The three platforms in brief
Zapier, Make, and n8n are the three most often chosen automation platforms for SMBs. Each solves the same problem (connect app A to app B and automate a process) but takes a different approach. Zapier bet on the largest catalog of integrations (more than 5000 apps) plus the simplest UI for a non-technical user. Make (formerly Integromat) picked the middle path: a visual editor with full branching and iterators, around 1500 integrations, mid-range pricing. n8n is open source and self-hostable: about 500 native nodes plus a universal HTTP node that connects to any REST API, the lowest long-term cost, but it requires a developer or the willingness to learn.
The real boundary between the platforms is not the number of integrations but the way logic is defined. Zapier thinks in trigger plus action terms: a linear chain of steps where each step returns data to the next. Make thinks in scenario terms: a graph where you can have branching, parallel branches, iterators over lists. n8n thinks in workflow terms with full flow control: nodes can branch, merge, loop back, and you can write custom JavaScript in a Function node. These differences decide what you can build without touching code, and what requires a workaround or a plan upgrade.
For a business decision three more things matter alongside logic: pricing (Zapier prices by tasks/mo, Make by operations, n8n self-host by workflow executions but the infrastructure runs on your own VPS), control over data (Zapier and Make are SaaS where data flows through their servers, while n8n self-host never lets data leave your VPS), and learning curve (Zapier takes 2-4 hours to build a first working workflow, Make 8-16 hours, n8n 16-40 hours if you do not know Docker plus reverse proxy plus webhooks).
Pricing on a three-year horizon
A surface comparison of pricing pages does not show TCO (total cost of ownership). The real cost depends on the number of workflows, frequency of executions, tasks or operations per workflow, and whether premium-only features come into play. We benchmarked three SMB scenarios: 5 light workflows (each 100 executions/mo), 20 medium workflows (each 500 executions/mo), 50 plus heavy workflows (each 2000 executions/mo).
Scenario 1 (5 workflows, 500 executions/mo total): Zapier Starter 30 dollars/mo (750 tasks, barely enough), Make Core 12 dollars/mo (10k operations, plenty of headroom), n8n self-host on a Hetzner CPX21 12 euros/mo (zero hard limit). Three-year cost: Zapier 1080 dollars, Make 432 dollars, n8n 432 euros (around 460 dollars). Make and n8n very close, Zapier pays for UI simplicity.
Scenario 2 (20 workflows, 10000 executions/mo): Zapier Professional 50 dollars/mo (2000 tasks, but 20 workflows by 500 equals 10000 tasks which forces an upgrade to Team at 175 dollars/mo). Make Pro 29 dollars/mo (10k operations) or Teams 99 dollars/mo (50k operations). n8n self-host unchanged at 12 euros/mo. Three-year cost: Zapier Team 6300 dollars, Make Pro 1044 dollars, n8n 460 dollars. Zapier visibly the most expensive, n8n visibly the cheapest.
Scenario 3 (50 plus workflows, 100000 plus executions/mo): Zapier requires Business at 600 plus dollars/mo, Make Enterprise at 299 plus dollars/mo, n8n self-host still 12-30 euros/mo (vertical scaling of the VPS if execution time grows). Three-year cost: Zapier at minimum 21600 dollars, Make 10764 dollars, n8n 720-1080 euros. Here the difference is a five-digit number.
Conclusion: for SMBs starting with 5-10 workflows Zapier and Make are similar on cost, n8n requires extra DevOps work. Above 20 workflows n8n becomes economically dominant, but you still need someone to maintain the infrastructure. We recommend a hybrid: Make for 1-10 workflows (the client edits them), n8n self-host for 20 plus workflows (we maintain it under an 800 PLN/mo retainer). More on retainers in B2B customer service automation.
Functionality: branching, loops, error handling
Branching is the foundation of any non-trivial workflow. Example: an invoice arrives by email, we parse it with OCR, if the amount is below 1000 PLN we approve automatically, if above it goes to a manual approval queue. That is a simple conditional split. Zapier offers Paths (branching) only from Professional at 50 plus dollars/mo, where you define a maximum of 3 paths per level. Make has routers natively from the first paid plan, unlimited branching, every branch independent. n8n has an IF node plus a Switch node, branching and merging in both directions, with no cap on the number of branches.
Loops (iterations) are the second key mechanism. Example: pull a list of 100 clients from a CRM, for each one check whether they have an overdue invoice, if so send a reminder. Zapier’s Sub-Zaps or Looping by Zapier costs extra (tasks count times iterations, costs grow fast). Make has Iterator and Array Aggregator natively, each iteration counts as one operation. n8n has Split In Batches plus Loop nodes, each iteration is a single execution, with no per-iteration upcharge.
Error handling is the area where the difference is largest. Zapier from Starter gives Auto-retry plus error-halts-the-zap (Halt). Make has Error Handlers per module with configurable retry, ignore, rollback. n8n has Error Workflow (a separate workflow triggered when the main one fails), retry policy per node, a dead letter queue into a database. For business-critical workflows (invoice dispatch, accounting integration) error handling decides reliability.
The practical consequence for SMBs: without a sensible error handler the first serious incident (an external API returning 503 for an hour, a network blip, a rate limit) leaves the workflow in a half-done state. The client got a confirmation that the invoice arrived, but the CRM did not update, and accounting received a duplicate notification. A full audit in n8n lets you reconstruct state and repair manually. In Zapier and Make recovery requires a manual replay from the UI (only on certain plans), and in Zapier Starter often means fully manual repair on production.
AI nodes, which is strongest in 2026
By 2026 all three platforms have native AI nodes, but they differ in integration depth. n8n has the most developed AI nodes: a dedicated AI Agent node with LangChain integration, chain nodes (RAG, conversational retrieval, agent toolkit), memory nodes (buffer, vector store, postgres), Tools agent with function calling, output parsers, splitter nodes for embedded documents. You can build a full RAG pipeline in 10-15 nodes without external code.
Make has basic AI nodes: OpenAI, Anthropic (Claude), Cohere, Mistral, plus generic HTTP to any API. There is no built-in vector store, no agent abstraction with multi-turn memory, no ready-made chains. Building your own assistant requires gluing 10-20 modules from HTTP plus your own SQL/Postgres for memory. For simple use cases (email classification, content generation, transcription) Make is sufficient; for stateful agents it requires workarounds.
Zapier has integrations with OpenAI, Anthropic, Hugging Face, plus its own Zapier AI layer (Zapier Agents, Copilot). Functionally similar to Make, with custom prompts under less control (Zapier wraps its own template around your prompt, changing model behavior). Limited control over temperature, max tokens, system prompt structure. Good for 80 percent of common use cases, weak for advanced ones. More on building agents in our article on Claude Agent SDK.
For advanced AI projects where you need full control over the system prompt, costs, caching, and tool use, none of the three platforms suffices on its own. The hybrid pattern: an agentic orchestrator in Claude Agent SDK handles conversation and logic, n8n handles integrations with client systems (CRM update, calendar, mailing), communication over webhooks or an MCP server. We cover this pattern in our MCP article. Our AI Assistant V0.1 (June 2026 launch) uses exactly this stack: SDK plus n8n plus MCP, with no Make or Zapier.
Self-host versus cloud, security plus GDPR
n8n self-host is the only option where data never leaves your infrastructure. The workflow runs on your VPS (Hetzner, OVH, your own server in the office), credentials are stored encrypted in a local database (Postgres/SQLite), and no API call goes through n8n.io’s servers. That matters for DACH clients (DSGVO compliance) and for Polish clients in regulated industries (medicine, finance, law firms).
Zapier and Make are SaaS where data flows through their servers (US for Zapier, EU for Make), with credentials stored in their cloud. Both offer DPA (Data Processing Agreement) and claim GDPR compliance, but the actual data path is: your system to their server (parsing, branching, transformation) to the target. For every workflow that touches PII (client data, invoices, contracts) you need a signed DPA and alignment with the client’s policy.
An additional self-host aspect: version control and audit. n8n keeps workflows in a database, and you can export them to JSON and version them in git. You can have staging and production environments, run code review on changes before deploy, and have a full audit log of who, when, and which workflow changed. Zapier and Make have change history, but no native git workflow and no staging environment.
A third aspect: secret management. n8n self-host lets you plug in Vault, AWS Secrets Manager, or any external secret store. API keys for Stripe, Gmail, or a CRM are never stored in plaintext, and key rotation is a first-class operational process. Zapier and Make keep credentials in their cloud, and the client must accept external trust. For firms in regulated sectors (banks, insurance, healthcare providers) this detail is often a hard reason to rule out SaaS on compliance and risk grounds.
Integrations, does it have what I need
Integration count is the first filter but rarely the final one. Zapier advertises more than 5000 apps, Make around 1500, n8n around 500 native. In real terms for SMBs in Poland in 2026 the must-have list is 30-50 integrations: Gmail, Outlook, Google Calendar, Microsoft 365, Slack, WhatsApp, Telegram, Trello, Asana, ClickUp, Notion, Airtable, Google Sheets, Excel Online, Stripe, PayPal, Tpay, Przelewy24, Allegro, Shopify, WooCommerce, MailChimp, Active Campaign, HubSpot, Pipedrive, Salesforce, Twilio, OpenAI, Anthropic, Claude, ChatGPT, plus 10-15 Poland-specific ones (Fakturownia, iFirma, Comarch Optima, wfirma, GUS API, NIP lookup, couriers InPost, DPD, DHL). All three platforms cover this list.
The difference shows up in niche integrations: a client’s proprietary API, a German accounting system (DATEV, Lexware), a French ERP (Cegid, Sage), a Poland-specific CRM. Here Zapier and Make either have it or do not (then you wait for them to add it), n8n has a universal HTTP node that connects to any REST API plus the option of writing custom community nodes. In practice: for 80 percent of clients all three platforms are enough, for 15 percent of clients n8n wins thanks to its HTTP fallback, for 5 percent of clients none is enough and you must write a custom integration (we deliver that under the automation package at 3000 PLN per workflow).
Custom code nodes are the second mechanism for closing integration gaps. Make has Custom function (inline JavaScript, up to 100 lines of code), n8n has Function node plus Code node (full Python or JavaScript with no cap). Zapier has Code by Zapier (Python/JavaScript), with runtime limits (10 seconds max, no npm packages). For complicated transformations n8n wins, for simple ones Make is enough, Zapier is the last option.
Finally: testing integrations in a staging environment. n8n self-host lets you run a second instance on a separate VPS with a copy of workflows and sandbox credentials, so the client can safely test changes before deploying to production. Make and Zapier have no native staging, the client has to create a duplicate workflow with “_test” in the name, which does not eliminate the risk of a production mistake.
Recommendation per scenario
SMB below 10 workflows plus a non-technical user who wants to edit themselves: Make. The most approachable UI for non-developers, predictable pricing, basic AI nodes are enough. We build the workflows, the client maintains them, and when needed we modify them at an ad-hoc fee.
SMB above 10 workflows plus a tech-savvy team: n8n self-host. Hosting on Hetzner CPX21 (12 euros/mo), we configure the infrastructure and the first workflows under the 12000 PLN implementation package, retainer 800 PLN/mo for maintenance plus ad-hoc changes. The value materializes once the client scales past 20 plus workflows and Zapier/Make start to cost 200 plus dollars/mo.
Enterprise plus compliance-heavy (medical, finance, legal): n8n self-host is mandatory. There is no other option for sectors where data must stay inside the client’s infrastructure. Hosting on-prem or a dedicated VPS in the EU. Full audit log, MFA, SSO integration, role-based access control.
Quick prototype plus no developer available: Zapier. Expensive long term, but 2-4 hours from idea to a working workflow. It makes sense when you need to validate something fast before investing in a solid solution. For AI Assistant V0.1 launching in June 2026 we build on Claude Agent SDK plus n8n hybrid: SDK for the conversation orchestrator, n8n for integrations with client systems. We discuss pricing through the contact form.
Anti-patterns to avoid: do not build 100 workflows in Zapier “because it is fast” and then ask for a migration (the migration cost is often higher than starting with n8n). Do not try to self-host n8n without someone who can maintain the infrastructure (every outage is a manual repair, the client sees downtime). Do not pick Make for workflows that need strong AI control (custom system prompt, prompt caching, multi-turn memory): you will hit a wall after 2-3 months of iteration. A good starting point is an AI audit at 1500 PLN: 4-8 hours of discovery, a process map, a platform recommendation per use case, a roadmap PDF with prioritization. We deduct the audit cost from the implementation package if we go further.
FAQ
Is self-hosting n8n hard?
For someone who knows Docker compose plus a reverse proxy plus basic Linux, installation takes an hour. For a non-technical person it is a barrier that cannot be crossed unaided. We install, configure SSL, backups, and monitoring under the implementation package. Ongoing maintenance is then 1-2 hours/mo (updates, log review, debugging the occasional error).
How big is vendor lock-in on cloud platforms?
Zapier and Make let you export workflows as JSON, but the format is proprietary and not portable between platforms. Migrating Zapier to Make or to n8n requires a manual reconstruction of every workflow (one-to-one module mapping rarely works). For 5-10 workflows that is 8-20 hours of work, for 50 workflows it is 40-80 hours. We offer a migration package at 12000-18000 PLN for 20-50 workflows.
What about a workflow running 1000 times a day?
n8n self-host has no native limit, the only constraint is CPU/RAM on the VPS. Make Pro at 29 dollars/mo gives 10000 operations (333/day), Teams at 99 dollars/mo gives 50000 operations (1666/day). Zapier Professional at 50 dollars/mo gives 2000 tasks/mo (66/day), and to handle 1000 executions a day you need Team at 175 dollars or Business at 600 dollars. For high volume n8n is the only sensible option.
Can I migrate from Zapier to n8n?
Yes, but it requires manual reconstruction. There is no automatic migration tool between platforms (the workflow JSON format is proprietary and structurally different). We do this for 12000-18000 PLN for 20-50 workflows, the price including an audit of the current Zaps plus a rebuild in n8n plus testing plus documentation plus team training. The financial payoff shows up at 20 plus workflows (return on investment in 12-18 months).