Use Zapier if you want AI steps working this week with the least setup and your volumes are modest. Choose Make when workflows branch and volumes grow, because it's far cheaper per step. Pick n8n when runs have many steps, such as AI agents, or you need self-hosting, and someone technical can maintain it.
The deciding number is steps per run, because each platform charges for a different unit. Zapier bills every successful action as a task, Make bills every module that runs as a credit, and n8n bills a whole workflow run as one execution however many steps it contains. A five-step AI workflow therefore costs roughly four tasks, five credits or one execution each time it runs, and the gap widens with every step you add. Headline plan prices hide this completely.
The pricing unit is the real difference
Here's how the three bill, what their entry prices buy, and what happens when you run out. All three have changed their pricing in the past year, so these are list prices as of September 2026.
| Zapier | Make | n8n (cloud) | |
|---|---|---|---|
| Billed per | Task: each successful action step | Credit: each module that runs | Execution: one whole workflow run |
| Free | Triggers, filters, Paths and Formatter steps; Free plan gives 100 tasks a month, two-step Zaps only | Routers and filtered-out bundles; Free plan gives 1,000 credits a month and two active scenarios | Self-hosted Community Edition (you run the server) |
| Entry paid price | Professional, $29.99 a month monthly or $19.99 annually, for 750 tasks | From $9 a month for 5,000 credits; annual saves about 15% | Starter, EUR 24 monthly or EUR 20 annually, for 2,500 executions |
| Speed of triggers | Polls every 2 minutes on Professional, 1 minute on Team; webhooks are instant but not on Free | 1-minute minimum interval on paid plans (15 on Free); webhooks are instant | Webhook or schedule triggers, set per workflow |
| At the limit | Runs are held for replay, or pay-per-task continues at 1.25x (annual) or 2.5x (monthly) the base rate, up to 3x your plan's tasks | Scenarios stop unless you buy extra credits, which carry a 25% premium; auto-purchase buys 10,000 at a time | Executions fail immediately with an error; nothing queues, and limits reset at the start of each calendar month |
That last row matters more than most comparisons admit. On Zapier's pricing, running out usually costs money but loses nothing, because held runs can be replayed. On n8n cloud, a workflow that hits the ceiling mid-month simply fails until the 1st, so an enquiry form wired to it goes quiet without anyone noticing.
How each one handles the AI step itself
All three can call a large language model in the middle of a workflow. They differ in who you pay for it and how it's counted.
- Zapier offers AI by Zapier steps on Professional, Team and Enterprise (not Free). A run uses 1, 3 or 5 tasks depending on the model tier you pick, or one task if you connect your own API key. Zapier also sells Agents, Chatbots and an MCP server; each successful tool call through Zapier MCP uses two tasks.
- Make gives you two routes. Third-party modules for OpenAI, Claude or Gemini use one credit per operation, and you pay the AI provider for tokens on your own account. Make's own AI Provider, used by its AI Toolkit and AI Agents, charges a credit per operation plus credits based on tokens. After a repricing on 25 August 2026, Make's example extraction step (2,000 input tokens, 100 output) costs about 0.16 credits in tokens on the Medium tier.
- n8n builds AI in through its AI Agent and model nodes, normally with your own API keys. The AI step sits inside the same single execution as everything else, and n8n can pause an agent before a tool call for a person to approve it by Slack, Gmail or Teams.
With your own key, the model's own bill is small for short jobs. A classification-and-draft step using about 1,500 input tokens and 300 output tokens costs roughly $0.0003 a run on gpt-6-luna ($0.10 and $0.50 per million tokens) or about $0.003 on Claude Haiku 4.5 ($1 and $5). Even at 2,500 runs a month, that's under $1 or about $7.50. The platform's counting method, not the model, usually decides the bill. If you want to see what an AI step actually does inside a Zap, adding AI steps to Zapier to classify, summarise and draft walks through one.
One enquiry workflow priced at 100, 600 and 2,500 runs
Take an illustrative yoga studio's enquiry flow. A website form submission arrives; an AI step classifies it (intro offer, private session, teacher training, corporate class or complaint) and drafts a reply; the enquiry is added as a row in Google Sheets; a Gmail draft is created for the owner to check; and a Slack message alerts whoever is on the desk. A filter drops spam before the AI step.
Counted per run, with the AI step on your own API key and an instant (webhook) trigger:
- Zapier: trigger and filter free; AI step, Sheets row, Gmail draft and Slack message are 4 tasks.
- Make: trigger, AI module, Sheets, Gmail and Slack modules are 5 credits.
- n8n: 1 execution.
Now the three volumes, using monthly-billing list prices unless stated. A quiet summer month might bring 100 enquiries, a January rush 600, and a studio group with three sites 2,500.
| Runs a month | Zapier (4 tasks a run) | Make (5 credits a run) | n8n cloud (1 execution a run) |
|---|---|---|---|
| 100 | 400 tasks: Professional 750 tier, $29.99 ($19.99 annually) | 500 credits: entry plan at $9 (the Free plan covers it on paper but allows only two active scenarios) | 100 executions: Starter, EUR 24 (EUR 20 annually) |
| 600 | 2,400 tasks: 5,000 tier at $133.50 ($89 annually), or the 2,000 tier plus pay-per-task for about $110 ($61 annually) | 3,000 credits: entry plan at $9 | 600 executions: Starter, EUR 24 |
| 2,500 | 10,000 tasks: 10,000 tier at $193.50 ($129 annually), with no headroom | 12,500 credits: entry plan plus about $18 of extra credits, roughly $27; a larger tier may be cheaper | 2,500 executions: Starter's exact limit, so Pro at EUR 60 (EUR 50 annually) for headroom |
A few notes on the sums. The Zapier pay-per-task figure assumes the overage rate is a multiple of that tier's per-task price ($73.50 for 2,000 tasks, so about $0.037 a task, times 2.5 for 400 extra tasks). The Make extra-credit figure assumes your subscription's rate matches the entry plan's $9 per 5,000 credits, plus the 25% premium, which gives about $2.25 per 1,000; your subscription tab shows the real rate. Add the model's own token bill to every column if you use your own key, and see when Zapier gets too expensive for how to find your own tipping point.
If the AI step runs on one of Zapier's 3x model tiers instead of your own key, each run becomes 6 tasks: 600, 3,600 and 15,000 tasks, which lands on the 750, 5,000 and 20,000 tiers ($29.99, $133.50 and $283.50 monthly). Model choice inside Zapier can move you up a whole tier.
Decision table: nine things that should decide it
| Criterion | Zapier | Make | n8n |
|---|---|---|---|
| Owner with no technical help | Easiest to start; linear Zaps read like a checklist | Visual canvas; a learning curve with routers and data mapping | Hardest; cloud is manageable, self-hosting is not a DIY job |
| Cost at five or more steps per run | Highest; every action counts | Low; one credit per module | Lowest; one execution per run |
| Behaviour at the limit | Runs held or billed; nothing lost | Scenarios stop until credits are added | Executions fail until next month or an upgrade |
| Branching and loops | Paths (free steps), but deep logic gets messy | Routers cost nothing and suit branching well | Built for complex logic and code steps |
| AI agents with several tool calls | Each action is a task | Credits per operation, tokens and tool calls | One execution per run |
| Human approval before sending | Human-in-the-Loop on Professional sends approvals only to yourself | The Human in the Loop app is Enterprise-only (closed beta) | Approve or reject by Slack, Gmail or Teams, including before an agent's tool call |
| Error handling | Error handler steps; a Zap pauses itself if 95% of runs error over 7 days | Named error handlers: Skip, Retry, Resume, Commit, Rollback | Error workflows; failed executions don't count against the quota |
| Data kept on your own server | No | No | Yes, if self-hosted |
| How the price list works | Dollar tiers of tasks, from 750 to 2 million a month | Dollar credit slider, from 5,000 credits | Euro tiers of executions, plus a free self-hosted edition |
Where each platform wins outright
Zapier: the owner who needs it running by Friday
A personal trainer who wants new enquiry forms classified, logged and answered in draft, at 60 runs a month, will be done in an afternoon on Zapier and will pay $29.99 a month. The per-task premium doesn't matter at that volume, held runs protect against a busy month, and the linear layout means the next person can read it. If the business also leans on apps with less common integrations, check each one has the trigger you need on all three platforms before deciding; that alone sometimes settles it.
Make: branching workflows at volume, on a budget
A wedding planner whose enquiry form feeds five different routes (venue enquiries, full planning, on-the-day coordination, supplier pitches, spam) is a natural fit for Make's router, which costs nothing to run. At several hundred enquiries a month the $9 entry plan still covers it. Make is also the easiest of the three for seeing, visually, which branch a run took when something goes wrong, and building a first AI automation in Make shows the canvas in practice.
n8n: many steps per run, or data you'd rather host yourself
A multi-site studio group running an AI agent that looks up a member, checks class availability, drafts a reply and logs the outcome might make ten or more tool calls per enquiry. n8n charges that as one execution. It's also the only one of the three you can self-host, which matters if a client contract says data must stay on infrastructure you control. What n8n is, and whether a small business should use it covers the trade-offs, and n8n self-hosted versus cloud costs puts numbers on the server route.
A yoga studio's choice, reasoned through
Back to the illustrative studio. The owner runs everything herself with a part-time desk assistant; nobody on the team writes code. Volume swings from about 150 enquiries in summer to 600 in January. The studio uses a free personal Gmail account for bookings.
On price alone, Make wins easily: $9 a month covers the January peak, where Zapier's equivalent is $133.50 on monthly billing or $89 annually. n8n cloud is also cheap at EUR 24, but its hard stop at the execution limit makes a January spike risky on the Starter tier, and the owner doesn't want to manage it.
Then the snag. Make's help pages say accounts on a personal @gmail.com address need their own Google Cloud OAuth client to connect Gmail, Drive or Sheets. That's a one-off setup of perhaps an hour for someone comfortable in Google Cloud, and a wall for someone who isn't. The studio had two sensible paths:
- Pay someone once to set up the OAuth client and build the scenario in Make, then run at $9 a month.
- Stay on Zapier's 2,000-task tier on annual billing ($49 a month), trimming a step (the Slack alert became part of the Gmail draft) so each run uses 3 tasks. January's 600 runs then need 1,800 tasks, and pay-per-task covers any spike beyond that.
She chose the second, because it could be built and fixed by her own desk assistant. That's the pattern worth copying: the right platform is the cheapest one your team can maintain, not the cheapest one on paper.
Traps that make the cheap option expensive
- Polling triggers on Make. Each scheduled trigger check uses a credit even when nothing arrives. Checking every 15 minutes burns about 2,880 credits a month before a single enquiry, which is over half the $9 plan. Use instant (webhook) triggers wherever the app offers them.
- Sizing n8n Starter to your exact volume. At the limit, executions fail immediately rather than queue. Leave at least 30% headroom, or choose Pro.
- Zapier's model tier multiplier. A 5x model tier turns a 4-task run into 8 tasks. Use the cheapest tier that does the job, or your own key.
- Errors in handler paths still count. On Zapier, steps inside error handler paths and full Zap replays use tasks, so a flaky app can quietly double your usage.
- Silent failures. Zapier sends no error email when an error handler runs, and an n8n cloud workflow over quota just errors. Set up a daily check on each platform's run history; stopping Zapier and Make automations breaking silently lists the alerts worth wiring.
- Self-hosting without an owner. A self-hosted n8n that nobody updates is a security risk with your customer data inside it. If there's no named person to maintain it, use cloud.
A two-week trial that settles it
All three have free tiers or trials, so don't decide from tables, including this one. Build the same single workflow on your top two candidates and run both in parallel for two weeks on real data, with the AI drafts going to a test inbox rather than customers.
- Record the tasks, credits or executions each run actually used; compare with your estimate.
- Break it on purpose: send a form with a missing field and see how each platform reports the error.
- Time how long a change takes, such as adding a new enquiry category.
- Ask whoever will maintain it to make one change alone, without help.
- Multiply the real per-run usage by your busiest month's volume, plus 30%.
The platform that passes steps 2 and 4 comfortably is usually the right one, even if step 5 says it costs a little more.
Choosing between Zapier, Make and n8n: common questions
Does using my own OpenAI or Anthropic key change the cost?
Yes, on all three. On Zapier, an AI by Zapier step uses 1, 3 or 5 tasks per run depending on the model tier, but only one task with your own API key. On Make, third-party AI modules such as OpenAI or Claude use one credit per operation and you pay the provider for tokens separately. On n8n the step sits inside one execution either way. Your own key adds a per-token bill, usually pennies for short classification jobs.
Is n8n free?
The self-hosted Community Edition is free software, but you pay for the server and, more importantly, for someone who can install, update, back up and secure it; n8n's own docs say self-hosting needs technical expertise. The cloud version is paid, from EUR 20 a month billed annually (EUR 24 monthly) for 2,500 executions. For most owner-run businesses without technical help, cloud is the realistic comparison.
Which platform is best for AI agents that call several tools?
Agents take many steps per run, which favours n8n's per-execution billing: one run is one execution however many tool calls it makes. On Make, an AI agent run costs a credit per operation plus credits for tokens and for each tool it calls. On Zapier, each successful action counts as a task. If you expect loops and multiple tool calls, price a realistic run on each before committing.
Can I start on Zapier and move to Make or n8n later?
Yes, but plan on rebuilding rather than copying: each platform structures workflows differently, so every step, field mapping and error rule is recreated and retested. Keep a written description of each workflow (trigger, steps, fields, error handling) from day one. That document turns a painful migration into a few hours of rebuilding per workflow, and it helps whoever looks after the automations next.
Further reads
- Power Automate for Small Businesses: When It Beats Zapier — If you live in Microsoft 365, Power Automate may beat all three.
- Native Integrations vs Zapier: Which Should Connect Your Apps? — Check whether your apps already connect without a platform.
- What Is a Webhook? Why Some Automations Run Instantly — Why instant triggers save credits and speed up runs.
- How to Add Human Approval Steps to AI Automations — Add a person's sign-off before AI output reaches customers.
- AI vs Rule-Based Automation: Which Does Your Task Need? — Decide whether the step needs AI at all.
- How to Set Spending Limits and Alerts on Pay-As-You-Go AI Tools — Cap task, credit and token spend before a loop runs wild.
- Automation Audit: Find the Zaps and Scenarios Nobody Owns — An inventory template, platform-by-platform checks and a triage table for finding orphaned Zaps and Make scenarios before one quietly breaks.
- Can AI Work With the Tools Your Business Already Uses? — The four ways AI connects to the software you run on, a 30-minute stack audit, an insurance broker's tools mapped, and workarounds for old systems.
- What Is an API? Why It Matters When You Buy Software — A plain-English explanation of APIs for people buying software, with the vendor questions and pricing traps that decide whether tools can connect.
- How Driving Schools Use AI to Keep Instructor Diaries Full — Fill cancelled slots in an hour, replace pupils before they pass, cut dead travel between lessons and spot quiet weeks early, with the data each step needs.
- AI Email Triage for Shared Inboxes: Sales, Support, and Invoices — Split a hello@ inbox into sales, support and invoice lanes with AI labels, confidence thresholds and a fraud check on anything asking to be paid.
- How to Stop Retyping Data Between Apps With AI Automation — Find every place your team copies data by hand, match each to the cheapest fix, and use AI extraction only where the source is an email, PDF or free text.
- How Much Does an AI Agent Cost a Small Business? — The five ways AI agents are priced, the cost-per-task sum that makes them comparable, and the supervision time that usually costs more than the software.
- No-Code AI Tools vs a Custom Build: When Templates Stop Working — Six signs a Zapier or Make workflow has outgrown no-code, the middle options to try first, and a worked cost comparison for a care agency's shift offers.
- AI Vendor Lock-In: How to Keep Your Data and Prompts Portable — Where AI lock-in really comes from, which plans can export what in 2026, and a costed exit drill that keeps your prompts and data movable.
- What Connecting Two Business Apps Really Costs: Four Options — Price the whole app connection with four routes, a worked annual budget and tests for duplicates, delays and failures.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Zapier pricing page (task tiers, pay-per-task rates, AI features by plan) and Zapier help on task usage and held runs; Make pricing page, Make help on extra credits and the AI Provider token pricing update (25 Aug 2026); n8n pricing page and n8n support on execution limits; n8n docs on human-in-the-loop for tool calls. Checked September 2026.