In 30 seconds
- There’s no best tool: there’s one that suits each case better.
- Zapier is the easiest to start with; Make, the most visual with conditions; n8n, the most flexible and the only one you can install on your own server.
- Each one charges for something different: tasks, operations or executions.
- Before choosing a tool, map out the process. And sometimes it’s better not to automate.
It’s one of the questions I get asked most. I work with all three, and here I explain how I compare them, with the same example built in each one so you can see the difference in practice.
What all three do
All three are automation platforms built on the same idea: when something happens in one app (the trigger), actions run in others. For example:
The difference lies in how it’s built, how complex it can get, where the data lives and how you pay.
Zapier
A list of steps from top to bottom. Good for a few steps; gets complicated with lots of branches. Cloud only. Charges per task run.
Make
A diagram with branches, very visual. Very good with conditions, filters and lists. Cloud only. Charges per operation or credit.
n8n
A node diagram, visual and technical. Lets you add code. In the cloud or installed on your server. Charges per execution in the cloud, or free if you install it yourself.
The comparison
| Criterion | Zapier | Make | n8n |
|---|---|---|---|
| Ease of getting started | The easiest | Medium | Medium or low, depending on use |
| Visual building | A list of steps from top to bottom | A diagram with branches, very visual | A node diagram, visual and technical |
| Complex processes | Good for a few steps; gets complicated with lots of branches | Very good with conditions, filters and lists | The most flexible; lets you add code |
| Self-hosting | No, cloud only | No, cloud only | Yes, on your server, as well as the cloud version |
| AI capabilities | Built-in AI steps and agents | Built-in AI modules and agents | Highly configurable AI nodes and agents |
| Pricing model | Per task run | Per operation or credit used | Per execution in the cloud, or free if you install it yourself |
The same example in all three
Imagine a shop that sells online on Shopify or WooCommerce. When a new order comes in, we want to add a row to Google Sheets and let the shop know it needs to prepare it.
In Zapier
Each order field goes into a column: number, date, customer, amount, collection shop. The “Paths” step is only needed if the alert depends on the shop. It’s quick to set up; each order uses several tasks, one per completed action.
In Make
You see the whole process drawn out at a glance, with the branches per shop. Each module run uses operations or credits.
In n8n
Similar to Make. Installed on your server, the order data doesn’t pass through a third party’s cloud, except for the apps you connect. On n8n’s cloud, each order counts as one execution.
For something this simple, all three will do. The differences show up when the process grows: more shops, more conditions, more volume or more sensitive data.
Real cases with each tool
These are cases published by each brand, so they’re picked from their best customers. Use them as a reference for what can be achieved, not as a promise.
Autotorino, 74 sites: every customer comes in through the same door
This Italian car sales and after-sales group, with 74 sites and 23 brands, replaced a scattered lead intake with 109 automations that write to Salesforce. It then built another one with AI to handle calls that used to be lost.
The takeaway: with lots of sites, the first thing you notice is centralising how each customer comes in.
Source: Zapier, Autotorino case ↗. Figures published by Zapier.
Orders, invoices and stock for an Amazon seller
Using Make, the consultancy Techflow.ai connected Amazon batch orders with shipping labels, invoices, the warehouse and the stock of a German supplier that was running everything on a spreadsheet.
The takeaway: Make fits when the process has lots of steps and branches you want to see drawn out.
Source: Make, Techflow.ai case ↗. Figures published by Make.
Importing prices into an online shop: from a day to 10 minutes
The agency iMi digital built a flow in n8n that collects its client’s ERP files via SFTP, splits them into products, customers or prices and uploads them in batches to its Shopware online shop.
The takeaway: n8n fits when the volume is large and the process needs custom logic.
Source: n8n, iMi digital case ↗. Figures published by n8n.
Checklist to decide
Tick the questions you answer yes to:
- Do you want something running this very afternoon and the process has fewer than five steps? Start with Zapier.
- Does the process have several conditions (by shop, by amount, by customer type)? Make or n8n.
- Do other people on the team need to understand and review the automation? Make is usually the easiest to read.
- Do you work with sensitive customer or employee data and prefer it not to leave your infrastructure? n8n installed on your server.
- Do you have someone to maintain a server, make backups and run updates? If not, rule out self-hosted n8n and use a cloud version.
- Are you going to move a lot of volume, hundreds or thousands of records a day? Compare what each one would charge you in your specific case.
- Do you need AI with custom logic, such as an agent that checks several sources? n8n or Make.
- Do the apps you use have a connector in the tool? Check that first.
Common mistakes
First-week review
Whichever you choose, the first week is the one that teaches you most. These are the checks I’d do every day:
- Run history: all three keep a log. Check there are no errors or empty runs.
- Data at the destination: compare three or four orders in the sheet with those in the online shop. Dates, amounts and names should match.
- Duplicates: if an order appears twice, check the trigger. It’s one of the most common faults.
- Usage: see how many tasks, operations or executions you’ve used and multiply by the weeks in the month.
- Alerts: ask the shop whether they arrive, whether they’re clear and whether they arrive in time to prepare the order.
- Access: all connections on company accounts and at least two people with access to the tool.
How to switch tools
Automations can’t be exported from one platform to another: they have to be rebuilt. In order, with no nasty surprises:
- Inventory: each automation, what it does, how often it runs and who depends on it.
- Clean-up: drop the ones no longer in use. There are usually more than it seems.
- Learning: start with a simple, non-critical one.
- In parallel: rebuild it and run it alongside the old one for a few days, writing to a test sheet.
- Switchover: when the results match, turn off the old one and switch on the new one.
- The rest: repeat with the others, leaving the most critical ones until you’ve mastered the tool.
When NOT to automate
Automating isn’t always the answer. I wouldn’t do it in these cases:
The process changes every week
Stabilise it first. Automating something undefined only moves the mess elsewhere.
It’s done rarely
A 10-minute task once a month is rarely worth the time it takes to build and maintain.
Unreliable source data
If the stock or sales have errors, the automation will multiply them.
It needs judgement in every case
A sensitive complaint or a decision about someone on the team. The automation can prepare the information, but not decide.
Nobody is going to be responsible
Every automation needs someone who receives the error alerts and knows what to do.
How I decide
Zapier
The process is simple and you’re in a hurry to start.
Make
The process has several conditions and the team needs to understand it.
n8n
There’s sensitive data, a lot of volume or custom AI, and someone to maintain it.