n8n vs Zapier vs Make: Which One Go Save You Money?

The one difference wey decide this

n8n, Zapier and Make dey do the same work. Trigger go fire, data go move between APIs, and something go land for another system. The choice between dem usually depend on one thing: the billing unit. Zapier and Make charge for every unit of work wey your automations perform. Self-hosted n8n no charge anything per unit of work, because na server you dey pay for instead.

Na this one difference dey drive everything wey follow. Metered plan fit cost almost nothing when you run ten workflows per day, but e fit become your biggest software cost when workflow start firing every minute. Flat server cost dey do the opposite: e fit look expensive on day one, but e no dey increase as volume grow. No be any of the two models dey deceive person. Dem just bill different users better.

Wetín each product dey actually charge for

Zapier dey bill per task. Zapier own definition be say task dey count anytime Zapier successfully complete one unit of work. Each successful action step for one Zap dey use one task, and some actions dey use more than one. Triggers and polling no dey count, failed actions no dey count, and the built-in tools (Formatter, Paths, Filter, Delay, Looping, Storage) no dey count. So, if one Zap get one trigger and three action steps, e go cost three tasks every time e run. As of July 2026, the free tier get 100 tasks per month, and the paid Professional plan start around 20 US dollars per month when dem bill yearly, for 750 tasks.

Make dey bill per credit, wey be the unit wey e dey call operation before. Make documentation talk say each action wey one scenario perform dey use a particular number of credits, and most actions dey use one. Reading, searching, creating, updating, deleting, transforming, aggregating and iterating each cost one credit. Routers and error handlers no dey. The important word na iterating: one module wey dey loop through 200 rows go spend roughly 200 credits, no be one. As of July 2026, the free tier get 1,000 credits per month, and the Core plan start around 12 US dollars per month for 10,000 credits.

Self-hosted n8n no dey bill anything per execution. The Community edition free and e carry almost the complete feature set. Workflow runs, nodes inside one run, loop iterations and retries all cost zero, because na only your server CPU, memory and disk dey act as meter. Paid self-hosted tiers dey exist and dem need license key, but dem dey sell governance features like single sign-on and separate environments, no be throughput. If you don already set up n8n for VPS with Docker and HTTPS, you don already pay the complete recurring bill.

The install command from n8n own docs short reach to show the point:

docker volume create n8n_data

docker run -it --rm \
 --name n8n \
 -p 5678:5678 \
 -e GENERIC_TIMEZONE="Europe/Berlin" \
 -e TZ="Europe/Berlin" \
 -e N8N_ENFORCE_SETTINGS_FILE_PERMISSIONS=true \
 -e N8N_RUNNERS_ENABLED=true \
 -v n8n_data:/home/node/.n8n \
 docker.n8n.io/n8nio/n8n

The editor go then answer for port 5678. That command na only for first look, because e dey use the default SQLite file and --rm go throw the container away when you stop am. Server wey you rely on need Postgres and reverse proxy for front.

A realistic month, wey you count

Vague comparisons no useful, because the answer depend completely on your volume. So count am. Na four automations be these wey small company really dey run, together with how often each one dey fire and how many billable steps each run dey perform.

[
  {
    "label": "Form to CRM to Slack",
    "runs_per_month": 1200,
    "billable_steps": 3,
    "units_per_month": "3,600"
  },
  {
    "label": "Order sync, every 15 min",
    "runs_per_month": 2880,
    "billable_steps": 4,
    "units_per_month": "11,520"
  },
  {
    "label": "AI inbox triage",
    "runs_per_month": 900,
    "billable_steps": 5,
    "units_per_month": "4,500"
  },
  {
    "label": "Daily revenue report",
    "runs_per_month": 30,
    "billable_steps": 6,
    "units_per_month": "180"
  }
]

The order sync alone dey spend 11,520 billable units every month, because schedule trigger wey dey run every 15 minutes fires 2880 times and each run dey touch 4 paid steps. The daily report, wey feel like the heavy workflow because e get the most steps, dey spend 180. Frequency dey beat complexity every time. Na why estimating from the shape of workflow instead of the schedule dey make people receive invoice wey dem no expect.

Together, those four automations dey spend 19,800 billable units every month. For Zapier, that one put you for 20,000-task tier of paid plan. For Make, na roughly twice the credits wey Core plan include, so you go need buy higher plan. For self-hosted n8n, na the same monthly server bill as running one workflow, and the server no go notice.

Now change one number. Move the order sync from every 15 minutes to every 5 minutes. Zapier and Make go triple that line to 34,560 units, and the invoice go move to the next tier. The n8n box go add only few percent CPU. That difference, instead of the headline price of any plan, na the real decision.

Where crossover dey happen

No universal break-even point dey, but the pattern dey predictable.

• Below roughly 1,000 units per month, hosted free tiers win completely. Nothing beat zero cost, and you no get server to patch.
• Between roughly 1,000 and 10,000 units, entry paid plans cheap, and hosted tools still better value, because VPS plus your time cost pass 12 to 20 dollars per month.
• Above roughly 20,000 units per month, metered billing dey add up, and fixed server cost usually win. E go continue to win as volume dey grow.
• Any workflow wey dey process rows one by one (a CSV import, a list of 500 orders, a paginated API) dey skip these ranges completely, because one run fit use thousands of units.

Na this last case make self-hosting stop to be just preference. A nightly job wey process 5,000 records go use 5,000 credits for metered plan, but e go use only CPU for one execution on n8n.

The same logic apply to AI steps, and na there automation volume dey grow fastest. Every model call inside hosted scenario na billable unit on top of the model provider own token charge, so dem dey meter you twice for the same work. If you run the Claude API inside n8n workflows, you go pay only the model provider.

The n8n license, plain

n8n no be open source under the usual definition, and to pretend say e be na how people dey get surprise later. E dey ship under Sustainable Use License, wey talk say:

You fit use or modify the software only for your own internal business purposes or for non-commercial or personal use.

And about passing am give other people:

You fit distribute the software or provide am to other people only if you do am free of charge for non-commercial purposes.

Compare this with wetin you plan to do. If you run n8n to automate work for your own company, na internal business purposes be that, and e dey permitted. This cover almost everybody wey dey read this. If you run n8n as the engine behind product wey you dey sell, or you resell hosted n8n instances to customers, the free license no cover am. You go need commercial agreement with n8n. Check the current LICENSE.md for the n8n repository before you build business on top of am, because terms fit change and na only the text for the repository count.

Another milder issue dey. The free Community edition get almost every feature, and free registration unlock few more, like folders and debug inside editor. Features wey target bigger teams, including single sign-on, log streaming, and separate environments, dey behind paid license key. Those ones na governance features. Dem no dey sell throughput at all. If you no sure which side of that line the feature wey you depend on dey, e good make you check the community and enterprise limits feature by feature before you design around one.

Wetin the flat server cost no cover

The honest version of the self-hosting argument include the parts wey no free.

You own uptime. When hosted platform get outage, you go wait and person for the other side go receive alert. When your VPS get outage, na you go fix am, any hour wey e happen. You own upgrades, because n8n dey release often and container image wey you never update go become security problem. You own backups: the n8n encryption key and the Postgres database together na the only copy of your credentials and workflows, and if you lose the key, every stored credential go become unreadable. Point encrypted restic backups to both from the first day, no be when first incident happen.

You still own the parts wey hosted plan quietly include. Zapier and Make maintain thousands of app connectors and repair dem when vendor change API. n8n get large node library and generic HTTP node wey cover the rest, but when integration break, na your afternoon e go take. Hosted plans also include compliance paperwork, support with response time, and another person’s on-call rotation.

Budget the real number honestly: the server, plus about one hour maintenance every month, plus the day wey you go spend for initial setup. If that total still beat the metered invoice, self-hosting na the right choice. For 19,800 units every month, e usually do. For 500 units every month, e usually no do.

Which one you suppose pick

Pick Zapier when na connector list be the main product. E get the widest app coverage and e easy to learn, and if your automations short and no dey happen often, per-task billing truly cheap.

Pick Make when your logic dey visual and dey branch. Its canvas handle routers, iterators and error handlers well, and its credits cost less per unit than Zapier tasks for similar tiers. Watch the iterators, because na there credit usage fit grow without warning.

Pick self-hosted n8n when volume high, when data no fit leave your own infrastructure, or when workflows dey call models and databases wey you already run. Na only this one among the three wey let you read the code path wey workflow dey follow and change am. Budget one VPS and one hour every month, and read the license first. E deserve place for any 2026 shortlist for self-hosting because of the reason above: the meter stops.

FAQ

Self-hosted n8n really free na?

Community edition free to run, and e no get charge per execution with almost the complete feature set. You still go pay for the server. Features wey target bigger teams, like single sign-on, log streaming, and separate environments, need paid license key. Dem no dey meter your workflow volume, and na the main reason people dey self-host.

How many Zapier tasks one Zap dey use for each run?

Na one task for each successful action step, but some actions fit cost more than one. Trigger free, polling free, failed actions free, and built-in tools like Formatter, Paths, Filter, and Delay free. So, Zap wey get one trigger and three actions go cost three tasks per run. If e fire 1,000 times for one month, na 3,000 tasks e go use.

Make credits and Zapier tasks dey count the same way?

Dem dey similar, but dem no exactly the same. Both dey charge for each action wey workflow perform. Make dey count each module action as one credit, including transformations and every iterator pass. Zapier no dey charge for its built-in formatting and filtering tools. The difference matter pass for workflows wey loop through lists, because Make per-iteration counting dey add up quickly.

I fit use self-hosted n8n inside commercial product?

No, not under the free license. Sustainable Use License allow you use am for your own internal business purposes, or for non-commercial or personal use. Automating work for your own company dey okay. If you embed n8n inside product wey you dey sell, or resell hosted instances, you need commercial agreement with n8n. Read the current LICENSE.md for the repository before you commit to any design.

For what volume I suppose switch from Zapier or Make to n8n?

No fixed number dey, but the general pattern still hold. Below about 1,000 billable units per month, hosted free tiers dey win. Above about 20,000 per month, fixed server cost usually dey win. Any workflow wey dey iterate over rows fit break this rule immediately. One run over 5,000 records go spend 5,000 units for metered plan, but e no go spend anything extra for your own server.