form submission arrives, a record is created in the CRM, a row is written to a sheet, a Slack message is sent, a confirmation email goes out. On Zapier that is four tasks per run, because the trigger is free. On Make it is five credits, because the trigger module reading the submission counts. On n8n it is one execution. The table shows the cheapest published tier that covers each volume, the sterling conversion, and the cost per run. Where no tier is published, it says so.

Arithmetic: tasks = runs × 4; credits = runs × 5; sterling at $1.3530 and €1.1649 per pound. Zapier tiers from zapier.com/pricing; Make from make.com/en/pricing; n8n from n8n.io/pricing; server price from digitalocean.com. Self-hosting costs exclude the time of whoever maintains it.

At 1,000 runs Make is the cheapest hosted option by a distance, because 5,000 credits fit inside its 10,000-credit base tier at $9. Zapier is expensive at that volume not because of its rate but because of its tiers: 4,000 tasks sit just above the 2,000-task tier, so you pay for 10,000. Halve the workflow to 500 runs and Zapier drops to $49. Tier edges matter more than list prices at small volumes.

At 10,000 runs the per-step billing bites. Zapier needs the 50,000-task tier at $289 a month, which is £2,563 a year for one workflow. n8n Pro covers it for €50 a month, £515 a year, and self-hosting for £106 a year of server. Make’s price at 50,000 credits sits on its slider and was not published as text on the day we read it, so we cannot place it; it will sit between its $9 base and Zapier. Softomate’s UK guide puts the point at which migration off Zapier pays for itself at around £6,000 a year of Zapier spend, which this table reaches at roughly 25,000 runs of a five-step workflow.

At 100,000 runs none of the three publishes a cloud price that covers it. That is the honest crossover to code or to self-hosting: a Community edition n8n on a $12 to $24 server, or a small service written for the purpose, both of which cost the same whether they run ten thousand times or a million, with the caveat that somebody now owns a server.

n8n’s Community edition is free to run under its Sustainable Use License, which permits use for internal business purposes and prohibits offering it commercially to others. The documentation says the Community edition includes almost the complete feature set, and lists what it does not: environments, SSO with SAML or LDAP, Git version control, log streaming to external systems, projects, external secret stores, external binary storage and multi-main scaling are Business or Enterprise features. For a single business running its own workflows, none of those is a blocker on day one; log streaming and version control are the two you will miss first.

The server is cheap. n8n’s own Hetzner guide says a CPX11 is enough for a standard deployment and recommends at least 2 vCPU and 4 GB of RAM if the AI assistant sandbox is added. DigitalOcean lists a 1 GB droplet at $6 a month and a 2 GB droplet at $12, about £4.43 and £8.87. AI Agency Plus’s UK guide puts self-hosted n8n at £5 to £15 a month, which matches. Docker Compose with the built-in SQLite database and a Caddy reverse proxy is the documented path and takes an afternoon.

What you give up is somebody else’s pager. n8n’s Hetzner guide is blunt that self-hosting requires expertise in server configuration, container management and application security, and that mistakes can lead to data loss, security issues and downtime. Updates, backups, TLS renewal and the 3am restart are yours. The honest price of self-hosting is the server plus a few hours a month of somebody who knows what they are doing, or a supplier who includes it in a support arrangement you can leave.

For a UK business handling customer data, where the workflow runs and what it logs are not afterthoughts. Every automation that touches a customer record is processing personal data, and the platform it runs on is a processor whose location, retention and access controls you are accountable for under UK GDPR. The three vendors publish different amounts about it, and the gaps are as informative as the statements. The table records only what we read on the vendors’ own pages on 14 September 2026; where a cell says not stated, the right response is to ask the vendor, not to assume the worst or the best.

Only statements we read on the vendors’ own pages are recorded. "Not stated" means we did not find it on the pages listed in the sources, not that the vendor has no answer; ask before you sign.

A low-code platform is the right answer for most first automations: two well-supported applications, a trigger, a handful of steps, and a workflow your own team should be able to open and edit. It is fast to build, cheap at low volume and easy to change, and the person who understands the process can see the process on the screen, which is worth more than any technical property. Most of the automation we are asked for starts here and a good share of it should stay here.

Code wins in four situations, and they are recognisable in advance. The logic needs tests: once a workflow has more than a few branches, the only way to know it still works after a change is a test suite, and platforms do not have one. The volume is high: past tens of thousands of runs the per-unit tiers cost more per year than a small service costs to write. A failure is expensive: an automation that moves money, sends regulated messages or writes to a system of record needs idempotency, retries with backoff and an audit trail per record, which a platform gives you partially at best. And the data is sensitive: a self-hosted service in a UK region is the cleanest residency answer there is.

Code loses when the requirement is still changing weekly, when the people who will own it are not developers, and when the integration is to a SaaS product whose API the platform has already wrapped and maintains on your behalf. Rewriting a working Zap into a service because it feels more professional is a way to spend £4,000 saving £20 a month, and then owning a deployment nobody asked for. The test is whether anything in the four situations above is true today, not whether it might be true at ten times the volume.

The middle ground is the common case: n8n for the orchestration, with the two or three steps that need real logic written as small code nodes or as a separate service the workflow calls. That keeps the editable parts editable and the tested parts tested, and it means the platform decision is not a one-way door. A workflow that outgrows n8n Cloud moves to a self-hosted instance with a JSON export; a step that outgrows a code node becomes a service the same workflow calls. Neither move requires starting again.

The rule below is built from the tables above and from nothing else: the billing units, the published tiers, the residency statements and the four situations where code wins. It is deliberately blunt. Most of the cost of choosing wrong is not the monthly fee but the rebuild when the wrong tool hits its ceiling, so the rule errs towards the option that can grow. Apply it to each workflow separately; a business that runs Zapier for its lead forms and code for its invoicing has chosen well twice.

We build on n8n where it fits and in code where it does not, and the quote says which and why. A single agent or workflow is £4,000 to £15,000 fixed and a multi-agent system £8,000 to £25,000; the running cost is stated at your volume before go-live using the prices on this page, and the platform choice is part of that sum rather than a preference. Where n8n is the answer we set it up in your account, cloud or self-hosted in a UK region, so that you can leave with it. Where code is the answer, you own the repository and the hosting from day one.

We do not resell any of the three platforms and have no arrangement with any of them, which is worth stating because the recommendation an agency gives is often the one it bills a margin on. The recommendation here is the one the arithmetic gives, and the arithmetic is on this page so you can check it. When the volume is too small to matter, we say so and suggest you build it yourself on Make or Zapier in an afternoon.

Platform prices and units were read on 14 September 2026 from n8n.io/pricing, make.com/en/pricing and zapier.com/pricing, with Zapier’s task-counting rules from its help centre and n8n’s licence, hosting and edition details from its GitHub licence file and documentation. Make’s data centre and certification statements are from make.com/en/security; its slider prices for larger credit volumes were not published as text and are recorded as such. Server prices are DigitalOcean’s droplet list on the same day; n8n’s Hetzner guide names a server type but Hetzner’s price for it was not readable as text, so it is not quoted.

The per-run table assumes a five-step workflow with one trigger and four actions and shows the cheapest published tier covering each volume; a different step count changes the Zapier and Make columns in proportion and leaves the n8n column alone. Currency conversions use the Bank of England spot rates for 11 September 2026. The £6,000 migration threshold is Softomate’s and is labelled as a third-party guide figure. All prices exclude VAT. The date at the top of this page is the date of the last check.