For most UK small businesses, Make is the best starting point, Zapier is the easiest if the bill does not matter yet, and n8n is worth it only when someone on your side can run a server.
The price on the pricing page matters less than the unit each tool bills in, who fixes a broken workflow at 8am on a Monday, and where your customers' data ends up.
This comparison covers all three. We use Make.com ourselves, so we say so now and show the working for the rest, so you can check our reasoning rather than take our word for it.
How does each tool charge you?
Zapier charges per task, Make charges per operation and n8n charges per execution, and the difference changes your bill more than the headline plan price does.
A Zapier task is one successful action step. Triggers, filters and built-in formatting steps do not count. A Make operation is any module that runs, and that includes the trigger and, according to one independent comparison, polling checks that find nothing. An n8n execution is one complete run of a workflow, however many steps it has.
Take one ordinary workflow: a website enquiry arrives, the contact is checked in the CRM, a record is created, an email goes to the customer and an alert goes to the sales channel. That is five steps. Run it 500 times in a month.
| Tool | What counts | Usage for 500 runs |
|---|---|---|
| Zapier | Four action steps, trigger free | About 2,000 tasks |
| Make | Five modules including the trigger | About 2,500 operations |
| n8n | One execution per run | 500 executions |
Now set that against the plans. The comparison guides we checked in October 2026 list Zapier entry plans of 750 and 2,000 tasks a month, Make's entry plan at 10,000 operations, and n8n's cloud starter at 2,500 executions.
The same workflow uses most of a mid-tier Zapier plan, a quarter of Make's entry plan and a fifth of n8n's.
Prices are quoted in dollars or euros and change often, so check the vendor's page before you commit. The shape of the maths stays the same: the more steps a workflow has, the faster Zapier's bill grows, and the more volume you have, the more n8n's per-run model helps.
Which is easiest for a non-technical team to build in?
Zapier is the easiest, Make is a step up, and n8n is built for people who are comfortable with code.
Zapier has the largest catalogue of connected apps, with figures in the comparison guides ranging from about 6,000 to more than 8,000 depending on the date, against roughly 1,500 for Make and 1,000 for n8n.
For a small business using a niche booking or accounting system, that gap decides things. If your software is not on the list, you are writing custom API calls whichever tool you pick.
Make shows each workflow as a visual map, which makes branching and looping easier to see. It takes longer to learn than Zapier, but the comparison guides we read agree it handles data transformation better, so complex workflows need fewer workarounds.
n8n gives you full JavaScript and Python in a node, and is the usual choice for building AI workflows with several model calls. It rewards technical people and frustrates everyone else.
All three now offer steps that call a language model, so the AI part of a workflow is no longer a reason to choose one over the others.
The comparison guides we read give n8n the edge for workflows with several model calls chained together, while Zapier and Make make single-step uses, such as classifying an email or drafting a reply, easy to add. Price that in too, since model calls are billed separately by the provider.
What happens when something breaks?
All three will fail eventually, and the difference is how much each one helps you find out why.
Zapier retries a failed step automatically, up to three times, and emails you. It has no per-step error routing, so you cannot tell a workflow to take a different path if a step fails.
Make lets you attach an error route to each module: retry, ignore, roll back or commit what has been done so far. That matters for anything involving money or customer records.
n8n has error workflows that you set up yourself. On the self-hosted version, security patches and backups are your job. One of the guides we read notes that the self-hosted edition does not update itself, which is the sort of detail that surfaces eighteen months after the person who installed it has left.
Execution history is also worth checking, because it is how you investigate a complaint. History on cloud plans ranges from a week to a month or more depending on tier, so a customer who says "I never got the email" three weeks later may be asking about a run you can no longer see.
Who maintains it once it is built?
Whoever you name in writing at the start, and the cheapest tool is a false saving if the answer is "nobody".
This is the question most comparisons skip. A workflow tool is closer to a small piece of software than a subscription. It has credentials that expire, APIs that change and staff who need adding to it.
With Zapier or Make, you can ask a person in your team to own it, or pay an agency a retainer to monitor it. With self-hosted n8n you are also running a server, applying updates and testing restores.
If you have a developer on staff, that may be fine. If you do not, the saving on subscription fees is likely to be spent on repair bills.
Whichever you choose, make sure the account sits in your company name, with your email as owner, and that your agency is a team member, not the account holder. We cover this in more detail in what to ask any AI agency before you hire.
Where does your customers' data go?
Every workflow tool that handles customer details is a data processor under UK GDPR, which means you need a written contract with it.
The ICO says a controller-processor contract must cover the subject matter and duration of the processing, the types of personal data, and the processor's duties on security, sub-processors, breach assistance and deleting or returning data at the end. You can read the full list in the ICO guidance on what the contract must include.
For the hosted tools, ask for the vendor's data processing agreement and check where it runs. For self-hosted n8n, the processing happens on your own server, which can make the data question simpler, as long as you secure that server.
n8n's code is published under a "fair-code" licence called the Sustainable Use License, according to Wikipedia's n8n entry. It lets you self-host and modify the software for your own business use, and it bars you from reselling it as a competing hosted service. For an ordinary company running its own workflows, that is rarely a problem.
How hard is it to switch later?
Harder than the vendors suggest, because none of the three lets you export a workflow and import it into another.
Each tool stores your logic in its own format. Moving from Zapier to Make means rebuilding every workflow by hand, using the old one as a specification. For ten simple workflows that is a weekend. For sixty, with branches and error handling, it is a project.
That is a reason to write each workflow down in plain English before it is built: what triggers it, what it does at each step, what it does when a step fails and who gets told. The document is worth having whichever tool you use, and it turns a future migration from guesswork into a checklist.
It is also worth clearing out dead automations before you move. One of the guides we read recommends deleting unused "zombie" workflows first, and in our experience a business that has run a tool for two years has several that nobody remembers building.
Which one should you pick?
Pick by volume, by who will maintain it and by whether your apps are supported, in that order.
flowchart TD
A[Start] --> B{Is everything you use on Zapier's app list, and is volume low?}
B -- Yes --> C[Zapier]
B -- No --> D{Do you have someone who can run a server?}
D -- Yes --> E{High volume or heavy AI steps?}
E -- Yes --> F[Self-hosted n8n]
E -- No --> G[Make]
D -- No --> G
For most firms, the path ends at Make. That is why we build on it, alongside custom code where a workflow needs something the tool cannot do. We also move clients between tools when the numbers say they should.
Online retailers see this quickest. Ecommerce brands and retailers generate order, stock and customer events by the thousand, so a per-task price that looked sensible at 200 runs becomes a line item at 20,000.
Before you build anything, work out what is worth automating. Our list of five automations that save the most hours is a sensible starting point. And if you are weighing a subscription against a custom build, build vs buy for AI tools covers when each wins.
If you would like a second opinion on which tool suits your workflow and volume, see our AI automation services or contact us with a description of the process.
