n8n vs Zapier vs Make vs Power Automate vs Activepieces: Complete 2026 Workflow Automation Tools Comparison Guide

As of August 2026, workflow automation tools — the ones that wire up repetitive manual work without writing code — are no longer an early-adopter toy. Piping inbound inquiries into a CRM, pushing payment webhooks to Slack, having an LLM summarize something and drop it into Notion: for most teams these pipelines have crossed the line from “nice to have” to “the thing standing between us and burning out a human.” And yet the moment you actually sit down to pick a tool, the story falls apart. This guide takes on the workflow automation tools comparison not as a list of price tags, but from the only angle that matters: what will my workflows actually cost me every month?

There is exactly one reason this category is so uniquely confusing: the five products don’t bill in the same unit. Zapier counts one action step as one task. n8n counts one full run of a workflow as one execution. Make counts one module action as one credit. Power Automate bills in per-user and per-bot licenses, but governs usage separately through a daily request quota (Power Platform requests). And Activepieces uses credits but won’t publish the absolute numbers. Which means the second you line up headline figures like “$19/month vs $9/month,” the comparison is already wrong.

Summer 2026 added one more symbolic data point. Relay.app, widely seen as the category’s rising challenger, announced its shutdown on 16 July 2026. Free accounts end at 23:59 PT on 15 August 2026, paid accounts at 23:59 PT on 14 September 2026, and both new signups and free-to-paid upgrades are already closed. The silver lining: subscriptions were cancelled as of 16 July so there’s no further billing, and workflows, sequences, and MCP servers can be exported as JSON while tables come out as CSV. The company never disclosed why it shut down. The lesson is unmistakable: “will this vendor still exist?” and “can I walk out with my data?” now belong on your evaluation checklist right alongside the feature matrix.

So this article compares n8n, Zapier, Make, Microsoft Power Automate, and Activepieces using only figures verified on each vendor’s official pages as of 2 August 2026. It dissects the billing units, the real constraints hiding inside the free plans, integration counts, the legal differences between self-hosting licenses, and finally a recommendation for each type of user. Numbers without a clear source were left out entirely, and where a vendor simply doesn’t publish something, it says “not disclosed.”

1. Executive Summary: The 5 Major Workflow Automation Tools at a Glance (August 2026)

Here’s the summary for anyone short on time. Every price below is the official list price on annual billing straight from each vendor’s site, in whatever currency that vendor quotes (USD or EUR). n8n, Zapier, Make, and Activepieces have no official Korean-won pricing, so nothing was converted arbitrarily.

📊 Comprehensive Comparison of the Top 5 Workflow Automation Tools (August 2026)

There’s one sentence to take away from this table. There is no “cheap tool” — only a tool that’s cheap for the shape of your workflows. Pipelines with lots of steps and relatively few runs favor n8n by a mile against Zapier — but Make’s credit unit price is low enough that the total can still come out cheaper on Make under those same conditions (section 4 works the break-even out in numbers). A team firing off enormous volumes of simple one- or two-step notifications will do fine on a low Zapier tier or on Make.

2. Deep Dive: Core Capabilities of the 5 Workflow Automation Tools

🟠 1) n8n — A leading option for developer-oriented self-hosted automation

• Per-execution billing is a structural advantage: n8n counts one full workflow run as one execution, no matter how many steps it contains. Three steps or three hundred, same cost. That’s the exact opposite of Zapier’s philosophy of deducting a task per action step, and the gap widens the more complex your data pipeline gets. When n8n’s marketing claims it’s “dramatically cheaper,” the reason isn’t performance — it’s this counting model.
• The heart of n8n 2.0 is security and deployment safety: The n8n 2.x line, released during 2026, enables the task runner by default so Code nodes run in an isolated environment, blocks environment-variable access from Code nodes by default (you must explicitly allow it), and disables the ExecuteCommand and LocalFileTrigger nodes out of the box. In-memory binary data mode was removed entirely. This is the release where n8n confronted head-on the risk of a self-hosted automation tool executing arbitrary code on your own servers.
• Splitting Save from Publish is the change practitioners feel most: Previously, editing was production — one slip and you had an incident. From 2.x onward, Save preserves your edits without touching production, and Publish is the explicit deploy step. The canvas and sidebar UI were redesigned, and a migration tool ships to check compatibility before you upgrade.
• Multi-tier agents via the AI Agent node: Connect a chat model and tools to the LangChain-based AI Agent node and the agent decides for itself which tools to call. Use the AI Agent Tool subnode and you can build multi-tier setups where one agent calls another. Paid plans include AI credits: 2,300 per month on Starter, up to 13,700 per month on Pro.
• Capital and community as survival signals: In May 2026, a strategic investment from SAP valued the company at $5.2 billion. That’s roughly double the $2.5 billion valuation from its October 2025 Series C ($180 million). n8n’s own pricing page advertises 198,813 GitHub stars (a self-reported figure). Right after the Relay.app episode, this line item carries real weight.

⚡ 2) Zapier — 9,000+ apps and still the category leader

• The integration catalog isn’t even a contest: Official figures put it at 9,000+ apps and 30,000+ actions. Next to n8n’s 1,993 and Activepieces’ 700+, that’s a different order of magnitude. If your decision hinges on whether “that weird SaaS tool our company uses” is on the list, you effectively have no alternative.
• ⚠️ Start by relearning the plan names: The old Starter / Professional / Company three-tier structure is gone. As of 2026 it’s a four-tier lineup: Free / Professional / Team / Enterprise. Plenty of blog posts still quote stale tables with entries like “Starter $20,” so when you search for pricing, always check the date on the official page.
• A task is one successful action step; triggers don’t count: A five-step Zap running once burns five tasks. Critically, automation plus AI by Zapier plus Code steps plus Zapier MCP all share a single account quota. There’s no separate budget per product, so if Claude or ChatGPT hammers Zapier actions through MCP, your automation tasks dry up right along with them.
• Overage rates are punitive: Blow past your task limit and you shift to pay-per-task at 2.5× the base rate on monthly autopay, 1.25× on annual. This is the classic path to a shockingly large invoice in a month when traffic spikes.
• AI features are modular add-ons, billed apart from the main subscription: Agents are billed by “activities,” not tasks — 400 activities/month free, with Agents Pro starting at roughly $33.33/month on annual billing. Chatbots are also separate: two free, with paid starting around $13.33/month annually. On the flip side, Tables and Forms consume no tasks at all, and Canvas is available on every plan.

🟣 3) Make — A visual scenario canvas and an aggressive entry price

• $9/month (annual) for 10,000 credits: The Core plan is $9/month at the 10,000-credit tier. Buying that same “ten thousand units” from Zapier Professional costs $129/month; from n8n Pro, €50. The units aren’t equivalent, so it isn’t a direct comparison — but no product has a lower psychological barrier to entry.
• A credit equals one module action: As of 27 August 2025, “operations” were renamed “credits.” Prices and limits were frozen, and leftover operations converted 1:1 into credits. It looks like a pure rebrand, but one substantive change is hiding inside.
• Only the AI modules moved to variable billing: The native OpenAI/Anthropic connectors, image-processing modules, and similar now consume variable credits based on actual usage rather than a flat 1 credit. If your scenario leans hard on LLMs, credit burn becomes hard to forecast — so treat your first month as a measurement period and lock in a plan only after watching the consumption logs.
• What separates the plans is operational comfort, not features: Core gives you unlimited active scenarios and a 1-minute minimum interval; Pro adds priority execution and full-text search over execution logs; Teams adds team roles, shared scenario templates, and Make Grid. Every paid plan comes with unlimited seats, so your bill doesn’t scale linearly with headcount — a decisive difference from Zapier. Execution log retention is 30 days on Free through Teams and 60 days on Enterprise.
• The AI lineup reaches all the way down to the free plan: Make AI Agents is in beta but available on every plan including Free, and you can use Make’s own AI Provider or connect your own LLM API key (BYOK). Alongside it are Maia, which builds scenarios from natural language, and Make Grid (Teams and above), which visualizes your entire automation landscape.

🟦 4) Microsoft Power Automate — The only product where you buy seats, not usage

• The bill is a flat seat price, but usage is capped by a hard quota: Alone among the five, billing runs on flat per-user and per-bot licenses, so your invoice doesn’t climb linearly even when run volume explodes. That part is true. What isn’t true is the common claim that it “doesn’t meter usage at all.” Microsoft maintains a separate meter called Power Platform requests (PPR) and caps Power Automate Premium at 40,000 PPR per user per 24 hours and Process and Hosted Process at 250,000 PPR per license per 24 hours (there’s also a license-independent limit of 100,000 requests per five minutes). What counts as one request is essentially the same as a Zapier task. In the official docs’ own words, it covers connector calls and HTTP actions plus “built-in actions from initializing variables to a simple compose action,” both successful and failed actions count, and retries and pagination requests count too. In this product, actions are metered by quota rather than by price.
• Exceed the limit and your flows react before your invoice does: Overages don’t raise your bill — they get you throttled. To fix it you either buy the Power Platform request capacity add-on, which raises the limit by another 50,000 per 24 hours, or switch to pay-as-you-go (Process licenses can also be stacked on a single flow, each adding 250,000 PPR). Every organization is currently in a transition period with more generous limits than the official ones, but Microsoft itself tells you to design your flows against the official numbers. So for the team whose “triggers fire tens of thousands of times a day and per-task pricing terrifies us,” the fixed invoice is still a genuine advantage — but the problem doesn’t vanish, it converts from a pricing problem into a quota-design problem. A five-step flow running 10,000 times a day is 50,000 PPR on its own, already past the official limit of a single Premium seat. Two mitigating notes: desktop flow (RPA) executions don’t consume PPR, and Dataverse capacity and Copilot credits are consumed separately from this quota anyway.
• The only product with official Korean-won pricing: Power Automate Premium is $15.00 (₩20,300) per user/month, Power Automate Process is $150.00 (₩202,700) per bot/month, and Hosted Process is $215.00 (₩290,500) per bot/month. The other four have no official KRW list price at all, which means the buying team absorbs the FX and VAT risk. If you’re evaluating for the Korean market, where procurement teams often require a local-currency quote, that difference matters more than it sounds.
• The only product with native RPA: Desktop-flow-based RPA is built in. Attended desktop flows — the kind where a human is at the machine — are included in Premium, while unattended runs require Process ($150/bot). If you need to automate systems with no API, such as legacy groupware or in-house Windows applications, the other four products can’t even get you off the starting line.
• Process mining is part of the product: Premium includes process and task mining with 50 MB of storage, and using it seriously requires the Process Mining add-on at $5,000/month per tenant (₩6,756,800), which provides 100 GB of data storage, 2 GB of Dataverse DB, and 1 TB of file storage. It’s a Premium-only add-on.
• You have to understand the 2026 strategic shift: Microsoft is migrating from pure per-seat SaaS to a “seats + agents + consumption” hybrid. As of 1 July 2026, Copilot pricing was restructured around four SKUs — Business Standard + Copilot at $23.50, Business Premium + Copilot at $32, the E7 Frontier Suite at $99, and Copilot Business at a $21 list price — and the $18/user/month promo for organizations under 300 seats ended on 30 June 2026. On 16 June 2026, Copilot Cowork and the Work IQ API hit GA, and both consume Copilot Credits on top of seat fees. In other words, the era of “buy the seats and you’re done” is over.
• Governance is still in a class of its own: Integration with Entra ID, Purview, Teams, SharePoint, and the rest of the M365 stack makes this the easiest option to push through an enterprise security review. Conversely, reviewers consistently note that the licensing, Dataverse, and connector-tier (standard vs. premium) structure is a bigger barrier to entry than the tool itself.

🟩 5) Activepieces — The only genuinely open-source option of the five

• The MIT license is the decisive differentiator: It’s the only OSI-certified open source product of the five. What n8n’s Sustainable Use License forbids — charging for access to the runtime itself, i.e. reselling hosting — is perfectly legal with Activepieces. For a company that wants to bolt an automation engine into its own product and sell it, that isn’t a feature difference; it’s the difference between having a business and not.
• Community Edition is forever free with unlimited flows — but it stops at “the automation core”: The self-hosted Community Edition is completely free with no cap on flows, runs, or users. That said, the official pricing page calls CE “the automation core” and spells out what’s left out: Agents and Chat, Projects, API access, and the entire team and admin layer — meaning SSO, user roles, audit logs, secret managers, branding, and Git Sync. That list deserves exactly the same weight as the fact that n8n Community lacks SSO, Git, and environment separation. The cloud has a separate Free plan (1 seat, unlimited flows, 700+ integrations).
• AI agents weren’t carved out into a paid add-on: Alone among the five, AI agents ship with every cloud plan, cloud Free included. That’s the polar opposite of Zapier selling Agents separately by activities, and Activepieces also offers native MCP (Model Context Protocol) support. Be precise about the scope, though: this is a cloud story, and the self-hosted CE contains no agents and no chat. Standing up CE in Docker to get free AI agents will not get you the feature you came for.
• A hybrid where even paid plans can run on-prem: Plus at $16/month (up to 5 users) and Team at $166/month (25 users included, $25/user/month beyond that) can both run in the cloud or on-premises. Team adds 5× the credits of Plus, SSO, standard roles, and unlimited projects, while Ultimate is quote-based and provides annual rather than monthly credits, SCIM provisioning, custom RBAC, audit logs, and Git Sync. For embedding in your own product, Embed starts at $36,000/year.
• Two weaknesses worth naming honestly: First, at 700+ integrations (“pieces”) it has the smallest catalog of the five. Second, the official pages don’t publish the absolute credit amounts for any plan. They say only “monthly standard credits” and “5× Plus,” which makes pre-purchase capacity planning impossible — get the numbers from sales before you commit. The company was founded in 2022 by Mohammad AbuAboud and Ashraf Samhouri and went through Y Combinator.

3. Head-to-Head: Billing Units, Integration Coverage, and Free Plans

The most common misconception in any workflow automation tools comparison is “which one is faster.” But no neutral, second-by-second performance benchmark exists in this category. Even n8n’s “5/20/200+ concurrent executions” is a plan specification, not a performance measurement. So this section pits the five against each other on the three axes that actually decide purchases — billing efficiency, integration coverage, and free plans — using official figures only.

[Billing efficiency for step-heavy workflows]
n8n            : ⭐⭐⭐⭐⭐ (one full workflow run = 1 execution, unlimited steps)
Power Automate : ⭐⭐⭐☆☆ (flat seat billing, but actions metered as PPR — 40k/user/24h on Premium)
Activepieces   : ⭐⭐⭐⭐☆ (zero license fee self-hosted / credits on cloud, Free has a daily cap)
Make           : ⭐⭐⭐☆☆ (one module action = 1 credit, but cheap per unit)
Zapier         : ⭐⭐☆☆☆ (one action step = 1 task, multiplied by step count)

[Integration coverage]
Zapier         : ⭐⭐⭐⭐⭐ (9,000+ apps / 30,000+ actions)
Make           : ⭐⭐⭐⭐☆ (2,000–3,000 apps / 30,000+ actions, 350+ AI apps)
n8n            : ⭐⭐⭐☆☆ (1,993 nodes)
Activepieces   : ⭐⭐☆☆☆ (700+ pieces)
Power Automate : No official connector count available (best-in-class inside M365)

📈 Official Integration Counts Compared (2026-08)

There’s a simple reason Make’s integration count isn’t pinned to a single number: even Make’s own site quotes it differently on different pages. The AI Agents page says “3,000+ integrations”; the general pages and the free plan description say “2,000+ integrations.” Be skeptical of any comparison article that confidently commits to one of them.

🆓 Free Plans Compared — The real constraint isn’t the task count

The most common mistake when picking a free plan is reasoning like “100 Zapier tasks is three a day, that’ll do.” The real wall on Zapier Free isn’t tasks, it’s the two-step limit. “Form submitted → write to a sheet → notify Slack” is three steps, which means you simply cannot build it on the free tier. Likewise, the real wall on Make Free isn’t the 1,000 credits — it’s the 2 active scenarios and 15-minute minimum interval. Translation: it’s unusable for real-time alerting.

And that same yardstick has to be applied to Activepieces Free to be fair. Activepieces Free gives you unlimited flows plus AI agents, which on the feature list looks like the most generous free tier of the group — but it has a wall of exactly the same kind. The credits the official page describes are not monthly; they’re a daily hard cap refilled with a fresh batch every day, and once the day’s allotment is gone, your flows stay switched on while any remaining runs simply don’t go through until credits refresh the next day. The absolute amount is undisclosed too, so you can’t know in advance how many runs a day you get. For a learner firing off test runs several times a day, that bites as hard as Zapier’s two-step limit. On the feature axis, though, the verdict holds: Activepieces Free is still the least restricted free tier of the five. n8n has no forever-free cloud tier, so if you want to use it for nothing, standing up Community Edition on your own server is the only path. For a team that can spin up a container with one command it’s arguably the strongest free option in the category; for everyone else it isn’t an option at all.

🔓 Self-Hosting and Licensing — This is the real fork in the road

n8n’s Sustainable Use License permits internal business use, modification, and derivative works, plus free and non-commercial redistribution. A 2022 revision explicitly allows selling n8n-related consulting, implementation, and support services as well. Exactly one thing is forbidden: charging for access to the n8n runtime itself (reselling hosting), which requires a separate Enterprise/Commercial License. “It’s open source so I can do whatever I want” is the misunderstanding that gets billed most expensively here, so verify before you deploy.

The learning curve is a qualitative judgment rather than a number, but most reviews agree on the ordering, easiest to hardest: Zapier < Activepieces < Make < Power Automate ≈ n8n. Zapier has a linear step UI and a natural-language Copilot builder; Make requires you to internalize routers and iterators on a node canvas; and n8n effectively becomes a developer tool the moment JSON data structures and the Code node enter the picture. With Power Automate, the licensing scheme is harder than the tool. Because tool choice is inseparable from team composition, you should ask the same question here that you’d ask when doing a team collaboration tools comparison — who on this team is actually going to maintain the thing?

4. Complete Pricing Comparison (Official List Prices, August 2026)

[Entry paid prices at a glance — official annual billing]
Make Core              : $9 / mo        (10,000 credits, unlimited seats)
Power Automate Premium : $15 / user/mo  (₩20,300, attended RPA included)
Activepieces Plus      : $16 / mo       (up to 5 users, credits + PAYG)
Zapier Professional    : $19.99 / mo    (750 tasks, 1 seat)
n8n Starter            : €20 / mo       (2,500 executions, unlimited users)

💳 n8n Official Pricing (EUR, annual billing / monthly billing in parentheses)

Companies with fewer than 20 employees can get 50% off Business through the Startup plan. Every paid plan includes unlimited users, workflows, and steps, plus the full integration catalog. Do note that on Starter and Pro, exceeding your limit can pause your workflows.

💳 Zapier Official List Prices by Task Tier (USD/month)

The top tier scales all the way to 2 million tasks per month, with Professional maxing out at $3,389 annual / $5,099 monthly and Team at $3,999 annual / $5,999 monthly. Annual billing runs about 33% cheaper than monthly, Free includes 100 tasks and 1 seat, and Team includes 25 seats. The catch to watch: Professional is a single seat. The moment two or more people need to co-manage Zaps, you’re effectively forced up to Team (from $69).

💳 Make Official Pricing (USD, at the 10,000-credit tier)

Make’s official page shows $9/$16/$29 alongside a “Save 15% or more!” annual toggle, and monthly billing is generally reported at roughly $10.6 / $18.8 / $34. Since the vendor doesn’t print both figures side by side, this article treats only the annual numbers as confirmed.

⚠️ Remember that this table shows exactly one tier: the base 10,000 credits. Zapier prints its whole ladder, from 750 tasks up to 2 million, right on the page. Make works differently. Its official pricing page has a credit slider that steps through 10k → 20k → 40k → 80k → 150k → 300k → 500k → 750k → 1M → … → 8M+, and moving the slider moves the Core/Pro/Teams prices with it. Higher-tier amounts only render once you drag the slider — they aren’t printed statically — so this article won’t publish them as confirmed figures. If your calculation lands above 10,000 credits, go to make.com/en/pricing, set the slider to your own credit number, and read the real price yourself. Activepieces goes a step further: the absolute credit amounts are undisclosed entirely, so there’s no way to size capacity without talking to sales.

💳 Power Automate and Activepieces Official Pricing

Copilot Studio also offers Azure pay-as-you-go at $0.01 per credit (standard overage $0.01, generative overage $0.02).

💸 Hidden Costs — It isn’t the list price that inflates your invoice

First, the illusion created by different billing units. Here’s a worked example (arithmetic derived from each vendor’s official definitions, not measured data). Run a five-step workflow 1,000 times a month and you consume 5,000 Zapier tasks, 1,000 n8n executions, and 5,000 Make credits. Power Automate bills a flat seat price, so its cell holds not a dollar figure but 5,000 PPR per month (comfortably inside Premium’s 40,000-per-day limit), and Activepieces is the one cell that can’t be computed at all because its credit amounts are undisclosed.

Now bump the same workflow to 2,000 runs a month. Zapier hits 10,000 tasks, landing you in the $129/month Professional annual bracket, while n8n is at 2,000 executions and still comfortably inside Starter (€20/mo, 2,500 executions). That’s the calculation everyone quotes — and this is exactly the point where Make quietly drops out of the comparison, which is the most common trap in this category. Under the same conditions Make is at 10,000 credits, i.e. Core at $9/month. Less than half of n8n’s €20.

So “the more steps you have, the more overwhelmingly n8n wins” is unconditionally true only against Zapier. Against Make it comes with conditions, and the break-even falls straight out of the shape of each vendor’s base tier. n8n Starter (€20) covers 2,500 executions a month regardless of step count, while Make Core ($9) holds only as long as (steps × monthly runs) stays under 10,000 credits. That means n8n overtakes Make in the band where (steps × monthly runs) exceeds 10,000 while monthly runs themselves stay at or below 2,500. For instance, 15 steps × 1,000 runs a month is 15,000 Make credits — slider goes up — while n8n is at 1,000 executions and stays at €20. Flip it around and shapes with few steps and many runs, like 5 steps × 2,000 runs (10,000 credits) or 2 steps × 3,000 runs, stay cheaper on Make. The widening gap as step count grows is a Zapier comparison; against Make you have to solve those two inequalities with your own numbers.

Second, overage rates. Zapier sells extra tasks at 2.5× on monthly autopay and 1.25× on annual once you exceed your limit. n8n Starter/Pro may pause your workflows instead. Activepieces charges PAYG on overage. Which of those is scarier depends on your team’s temperament — decide in advance whether you’d rather have the service quietly stop or the invoice quietly grow.

Third, AI add-ons. Zapier bills Agents (activities) and Chatbots outside the main subscription, Make’s native AI modules run on variable credits, and Power Automate has Copilot Cowork and Work IQ eating Copilot Credits on top of seat fees. It’s most accurate to think of 2026 automation pricing as a three-layer structure: “base fee + agents + consumption.” Conversely, n8n bundles AI credits into paid plans and Activepieces ships AI agents on every cloud plan, so the heavier your AI usage, the more predictable those two look by comparison.

Fourth — and this is the item this article has most often called “free” — the total cost of ownership of self-hosting. With n8n Community and Activepieces CE, the license fee is zero; the cost is not. A server that has to stay up 24/7 (VPS or your own hardware), database and credential backups, TLS certificates and domains, monitoring and someone on call for incidents, and queue/worker separation once run volume grows all land on your team. Server pricing itself varies so much by spec and provider that this article won’t pin a figure to it — because the underestimated cost isn’t the money, it’s the person-hours.

The signature line item there is version-upgrade labor. The n8n 2.x migration covered in Tip 3 of the next section — Code nodes losing environment-variable access, ExecuteCommand and LocalFileTrigger disabled by default, in-memory binary mode removed — is that cost. On managed cloud the vendor absorbs that change; when you self-host, your team pays it in auditing, fixing, and redeploying every workflow. So answer this before you commit: “how many hours a month will we spend on server upkeep and upgrades in order to avoid a €20 n8n Starter (or a $16 Activepieces Plus)?” If the labor cost of those hours exceeds the subscription, then the reason to self-host has to be data sovereignty, not savings. For a team whose data genuinely cannot leave the building, that price is well worth paying; for a team just trying to save twenty euros, it usually isn’t.

5. Final Recommendations by User Type

This is where the workflow automation tools comparison turns into an actual decision. Pick the single persona below that’s closest to your situation and your shortlist naturally collapses to one or two candidates.

🎯 1) A startup with one or two developers running 10+ step data pipelines

• Best pick: n8n
• Why: One full workflow run is one execution, so adding steps costs you nothing. Starter at €20/month gives unlimited users, workflows, and steps, plus 2,300 AI credits. Build the same job in Zapier and your tasks drain as a multiple of your step count.

🎯 2) A team with no developers that needs to connect 20 marketing and sales SaaS tools

• Best pick: Zapier
• Why: 9,000+ apps and 30,000+ actions puts its coverage an order of magnitude above the other four. The linear step UI and natural-language Copilot builder let a non-technical user ship their first automation the same day. Just note that with two or more people you should be looking at Team (from $69, 25 seats) rather than single-seat Professional.

🎯 3) A small team on a tight budget that wants to design branching and looping logic visually

• Best pick: Make
• Why: Core is $9/month on annual billing for 10,000 credits, and every paid plan has unlimited seats, so headcount growth doesn’t raise the bill. Once you’ve absorbed routers and iterators, you can design complex branching and literally watch it on the canvas. If you plan to lean on AI modules, be sure to measure variable credit burn during month one.

🎯 4) An enterprise already on M365 that also needs to automate API-less legacy systems

• Best pick: Power Automate
• Why: It’s the only one of the five with native RPA (desktop flows), and attended RPA is included in Premium at $15 (₩20,300) per user/month. Add Process ($150/bot, ₩202,700) if you need unattended runs. Entra ID and Purview-based governance makes security review the smoothest of the group, and since it’s the only product with an official Korean-won price, internal purchase approval is far easier for Korea-based buyers. One caveat: a flat seat price does not mean unlimited runs. Premium caps you at 40,000 Power Platform requests per user per 24 hours and Process at 250,000 per license (built-in actions, failures, and retries all count), and exceeding that gets you throttled — so if you have high-frequency flows, design the capacity add-on (+50,000/24h) or pay-as-you-go into your licensing plan from day one.

🎯 5) A company embedding an automation builder into its own SaaS, or reselling one

• Best pick: Activepieces
• Why: The MIT license puts no restrictions on commercial resale or productizing it as SaaS. The “selling runtime access” that n8n’s fair-code license forbids is entirely legal here. If you need full white-labeling, Embed (from $36,000/year) provides a JavaScript SDK and an embeddable builder.

🎯 6) A team handling customer PII or internal databases where data can’t leave the building

• Best pick: n8n Community or Activepieces Community Edition
• Why: Both are free to self-host with no cap on executions or flows. But you have to look at what each community edition leaves out at the same eye level. n8n Community has no SSO, no Git version control, and no environment separation (dev/prod) — if you need those, you’re moving up to Business (€667/mo) or Enterprise. Activepieces Community Edition, in the vendor’s own words, stops at “the automation core” and excludes Agents and Chat, Projects, API access, SSO, user roles, audit logs, secret managers, branding, and Git Sync. Two of those are lethal for this specific persona: with no secret manager you cannot delegate credentials to an external vault (that’s precisely the “where are credentials stored” item on the checklist below), and with no API access you’re locked out of deployment automation and external audit tooling. If audit logs, SCIM, Git Sync, and secret management are hard requirements, the real candidates are Activepieces Ultimate or n8n Enterprise. Weigh self-hosting TCO (section 4) and license character (fair-code vs. MIT) as part of the same decision.

🎯 7) A solo operator who has never automated anything

• Best pick: Start with Activepieces Free or Make Free
• Why: On Zapier Free, the two-step limit stops you long before the 100 tasks do; on Make Free, the two active scenarios and 15-minute interval stop you first. Activepieces Free is one seat but includes unlimited flows and AI agents, making it the least restrictive place to learn on the feature axis. It has one wall of its own, though: its credits are a daily hard cap refilled each day, not a monthly pool, so once the day’s allotment is gone your flows stay switched on while the remaining runs silently don’t go through (the amount is undisclosed). If executions suddenly stop mid-lesson, it’s probably exhausted credits rather than a bug — try again tomorrow. Get a feel for it, look at the shape of your workflows, and only then pick a paid plan — that order is the cheapest.

6. Practical Tips and Common Mistakes

✅ Tip 1) Before choosing a tool, write down “step count × monthly runs” on paper

You can read every workflow automation tools comparison ever written, and it won’t help if you skip this calculation. Most failed rollouts in this category happen because people look at the pricing page before they measure the shape of their workflows. Flip the order. Pick three to five workflows you want to automate and write down, for each, how many action steps it contains and how many times a month it triggers. Then compute tasks for Zapier as (steps × runs), executions for n8n as runs alone, and credits for Make as (modules × runs). Plug in the worked example — 5 steps × 1,000 runs a month equals 5,000 Zapier tasks, 1,000 n8n executions, 5,000 Make credits — and 30 minutes will give you the outline of which tool is cheap for your team.

One honest caveat, though. That calculation converts straight into a dollar figure for Zapier and n8n only — they’re the two that print their whole tier ladder on the page. Make requires you to drag the credit slider on the official page the moment your result exceeds 10,000 credits. Power Automate turns the exercise into a different question entirely: not “how much,” but “does this fit inside 40,000 PPR per user per day on Premium?” And Activepieces won’t yield a number at all without a sales conversation, because its credit amounts are undisclosed. Do the arithmetic for all five, but expect only two to give you an answer on paper — put the verification step for the other three on your calendar.

✅ Tip 2) If you’re on Zapier, harvest everything that doesn’t consume tasks

In Zapier, Tables and Forms consume no tasks. If you park intermediate data in an external Google Sheet, every read and write costs a task; keep it in Zapier Tables and that part is free. Canvas is available on every plan, so use it to sketch the workflow first and cut step count. Conversely, AI by Zapier, Code steps, and Zapier MCP all share the same task pool as your automations — hook MCP up to Claude or ChatGPT and forget about it, and your automation budget will quietly evaporate.

✅ Tip 3) Run the migration tool before upgrading to n8n 2.x

n8n 2.x tightened security substantially. Environment-variable access from Code nodes is blocked by default, the ExecuteCommand and LocalFileTrigger nodes are disabled by default, and in-memory binary data mode has been removed entirely. Workflows that leaned on those features back in the 1.x days will fail silently right after you upgrade. Always run n8n’s pre-upgrade compatibility migration tool first, and make the new Save / Publish split a hard team rule. A one-line rule — “Save is a draft, only Publish is production” — cuts your incident rate in half.

✅ Tip 4) If you’re using AI modules in Make, treat month one as an observation period

When operations became credits in August 2025, prices and limits were frozen — but the native OpenAI/Anthropic connectors and image-processing modules moved to variable credits based on actual consumption. Which means the same scenario burns different amounts depending on input length. Run it for a month on Free (1,000 credits) or Core ($9), confirm real consumption from the logs, then lock in a tier. Note that execution log retention is 30 days on Free through Teams, so finish your analysis inside that window. If you’re unsure which LLM to wire up, checking model characteristics in our AI models comparison first will save you a lot of wasted credits.

✅ Tip 5) With Power Automate, licensing design is the implementation

The single most common mistake with this product is confusing attended and unattended. Attended RPA, which runs on a PC where a human is logged in, is included in Premium ($15/user/month) — but unattended RPA running overnight requires Process ($150/bot/month) as a separate purchase. Most “it worked in testing but fails as a nightly batch” support tickets trace back to exactly this. Also, Dataverse capacity differs by SKU (Premium gets 250 MB DB; Process and Hosted Process get 50 MB), so if your flow accumulates a lot of data, size your capacity first. Some M365 plans include limited rights to standard connectors, but it varies by plan — check your own contract.

❌ Common mistake 1) Trusting an outdated price table

Zapier’s Starter / Professional / Company three-tier structure has already been retired, yet a great many comparison articles still quote “Starter $20.” Activepieces has the same problem: numerous review sites still publish the old pricing scheme — “Plus $25/mo, Business $150/mo, 1,000 free tasks, 2 active flows” — as if it were 2026 pricing, in direct contradiction to the official page ($16/$166, unlimited flows, credit-based). Power Automate’s old “per user plan / per flow plan” SKU names have been retired too. Always confirm pricing on the official page, dated today.

❌ Common mistake 2) Putting everything into one tool with no migration plan

The practical lesson from Relay.app’s shutdown isn’t “avoid small vendors” — it’s “make sure you can always walk out.” Relay.app at least opened the door, letting customers export workflows, sequences, and MCP servers as JSON and tables as CSV. At adoption time, check three things: (1) can you export workflow definitions as code/JSON, (2) where are credentials stored, and (3) is there a self-hosting escape hatch? Viewed through that lens, n8n Community and Activepieces Community Edition are structurally advantaged. If you’re worried about where your team’s working data piles up, the data-portability lens from our project management tools comparison applies just as cleanly to automation tools.

7. Frequently Asked Questions (FAQ)

Q. Which is cheaper, n8n or Zapier?

It depends on your step count. n8n counts one full workflow run as one execution; Zapier counts one successful action step as one task. Run a five-step workflow 1,000 times a month and n8n sees 1,000 executions while Zapier sees 5,000 tasks. The more steps you have, the more n8n wins; with simple one- or two-step notifications the gap shrinks dramatically. Also factor in that Zapier Professional is a single seat, so team usage pushes you up to Team (from $69).

Q. Does n8n have a forever-free plan?

As of 2026, n8n Cloud has no forever-free plan. You get a 14-day free trial (1,000 executions, no card required), and compute during the trial is Starter-grade. The only way to keep using it for free is the self-hosted Community Edition, where executions are unlimited but SSO, Git version control, and environment separation are not included.

Q. Is n8n open source? Can I sell a service built on it to customers?

n8n is fair-code under the Sustainable Use License, and it is not open source by the OSI definition. Internal business use, modification, and derivative works are permitted, and a 2022 revision also allows charging for n8n-related consulting, implementation, and support services. However, selling access to the n8n runtime itself (reselling hosting) is prohibited and requires a separate Enterprise/Commercial License. If resale is the whole point, MIT-licensed Activepieces is your answer.

Q. If I’m starting automation for free, which tool is best?

The constraints differ in kind. Zapier Free gives you 100 tasks and 1 seat but permits only two-step workflows, so anything with three or more stages is off the table. Make Free gives 1,000 credits but caps you at 2 active scenarios with a 15-minute minimum interval, making it unsuitable for real-time alerts. Activepieces Free is one seat but includes unlimited flows and AI agents, making it the least restrictive on the feature axis. Its credits, however, are a daily hard cap refilled each day rather than a monthly pool, so once the day’s allotment is spent your flows stay on while remaining runs silently don’t go through (amount undisclosed). If you can stand up your own server, n8n Community is the most powerful of all with unlimited executions — at the price of your team absorbing server operations and version-upgrade labor.

Q. I heard Make got rid of “operations” — did prices go up?

Nominally, no. As of 27 August 2025, operations were simply renamed credits, prices and limits were frozen, and leftover operations converted 1:1. What did substantively change is that native AI modules (OpenAI/Anthropic, image processing, and so on) moved from a flat 1 credit to variable credits based on actual consumption. If your scenarios use a lot of AI, your effective cost can rise.

Q. Is Power Automate worth it for a company that doesn’t use M365?

Not really. Its greatest strength is integration with the M365 governance stack — Entra ID, Purview, Teams, SharePoint — and without M365 that advantage evaporates, leaving only the seat cost of Premium at $15 (₩20,300) per user/month. That said, if you absolutely need RPA for legacy systems with no API, it’s the only one of the five with native RPA, so it can still be a candidate regardless of M365.

Q. Which tool is best if I want to use AI agents?

The billing models diverge sharply. Activepieces includes AI agents on every cloud plan, cloud Free included, and supports native MCP. Mind the scope, though: the self-hosted Community Edition contains no agents and no chat, so spinning up CE to get free AI agents leaves you without the feature. Free agents means the cloud Free plan. Make AI Agents is in beta but usable on every plan including Free, and you can attach your own LLM key (BYOK). n8n lets you build multi-tier agents with its LangChain-based AI Agent node and bundles AI credits into paid plans. Zapier Agents, meanwhile, are billed separately by activities rather than tasks (400/month free, Agents Pro from roughly $33.33/month annually), which makes its cost structure the most complicated.

Q. What happens if an automation tool suddenly shuts down?

Relay.app’s shutdown, announced 16 July 2026, is the live case study. Free accounts end at 23:59 PT on 15 August 2026, paid accounts at 23:59 PT on 14 September 2026, and new signups were closed. Subscriptions were cancelled as of the announcement date so there was no further billing, and workflows, sequences, and MCP servers could be exported as JSON with tables as CSV. No reason was given for the shutdown. The lesson is clear: before you adopt anything, confirm you can export your workflow definitions to files, and where possible pick a product with a self-hosting escape hatch.

Q. What does self-hosting actually cost?

The license fee is zero; the total cost isn’t. The software itself is free for both n8n Community and Activepieces CE, but a server that stays up 24/7 (VPS or your own hardware), database and credential backups, TLS certificates and domains, monitoring and incident response, and queue/worker separation once volume grows are all team costs. Server pricing varies so much by spec and provider that this article won’t pin it to one figure. The decision rule compresses to a single line instead: if the labor cost of the person-hours you spend each month on server upkeep and version upgrades exceeds an n8n Starter at €20 (or an Activepieces Plus at $16), then the reason to self-host has to be data sovereignty, not cost savings. Layer on the features each community edition drops (n8n: SSO, Git, environment separation; Activepieces CE: agents and chat, projects, API access, SSO, roles, audit logs, secret managers, Git Sync) and you have an honest comparison.

Q. Can I migrate my workflows from Zapier to n8n?

There is no automatic converter. Neither vendor offers an official tool to turn a Zapier Zap into an n8n workflow JSON, so migration is effectively a manual rebuild. And that’s the better path anyway: the moment the billing unit changes from tasks to executions, a design that punished you for splitting into many steps on Zapier flips to favoring consolidation into one workflow on n8n. Port it verbatim and you leave half the billing advantage on the table. The realistic sequence is (1) rebuild only your two or three highest-traffic Zaps in n8n first, to measure real execution counts and maintenance burden, (2) reissue credentials in the new tool, and (3) run both tools side by side for a while, comparing outputs, before you cut over. Documenting each existing Zap’s trigger and action layout — notes or screenshots — before you start will speed the rebuild up considerably.

Q. If I really had to pick just one, which is it?

Answering “it depends” is the same as not answering, so here’s a named default. For a team with no distinguishing conditions, the default is Make Core ($9/month on annual billing, unlimited seats, unlimited active scenarios). This article’s own price tables explain why: it has the lowest entry price of the five, the bill doesn’t rise as headcount grows, and it’s cheapest in total for the most common workflow shape — few steps, many runs. If you have no developers, aren’t on M365, have no resale plans, and haven’t settled the shape of your workflows yet, starting here is the correct bet on the odds.

There are three conditions under which you should not pick Make: if you lean hard on AI modules and can’t forecast variable credit burn; if your data cannot leave for an external cloud (Make has no self-hosting); and if an integration you absolutely need isn’t in Make’s catalog.

For the other four, here’s a “don’t pick this if” line each. n8n: with no developer to maintain it, or with one- and two-step workflows, the learning curve and self-hosting operations cost more than you save. Zapier: with many steps, or with two or more people managing Zaps (Professional is a single seat), it becomes the most expensive option on both the task and seat axes. Power Automate: without M365 and without a need for RPA, all that’s left is seat cost and licensing complexity. Activepieces: with no resale or embedding plan, and with the app you need missing from the 700 integrations, there’s no particular reason to choose it.

And in practice, plenty of teams run two in parallel. The classic split is internal pipelines on self-hosted n8n and the marketing team’s SaaS connections on Zapier.

8. Conclusion: Wrapping Up the 2026 Workflow Automation Tools Comparison

As of August 2026, the conclusion in this category isn’t “which product is best” but “which billing unit is a good match for the shape of my workflows.” Zapier still guards its moat of 9,000+ app integrations; n8n has been validated at a $5.2 billion valuation by SAP’s investment while sharpening its structural weapons of per-execution billing and self-hosting; Make holds onto small teams with a $9/month entry point and unlimited seats; Power Automate defends the Korean enterprise market with RPA and official won-denominated pricing; and Activepieces sells a freedom under the MIT license that the other four cannot imitate.

One trend shows up across all five. Pricing is being restructured into a three-layer model of “base fee + AI agents + consumption.” Microsoft has officially shifted from per-seat SaaS to a “seats + agents + consumption” hybrid, Zapier has detached Agents and Chatbots from the main subscription, and Make has moved its AI modules alone to variable credits. Relatively speaking, the more predictable players in this trend are n8n, which folds AI credits into its plans, and Activepieces, which ships AI agents on every cloud plan by default (though Activepieces’ self-hosted CE contains no agents either).

Finally, let me put Relay.app’s question back on the table. Nobody knows whether the tool you pick today will still exist in two years. So on day one, confirm whether you can export workflow definitions, where credentials live, and whether there’s a self-hosting escape hatch. Handle those three and you’ll avoid the worst outcome no matter which tool you choose. Automation works quietly for you once you’ve built it — but real design means preparing for the day that quiet breaks.

Summary: the right pick for your situation
────────────────────────────────────────────────────────────
No distinguishing conditions (default)    → Make Core ($9/mo, unlimited seats & scenarios)
Developers on staff + 10+ step pipelines  → n8n (Starter €20/mo, per-execution billing)
Non-technical team + 20+ SaaS to connect  → Zapier (Professional $19.99 / Team from $69)
Minimum budget + visual branching design  → Make (Core $9/mo, unlimited seats)
Already on M365 + need RPA                → Power Automate (Premium ₩20,300/user/mo)
Embedding or reselling in your own product→ Activepieces (MIT / Embed from $36,000/yr)
Data must never leave your infrastructure → n8n Community or Activepieces CE
                                            (zero license fee; server, ops and upgrade labor extra)
Complete beginner, learning by doing      → Activepieces cloud Free
                                            (unlimited flows + AI agents; credits are a daily hard cap)
────────────────────────────────────────────────────────────
* All prices per each vendor's official pages as of 2026-08-02.
  Power Automate is the only product with an official Korean-won price.
  Power Automate is flat per seat but carries a separate quota of 40,000 PPR per user per 24h on Premium.

Sketch your team’s workflows on paper and multiply step count by run frequency. The answer to the workflow automation tools comparison is already sitting inside that calculation. 🚀