n8n vs Zapier vs Make vs an AI Agent: Which to Use (and What It Really Costs)

n8n vs Zapier vs Make vs a custom AI agent: current prices, a worked cost example at 2,000 runs a month, and how to choose for your volume and team.

By Innovixus Team · 2026-10-28

In short: For the same five-step workflow at 2,000 runs a month, list prices come out at about $9 on Make, €20 on n8n Cloud, $129 on Zapier, and server plus maintenance costs for self-hosted n8n or a custom script. The difference comes from how each platform counts usage. Pick on volume, team skill, where your data must live, and whether the job needs AI judgement or fixed rules.

Most small and mid-sized businesses we talk to have already tried one automation tool, and many are now paying more than they expected for it. The confusion is understandable: n8n, Zapier and Make all count usage differently, and a custom AI agent has no subscription at all, just build and running costs. This guide puts current list prices side by side, prices one real workflow on each, and covers the reliability problems that cost money whichever tool you choose.

The short answer

Current pricing at a glance

Prices below were checked on 28 October 2026 on each vendor's official pages. Zapier and Make list prices in US dollars and n8n lists prices in euros. Zapier says it can bill in local currencies at the exchange rate on the payment date.

Make's own comparison page lists the $9, $16 and $29 prices but does not say whether they are monthly or annual billing, and its pricing page says annual payment is cheaper. Check Make's pricing page for the figure that applies to you. n8n's pricing page shows annual prices only; check current pricing if you want to pay monthly.

Worked example: one workflow, 2,000 runs a month

Take a common lead-handling workflow: a website form submission triggers it, then four actions run. It looks up the contact in your CRM, creates or updates the record, posts an alert to Slack and adds a row to a Google Sheet. At 2,000 leads a month:

On list prices, the same workflow costs $9 on Make and $129 on Zapier. That gap comes almost entirely from the counting method. Zapier charges per action, so every extra step multiplies the bill. Make charges per module, but at a much lower price per unit. n8n charges once per run, so adding steps costs nothing extra.

The Make figure counts the trigger as a credit too, so it is an upper bound. Past 10,000 credits you move to a higher tier.

Volume is not the only cost. Zapier is designed so a non-technical person can build and fix a workflow, and that saves money if the alternative is paying a developer each time something changes.

What self-hosted n8n really costs

The licence

n8n's Community edition is published under the Sustainable Use License, which allows use "only for your own internal business purposes or for non-commercial or personal use". Automating your own company's work is covered. If you plan to host n8n for paying clients or build it into a product you sell, read the licence carefully and talk to n8n about a commercial licence first. Features in files marked .ee need an n8n Enterprise licence.

The paid Business plan is self-hosted only, and its licence key checks in with n8n's licence server daily. Plan for that if your server sits on a restricted network.

The server

A small cloud server is enough to start. As an example, a DigitalOcean Basic droplet with 2 GiB of RAM is $12 a month, and 4 GiB is $24. Add backups for the database, and a separate managed database if the workflows matter to the business.

The maintenance

This is the cost most teams leave out. In October 2026 n8n published two batches of security advisories, on 14 October and 28 October, following another batch on 30 September. The 28 October batch included a high-severity issue where deprecated Code nodes allow code execution in the n8n main process, fixed in versions 2.42.6, 2.43.3 and 3.0.1. On n8n Cloud, the vendor handles patching. Self-hosted, someone on your team has to read the advisories and upgrade promptly.

Major upgrades need planning too. n8n published 3.0.0 as a pre-release on 27 October, followed by 3.0.1 on 28 October, while 2.45.5 remained the stable release. According to the 3.0 breaking changes page, 3.0 no longer supports installs run through npm, so those setups need to move to Docker first. The page also says to run the Migration Report under Settings before upgrading. Keep production on the stable release and test 3.0 on a copy.

Budget a few hours a month for updates, backup checks and failed-run reviews. If nobody on the team can do that, n8n Cloud or a managed service is the cheaper option in practice.

When an AI agent or custom script makes sense

The tools above follow fixed rules well. They are less suited to steps that need judgement, such as reading a free-text customer email, deciding whether it is a complaint, a quote request or spam, and drafting a reply.

A custom agent built with a developer (or a coding agent supervised by one) has no subscription, but you pay for the build, hosting and model usage. As a rough guide, if each lead needs about 3,000 input tokens and 500 output tokens on Claude Haiku 5.5, priced at $0.10 and $0.50 per million tokens, that is about $0.00055 per lead, or roughly $1.10 a month at 2,000 leads. A larger model can cost ten to forty times more for the same work, as our AI agent cost guide shows. Your token counts will differ, so measure them on real data before you budget.

An AI step inside Zapier, Make or n8n still incurs model costs; on Make, AI modules can use a variable number of credits. A common pattern is a hybrid: n8n or Make handles triggers and integrations, and calls an agent only for the step that needs reasoning. Our AI software development team builds that kind of agent when off-the-shelf steps are not enough.

Which one should you pick?

By volume. Under a few hundred runs a month, any tool's free or entry plan will do, so choose on ease of use. In the thousands, multi-step workflows get expensive on per-action pricing; price your real step count on each tool. At tens of thousands of runs, n8n's per-execution model or a custom service is often cheaper to run, provided someone can maintain it.

By team skill. If nobody on the team writes code, start with Zapier or Make. If you have a developer or a technical operations person, n8n gives more control, including code steps and self-hosting. A custom agent needs ongoing developer time.

By data residency and privacy. If customer data must stay on servers you control, for example because of client contracts in India or the Gulf, self-hosted n8n or a custom service is the straightforward route. For cloud plans, read each vendor's data-processing terms and hosting regions before sending personal data through them.

By the kind of decision. If the rule can be written as "if X, then do Y", use a standard workflow. If a person currently reads something and makes a call, that step is a candidate for an agent, with human review at first.

If you have not picked the first process to automate yet, start with our list of processes to automate first.

Production reliability, whichever tool you use

When automation builders compare notes, the same problems keep coming up: a retry loop that runs silently and runs up an API bill, duplicate webhooks creating duplicate records, an agent updating the wrong customer, and the fiddly job of connecting a client's WhatsApp number. None of these is solved by choosing a different tool. They are solved by how you build.

• Deduplicate webhook retries. Senders retry when they do not get a quick response, so the same event can arrive twice. Store each event's ID and skip any you have already processed.
• Alert on runs that "succeed" but do nothing. A run that finishes with zero records written shows as green. Check the output count and alert when it is zero for a workflow that should always write something, or when daily totals drop sharply.
• Pass a correlation ID through every sub-workflow. Generate one ID at the trigger and carry it into every sub-workflow, log line and record you touch. When something goes wrong, you can trace one lead end to end.
• Cap spend per agent. Set a monthly limit with your model provider, and a maximum number of steps or retries per run inside the workflow, so one loop cannot run unchecked.
• Require human approval before writes to the CRM. Have the agent propose the change and a person approve it, at least until you trust its accuracy. Match records on a unique ID such as email or customer number, never on a name.
• Treat WhatsApp as its own onboarding step. Make sure the business account and number are registered to the client, not to you or the automation tool, so access survives a change of supplier.

Moving off Pipedream Workflows

Pipedream has announced that Workflows will shut down on 31 March 2027, and that all Workflows data, including connected-account credentials, will be deleted by 30 April 2027. Pipedream Connect is not affected.

Pipedream provides a JSON export per project, limited to 500 workflows or 256 MB per export. Secrets, credentials and environment variables are redacted or excluded. Neither Pipedream's export documentation nor n8n's import documentation describes importing one into the other, so plan to rebuild, using the export as your specification. Re-create credentials, update every service that sends webhooks to a Pipedream URL, and run old and new side by side before switching off.

Where to go from here

If you want help choosing a tool or pricing a workflow at your real volume, see our smart automation services or get in touch with a description of the process. We will tell you which option fits, including when the answer is the cheapest one.

Frequently asked questions

Is n8n cheaper than Zapier?

Usually, once a workflow has several steps. n8n Cloud counts one execution per workflow run, while Zapier counts a task for each successful action. In our worked example of 2,000 runs a month, n8n Cloud Starter was €20 a month and Zapier Pro was $129 a month, both billed annually.

Is self-hosted n8n free?

The Community edition has no licence fee for internal business use under n8n's Sustainable Use License. You still pay for a server and backups, and you need someone to apply security updates. n8n published two batches of security advisories in October 2026 alone.

When should I use an AI agent instead of n8n, Zapier or Make?

When a step needs judgement, such as reading a free-text email and deciding what to do with it. For fixed rules, a standard workflow is cheaper and easier to debug. Many teams use both, with the agent as one step inside a workflow.

Can I import Pipedream workflows into n8n?

Not directly, as far as either vendor's documentation says. Pipedream's export is a JSON record of each workflow with secrets redacted, and n8n imports its own workflow JSON. Treat the export as a specification and rebuild before Pipedream Workflows shuts down on 31 March 2027.

Do these prices include AI model costs?

No. If a workflow calls a model such as Claude or GPT through your own API key, the model provider bills you separately for tokens, whichever automation tool you use.