n8n vs Zapier vs Make: how I choose for a small business

Clients often ask which automation tool they should use, usually after someone on their team has already built three Zaps and hit a pricing wall. There is no single right answer, but there is a reliable way to decide. These are the questions I ask, in order.

The short version

Pricing models change, so check each vendor’s current page — but the shape of the pricing is what matters.

Question 1: Who will maintain it?

If the person maintaining the workflows is an operations manager who wants to tweak a step on a Friday afternoon, Zapier is hard to beat. The interface is the most forgiving, the app directory is the largest, and almost every SaaS tool has an official integration.

If I will maintain it, or the client has someone technical in-house, the other two open up.

Question 2: How many runs per month?

This is where most small businesses get surprised. Zapier bills per task — roughly each action step that runs. A workflow with six steps that fires 2,000 times a month is 12,000 tasks. Make bills per operation (now counted in credits) in a similar way.

n8n counts a whole workflow run as one execution on its cloud plans, and costs nothing per run when self-hosted on a small server. For a CA firm sending reminders, syncing documents and replying on WhatsApp all day, that difference is often the entire business case.

Question 3: Where does the data have to live?

Client documents, PAN and GST numbers, salary data, medical records — some data you simply don’t want passing through a third-party automation cloud. Self-hosted n8n runs on your own server (or a server in your own cloud account), so the data stays where you control it.

Neither Zapier nor Make can be self-hosted.

Question 4: How much logic is involved?

Real business workflows are rarely “when X, do Y.” They are “when a document arrives, figure out which client it belongs to, check whether it’s the one we asked for, update the tracker, and if it’s the last missing document, notify the partner.”

• Make handles branching visually and well.
• n8n handles it visually and lets you drop into a code node when visual logic gets awkward, call any API directly, and build sub-workflows you can reuse.
• Zapier can do branching with Paths, but complex logic gets expensive and hard to read.

Question 5: Are there AI steps?

All three can call an LLM now. The difference is control. In n8n I can run an AI agent node with tools, keep conversation memory, choose the model provider (OpenAI, Groq, a local model), and log every prompt and response for review. For WhatsApp assistants and document classification, that control matters. I wrote more about this in WhatsApp AI chatbots that answer from your own documents.

My default

• Under ~10 simple workflows, team-maintained: Zapier.
• Visual workflows with moderate volume: Make.
• High volume, private data, real logic or AI: self-hosted n8n — with documentation and error alerts so you’re never dependent on one person to know what’s running.

The tool matters less than the mapping work before it. Most automation projects I take on start by drawing the current process with the people who do it, and finding that two of the five steps don’t need to exist at all. You can see one of those projects in the CA firm case study, or tell me about your workflow.