n8n, Make or Zapier, or a Custom AI Agent? When Each One Pays Off
If your process follows fixed rules and connects SaaS apps, a platform is enough: n8n from €20/month on its cloud (or free self-hosted for internal use), Make from $9/month or Zapier from $19.99/month. A custom AI agent, from €25,000 with us, only pays off with complex business logic, sensitive data, high volume or in-house systems. We implement both, so we will tell you which one fits.
- €25,000 Where a custom agent in production starts
Definition
n8n, Make or a custom AI agent?
Use Zapier or Make if you want to automate without coding and your data can live in their cloud; n8n if you want control over where your data lives (self-hosted or on its Frankfurt cloud) and to pay per execution rather than per step; and a custom AI agent when the process needs to reason over your data, integrate with in-house systems and measure the quality of every answer.
In numbers
The Comparison in Four Numbers
Entry plans checked on 25-Sep-2026. Careful: each platform counts something different.
- 2,500 n8n Starter executions €20/month billed annually (€24 monthly). One execution is the whole workflow, however many steps it has.
- 10,000 Make Core credits $9/month. Each module action uses one credit; some AI features use more.
- 750 Zapier Professional tasks From $19.99/month billed annually. Each action step uses tasks; the trigger does not.
- €25,000-50,000 Custom agent (Kiwop) One agent with 2-3 tools on one of your systems, 4-6 weeks to production. Published price.
Key points
Six Differences That Matter More Than the Integrations List
All three platforms connect to almost everything. This is what sets them apart.
- 01
Who Maintains What
Zapier and Make maintain the platform; you maintain your workflows. n8n Cloud, the same. Self-hosted n8n: you (or whoever implements it) also update and monitor the server. Custom: the code is yours, maintained by your team or by the vendor that built it.
- 02
Where the Data Lives
Zapier hosts data on AWS in the United States and does not offer EU-only storage. Make offers an EU or North America region on AWS. n8n Cloud stores data in Frankfurt; self-hosted, wherever you choose. Custom: wherever you deploy it, including your own server.
- 03
How Usage Is Counted (and How Cost Scales)
n8n charges per full workflow execution; the free self-hosted edition has no execution quota (your server sets the limit). Make charges per credit, one per module action. Zapier charges per task, for every action step; AI or code steps can use more. With long workflows and high volume, the gap multiplies.
- 04
AI Capabilities
All three ship agents: n8n has LangChain-based agent nodes and an agents module in testing; Make AI Agents is in beta; Zapier Agents is a separate add-on. They work for well-bounded agents. A custom agent adds what is usually missing: systematic evaluation, per-execution cost control and autonomy limits designed for your process.
- 05
Errors, Versions and Environments
n8n offers error workflows and automatic retries, but Git version control and separate environments (development and production) only come with its Business and Enterprise plans. Zapier includes Zap version control on its paid plans. Custom: Git, tests and staged deployments from day one.
- 06
Licence and Vendor Lock-In
n8n is not open source in the strict sense: it uses the Sustainable Use License ("fair-code"). It is free for your company's internal use and an agency can charge you to set it up, but you cannot resell it as a service for your clients to build their own workflows. Make and Zapier are SaaS: your workflows live on their platform. Custom: the code is yours.
Why
Same Workflow, Three Different Meters
Why the entry price does not tell you what you will pay.
A typical workflow: an order comes in, data is validated, the invoice is created, the CRM is updated and an email goes out. Ten steps including the trigger, 1,000 times a month. Using the definitions each platform publishes, n8n counts 1,000 executions (it fits in its 2,500-execution Starter plan), Make counts about 10,000 credits (it uses up the entry Core plan) and Zapier counts about 9,000 tasks (it needs a tier far above 750). This is our calculation based on their rules, not a real invoice: filters, AI steps or code steps change the count.
- 1,000 n8n
- ~10,000 Make
- ~9,000 Zapier
Key points
Who It Is For and Who It Is Not, Option by Option
None of them is the best in the abstract. It depends on the process, the data and who will maintain it.
- 01
Zapier
Yes: non-technical teams, few workflows and low volume, and well-known SaaS apps (it advertises 9,000+ connectors). No: if your data must stay in the EU, if your workflows have many steps and high volume, or if paying per task makes the bill spiral.
- 02
Make
Yes: marketing or operations teams that want a powerful visual editor with branches and loops, a low entry price and an EU region. No: if you want to self-host, if your per-step volume is high or if you need to version workflows in Git like the rest of your software.
- 03
n8n
Yes: companies with some technical capacity or a partner to maintain it, who want their data on their own server or in Frankfurt, to pay per execution and to add code when needed. It is the one we recommend for B2B. No: if nobody will maintain the server, or if you want to offer your clients a workflow builder inside your product (that requires a commercial licence).
- 04
Custom AI Agent
Yes: processes where AI makes decisions (classifying, drafting, resolving) with your data, with in-house systems that have no connector, sensitive or regulated data, high volume and a need to measure quality. No: if the process is fixed rules between SaaS apps; there a platform costs a fraction and is live in weeks.
Who it is for
Should You Build a Custom Agent Instead of Using n8n or Make?
If you tick two or more boxes on the left, it is worth a conversation. If you recognise yourself on the right, start with a platform.
Who it's for
- Complex business logic: exceptions, rules that change per client, decisions a person makes today.
- Sensitive data (health, finance, personal data at scale) that cannot pass through a third party's cloud.
- High volume, where paying per step or per task costs more than your own server.
- In-house systems: a legacy ERP, an internal database or an API with no connector on any platform.
- You need to measure the quality of every AI answer with evaluations and traceability, not just that the workflow "runs".
Who it's not for
- The process is fixed rules between SaaS apps that already have connectors.
- You move a few hundred or a few thousand records a month.
- You do not yet know which process to automate or what it costs you today.
- There is no budget to maintain anything: a managed SaaS beats an abandoned in-house system.
Investment
What It Costs to Do It With Kiwop
Only prices we already publish on our service pages.
Implementing an automation with n8n or Make costs €3,000-15,000 per process, depending on its complexity. Custom agents have the tiers below. Prices exclude VAT.
- Automation with n8n or Make
€3,000-15,000 per process
Self-hosted n8n on European servers or Make, depending on the case. Optional maintenance: €500-1,000/month.
- Self-hosted n8n with a fixed server cost of about €50/month and no execution limit
- Documented, exportable workflows with no vendor lock-in
- Optional managed maintenance: monitoring, fixes and adjustments
- First automation running in 2-4 weeks
- Single Agent in Production Most Common
€25,000-50,000
One agent with 2-3 tools on one of your systems. 4-6 weeks to production.
- Design of the goal, the tools and the autonomy limits
- Prototype validated with real cases from your business
- Integration with one core system (CRM, ERP, database or email)
- Human-in-the-loop on high-impact actions
- Deployment, monitoring and 30 days of support
- Agent With ML or Several Systems
€50,000-100,000
Classification or predictive analysis, and several systems connected to the same agent. Full projects: 2-4 months.
- Everything in the single agent
- Classification and predictive analysis on your data
- Multiple integrations: CRM, ERP, email and data warehouse
- Controlled cost per execution, with budgets and alerts
Enterprise multi-agent systems range from €100,000 to €500,000. No lock-in contract. Prices exclude VAT.
FAQ
FAQ: n8n, Make, Zapier or a Custom Agent
What people ask us before choosing.
Which is cheaper: n8n, Make or Zapier?
It depends on your workflow, because each one counts differently. Entry plans (25-Sep-2026): n8n Starter, €20/month billed annually for 2,500 executions; Make Core, $9/month for 10,000 credits; Zapier Professional, from $19.99/month billed annually for 750 tasks. Because n8n counts the whole workflow while Make and Zapier count steps, n8n is usually cheaper for long workflows with volume. For 2-3 step workflows at low volume, the difference is small.
Is n8n free? Is it open source?
The self-hosted Community edition is free for your company's internal use, but it is not open source by the OSI definition: it uses the Sustainable Use License, which n8n calls "fair-code". You may use and modify it for your business; you may not offer n8n as a service for others to build their workflows. Free is not zero cost: you pay for the server (about €50/month in our implementations) and someone has to maintain it. To learn more about the tool, see our guide on how n8n works (in Spanish).
Can an agency implement self-hosted n8n for me and charge for it?
Yes. n8n's licence FAQ allows charging for building, setting up and maintaining workflows, and for training and consulting, as long as your clients cannot create or edit workflows on that instance. That is what we do in process automation: n8n on European servers and documented workflows, for €3,000-15,000 per process (breakdown in How much does an AI project cost?) and, if you want, managed maintenance for €500-1,000/month.
Where is my data stored on each platform?
Zapier: on AWS in the United States; its own FAQ says it does not offer EU-only storage (it relies on the EU-US Data Privacy Framework). Make: EU or North America region on AWS. n8n Cloud: Frankfurt. Self-hosted n8n or a custom agent: wherever you decide. If you handle health or financial data, this usually settles the choice before price does.
Can n8n, Make or Zapier build AI agents?
Yes, and they work well for bounded agents. n8n has LangChain-based agent nodes, evaluations to compare answers against test cases, and an agents module in testing. Make has Make AI Agents (beta) on all plans, with its AI provider or your own key. Zapier sells Agents as an add-on (1,500 activities/month for $33.33/month billed annually). Where they fall short is continuous evaluation, fine-grained cost control and complex logic: that is where a custom agent comes in.
When is it worth paying €25,000 for a custom agent?
When the process it automates is worth well over that per year and a platform cannot handle it: decisions based on your data, systems with no connector, data that cannot leave your perimeter or volume that makes per-step pricing expensive. Our entry tier is €25,000-50,000: a first agent with 2-3 tools, in production in 4-6 weeks; full projects take 2-4 months. If €50/month of n8n solves it, we will tell you.
Can I start with n8n and move to a custom agent later?
Yes, and it is often the sensible route. The n8n workflow forces you to document the process and gives you real data on volume and exceptions. If one step later needs reasoning, you do not have to throw anything away: the workflow calls the agent via webhook only at that step. That is why n8n is part of our agent stack.
Who maintains each option when something breaks?
With Zapier, Make and n8n Cloud, the vendor maintains the platform and you maintain your workflows: if an API changes, the workflow fails and someone has to fix it. With self-hosted n8n, you also handle server updates and backups. With a custom agent, the team that built it or yours. We cover it with managed maintenance (€500-1,000/month for automation) or with support and monitoring for agents.
Sources
Sources for the Figures on This Page
Official pages of each vendor and Kiwop's published prices. Captured on 25 September 2026. Make and Zapier prices are shown in US dollars and exclude taxes.
- n8n: pricing (n8n.io/pricing) Captured 25-Sep-2026. Starter €20/month (annual) or €24 (monthly), 2,500 executions; Pro €50/month (annual) or €60 (monthly), 10,000 executions; Business €667/month (annual), 40,000 executions, self-hosted only; definition of an execution; Git and environments only on Business and Enterprise; cloud data in Frankfurt.
- n8n: Community licence FAQ (Sustainable Use License) Captured 25-Sep-2026. What the licence allows and forbids: internal use, charging for building and maintaining workflows, no offering n8n as a service.
- n8n: Sustainable Use License text (GitHub) Captured 25-Sep-2026. Use limitations and ".ee." files under the Enterprise licence.
- n8n: documentation on agents, evaluations and error handling Captured 25-Sep-2026. Agents in testing (preview); evaluations at docs.n8n.io/build/integrate-ai/test-and-improve-ai-workflows; error workflows at docs.n8n.io/build/flow-logic/handle-errors-gracefully.
- Make: pricing (make.com/en/pricing) Captured 25-Sep-2026. Free 1,000 credits; Core $9/month, 10,000 credits (the site does not say whether this is monthly or annual billing; annual saves 15% or more); 1 credit per module action; Make AI Agents (beta) on all plans; AWS hosting in the EU or North America.
- Make: Make AI Agents help centre Captured 25-Sep-2026. Agents with instructions, knowledge and tools (modules, scenarios and MCP).
- Zapier: pricing (zapier.com/pricing) Captured 25-Sep-2026. Free 100 tasks; Professional from $19.99/month (750 tasks); Team from $69/month; triggers do not count as tasks; 9,000+ apps; version control on paid plans; Agents Pro $33.33/month billed annually (1,500 activities).
- Zapier: data privacy overview Captured 25-Sep-2026. "Zapier hosts data in AWS servers located in the United States"; EU-only storage is not offered.
- Kiwop: RPA process automation Published prices: self-hosted n8n with a fixed cost of about €50/month and no execution limit; managed maintenance €500-1,000/month; first automation in 2-4 weeks.
- Kiwop: how much does an AI project cost? Published price: AI process automation, €3,000-15,000 per process, with self-hosted n8n or Python.
- Kiwop: AI agent development Published prices: single agent €25,000-50,000 (first agent in production in 4-6 weeks; full projects, 2-4 months); with ML or several systems €50,000-100,000; enterprise multi-agent €100,000-500,000.
- "Same workflow, three meters" example: our own calculation Applies the published definitions of execution (n8n), credit (Make) and task (Zapier) to a hypothetical 10-step workflow run 1,000 times a month. Not a real invoice.
Next step
Tell Us About the Process and We Will Tell You Which Tool Fits
We implement n8n and Make and we build custom agents, so we have no reason to push you towards the expensive option. If a €50/month workflow solves it, we will tell you on the first call.
- No commitment
- Response in 24h
- Custom proposal
Let's talk.
Initial technical consultation
AI, security and performance. Diagnosis with phased proposal.
- NDA available
- Response <24h
- Phased proposal
Your first meeting is with a Solutions Architect, not a salesperson.
Request diagnosis