The tool is the last decision, not the first
Almost everyone chooses badly by choosing early. They watch a video, fall in love with a tool and then go looking for what to automate with it. The right order is the reverse: first you decide what processes to automate with AI and to what result, and only then you pick the with-what. The tool is the hammer; the process is the nail. Buying the hammer before looking at the nail is how you end up with a drawer full of subscriptions nobody uses.
This guide doesn't tell you which to buy — that depends on your case — it gives you the criteria to decide it yourself and dodge the expensive mistake. There are five, they go in order, and the last one is the one almost nobody looks at. At the end you get a table to score your candidates and a link to the specific comparison that fits you.
The five criteria that actually decide
Forget the vendor's table of a hundred green boxes. The real decision fits in five questions. Answer them for your case before looking at any brand:
| Criterion | The question it answers | When it weighs most |
|---|---|---|
| Volume and flow shape | How much will it run and how complex is it? | Long flows or high volume |
| Integrations | Does it talk to what you already use without hacks? | You have a CRM/ERP/email to touch |
| Where the data lives / GDPR | Who controls the information? | Personal data or regulated sector |
| Real cost at scale | What does it cost at next year's volume? | You expect to grow |
| Who maintains it | Is there someone with a name to answer for it? | Always |
Volume, flow shape and integrations: what you'll actually use
The first two criteria tell you whether the tool works, not whether it's pretty. Volume and flow shape matter because each tool bills differently and has a different ceiling: a short flow that runs rarely fits in any of them; a long one, with loops and retries, running thousands of times a month, separates the tools fast. Before choosing, sketch your typical flow and estimate how many times it'll run. That number decides more than any review.
Integrations are the filter that eliminates candidates with no appeal. If the tool doesn't connect natively to your CRM, your ERP or your email, a custom connector awaits that someone will have to maintain — and that breaks the day the other system changes on its own. That's exactly the can of worms of integrating AI with your systems: the part that doesn't show up in the demo and where projects fall apart. Check the integrations you need before looking at the price; a cheap tool that doesn't talk to your stack is expensive.
Where the data lives and who's in control: the criterion that can veto the rest
This criterion doesn't nuance the decision: it can close it outright. There are two families of tools. Closed clouds (like Make, Zapier) host it themselves: you log in with your account, you maintain nothing and your data passes through their infrastructure. Self-hosted ones (like n8n) are built for you to host, with the data inside your perimeter. Neither is better in the abstract; what rules is your control requirement.
The short rule: if you need to be able to say "my data doesn't leave here" with GDPR in front of you, or your sector is regulated, or your security committee demands it, self-hosting stops being optional and vetoes closed clouds however convenient they are. If you don't have that requirement, self-hosting adds a server to look after in exchange for control you don't need. Decide this first: it can cross off half your candidates in one go.
The real cost and who maintains it in month six
The entry price almost always lies. They're all cheap on the first plan; the real cost shows up when volume grows, and it depends on how they charge: per operation (each step counts), per execution (one full run counts as one) or per server if you self-host. A flow that costs little on the cheap plan can balloon when it multiplies. Cost it at the volume you expect a year from now, not the demo's — it's the same discipline as calculating the ROI of automation before you sign.
And the criterion almost nobody looks at: who maintains it. Choosing a tool is half the decision; the other half is who takes care of it the day after. An automation with no owner degrades in silence no matter how good the tool was — the maintenance of automations guide covers it. The most powerful tool in nobody's hands loses to the humblest one with a clear owner. If your team has no technical profile, that fact outweighs any feature and pushes toward what can be handled without coding.
How to land the decision (and which comparison to go to)
With the five criteria answered for your case, the decision stops being a matter of taste and becomes a table. Score each candidate on the five axes, give more weight to whichever is non-negotiable for you (usually data/GDPR or integrations) and keep the one that wins your table, not the vendor's. In between there are plenty of cases that work equally well with two or three tools; there, the one your team will actually use wins.
Once you know the shape of your case, drop into the detail with the comparison that fits: if you're torn between the two big powerful no-code tools, Make vs n8n gives you the head-to-head on control, cost and technical ceiling. The logic is always the same: criteria first, brand after.