There are four paths to running workflow automation in 2026, and the honest cost gap between them at SME scale is now 10–20×. Zapier sits at one end, Cloudflare Workers and self-hosted n8n at the other, with Make and n8n Cloud in the middle. The choice isn't ideological — it's a function of monthly task volume, in-house technical capacity, and how much you value not running infrastructure.
Most comparison posts on this topic are written by one of the vendors and read accordingly. This one walks through what 100,000 monthly operations actually costs across all four options, with the migration math and the cases where each one still wins. It's a spoke off our broader SaaS replacement playbook — same logic, applied to the automation layer specifically.
TL;DR — At 100K monthly tasks: Zapier ≈ $599/mo, Make Pro ≈ $40-80/mo, n8n Cloud Pro ≈ $50-100/mo, self-hosted n8n ≈ $5/mo, custom Cloudflare Worker ≈ $5/mo. Pick by team shape (technical capacity + workflow stability), not by ideology. Live cost data: /data/automation-platform-pricing.
Key Takeaways
- At 100K monthly tasks/operations: Zapier ≈ $599+/mo (Company tier), Make Pro ≈ $40-80/mo, n8n Cloud Pro ≈ $50-100/mo, self-hosted n8n ≈ $5/mo, custom Cloudflare Worker ≈ $5/mo (Zapier Pricing, Make Pricing, n8n Pricing, Cloudflare Workers Pricing, retrieved 2026-05-23)
- Self-hosted n8n on a Hetzner CX22 (€4.51/mo) gives unlimited executions and 1,000+ pre-built integrations — feature-parity with Zapier for most SME use cases
- The decision isn't "which is best" — it's a function of monthly ops volume, technical capacity, and operational tolerance for managing infrastructure
- SMB SaaS spend per employee grew 21% YoY to $4,830 in 2025; workflow automation is one of the few categories where self-hosting genuinely competes on capability, not just price (Zylo 2025 SaaS Management Index)
Why does this comparison keep coming up in 2026?
Zapier's pricing has crept upward — Professional is now $49/mo for 2,000 tasks, Team $69/mo, Company $103.50/mo annually (Zapier Pricing, 2026). Meanwhile n8n has crossed 96K GitHub stars and ships 1,000+ pre-built integrations matching Zapier's catalog on the workflows most SMEs run (n8n GitHub). Make has held steady with a more transparent ops-based model (Make Pricing).
<!-- [UNIQUE INSIGHT] --> What we see on client calls in 2026 isn't founders asking "which is cheapest" — it's founders realizing that at 30K+ monthly ops, the math tips. A workflow that justified Zapier Team last year now sits inside Make Pro or self-hosted n8n with no functional loss.What does each tool actually do — and what does it cost?
All four execute the same core job: receive a trigger (webhook, schedule, polling), run logic, call APIs, write to destinations. They diverge on pricing model, hosting, and extensibility.
| Tool | Hosting | Pricing model | 2026 entry price | Catalog size |
|---|---|---|---|---|
| Zapier | Vendor cloud only | Per-task ("Tasks") | $19.99/mo (750 tasks, Starter) | 7,000+ apps |
| Make | Vendor cloud only | Per-operation ("Ops") | $9/mo (10K ops, Core) | 1,800+ apps |
| n8n Cloud | n8n-managed cloud | Per-execution + active workflow | €20/mo (2.5K executions, Starter) | 1,000+ nodes |
| n8n self-host | Your infrastructure | Hosting only (no per-use fee) | $5/mo (Hetzner CX22) | 1,000+ nodes |
| Custom code (e.g. Workers) | Your infrastructure | Compute time | $5/mo (Cloudflare Workers Paid) | Limited only by the APIs you write to |
Across the dozen-plus n8n and custom-code builds we've shipped, the average client uses 8–14 distinct integrations. The 7,000-app catalog is genuine, but the long tail rarely shows up in workflows that justify a paid plan.
How much do Zapier, n8n, and Make really cost at scale?
Pricing scales differently across the four. Here's what 100,000 monthly tasks/operations actually costs in 2026, billed annually where applicable:
| Tool | Tier at 100K ops/mo | Monthly cost | Annual cost |
|---|---|---|---|
| Zapier | Company (50K-150K tasks) | $103.50/mo (annual) up to ~$599 at higher Company tiers | $1,242–$7,188 |
| Make | Pro / Teams scaling — typically Make Teams or Enterprise | $40–$80/mo at most usage profiles | $480–$960 |
| n8n Cloud | Pro (15K executions, multi-workflow) | €50/mo (~$54) — bigger workflows push to Business | $648+ |
| n8n self-host | Hetzner CX22 + Postgres in Docker | €4.51/mo (~$5) | $60 |
| Cloudflare Worker (custom) | Workers Paid plan, 10M requests/mo included | $5/mo + $0.30 per million extra | $60 |
Two caveats. "Tasks" and "operations" aren't 1:1 — a single Make scenario can consume 5–15 ops depending on filters and iterators, while a Zapier Zap typically consumes 1 task per step. A 50K-task Zapier workflow is often 100K–150K ops in Make. Second, Zapier discounts annual billing by ~33%, which matters at higher tiers.
At 5–10K ops/mo, all four are within $50/mo of each other and the choice is about workflow complexity, not price. Above 30K ops/mo, the gap widens to 5–20× toward self-hosted or custom code.
What does the custom-code option actually look like?
A typical custom replacement is 800–2,000 lines of TypeScript or Python: a webhook receiver, a queue (BullMQ on Redis or a cloud equivalent), the workflow logic, retry handlers, and a thin dashboard if non-technical operators need to manage it. The codebase stays small because the abstractions are thin — HTTP in, HTTP out, logic between.
| Layer | What it does | Build effort | Hosting cost |
|---|---|---|---|
| Webhook receiver | Listens for triggers, validates, enqueues | 4–8 hrs | $0 (Cloudflare Workers, Vercel free) |
| Queue + dispatcher | Reliability layer — retries, dead-letter, observability | 8–16 hrs | $0–$10 (Redis on Upstash free, or Postgres LISTEN/NOTIFY) |
| Workflow logic | Branches, conditions, API calls | 8–24 hrs | scales with compute |
| Operator dashboard (optional) | View runs, retry failed, edit config | 8–16 hrs | $0 (deploy with the API) |
| Total | 30–60 hrs / $3,000–$8,000 build | $5–$25/month |
Break-even versus Zapier Team ($69/mo) sits at month 36–55 depending on build cost; versus Zapier Company at the $200+ tiers, it drops to month 15–30. The ongoing operational load is modest — 1–3 hours per month for monitoring, dependency updates, and the occasional API-schema change.
When does Zapier still win?
At under 5,000 monthly tasks, Zapier Starter ($19.99/mo for 750 tasks, Professional $49/mo for 2,000) is hard to beat on time-to-ship (Zapier Pricing, retrieved 2026-05-23). Three cases where Zapier remains the right answer even after the cost math runs against it:
- Non-technical ops where the operator cannot read code at all. Zapier's UI is best-in-class for shipping a new workflow on a Friday without a developer. Make is close but steeper. n8n is more developer-shaped. Custom code is out.
- Under 5,000 monthly tasks with high integration variance. Touching 30+ apps at low volume (agency-side client work, for example), the Zapier catalog is unmatched and Starter stays under $50/mo.
- Regulated workflows where vendor SOC 2 / HIPAA certifications are the buying signal. Self-hosting introduces compliance work that isn't worth it for a single workflow.
When is n8n (self-hosted) the right call?
n8n self-hosted wins above 20K ops/month with a technical operator who can manage a Docker container and a preference not to be in a vendor's pricing roadmap. The EU-hosting angle also matters for GDPR-sensitive workflows — n8n on Hetzner Helsinki keeps data on EU infrastructure end-to-end.
Downside: you own uptime, version upgrades, backups. One workflow = an hour a month. Forty workflows in production = 4–6 hours a month, which is real labor.
When does Make beat both?
Make wins the middle. At 10K–80K monthly operations, Make Pro at $16/mo (10K ops) or Teams at $29/mo (10K ops with collaboration) undercuts Zapier's equivalent task tiers by 50–70% (Make Pricing, retrieved 2026-05-23). The visual workflow design genuinely scales to complex branching without writing code, with a slightly steeper UI than Zapier.
Make's specific strength is iterators and aggregators — looping through arrays of items with collection-level transformations is cleaner in Make than in either Zapier or n8n. Marketing ops teams running drip campaigns with per-recipient personalization sit right in Make's sweet spot.
The constraint is the same as Zapier's: no self-host option, no escape hatch when pricing changes.
When is custom code the right answer?
At 100K+ monthly ops, a $5/mo Cloudflare Worker handling the same workload as a $599/mo Zapier Company plan represents a 120× cost gap (Cloudflare Workers Pricing, retrieved 2026-05-23). Custom code earns its keep on one of three conditions: that predictable high-volume scenario; workflow logic that doesn't map cleanly to drag-and-drop (multi-step LLM orchestration, complex state machines, database transactions across steps); or strategic workflows where vendor lock-in is catastrophic (billing, contracts, onboarding).
Across the NodeSparks engagements that ended in custom code, the trigger was rarely "save money on Zapier." It was a specific workflow — usually AI-orchestration heavy — that the vendor platforms couldn't model cleanly. The Apollo replacement we built and the Bill.com replacement running in Slack both followed this pattern. The Zapier replacement was a side effect.
How do migration paths between them compare?
Zapier → Make is typically a 1-week project for 10–20 Zaps — similar mental model, vendor-hosted on both ends. Zapier → n8n Cloud is comparable. Zapier → self-hosted n8n adds 1–2 weeks of infrastructure setup. Zapier → custom code is the longest path and the only one that eliminates vendor-pricing risk entirely.
| Migration | Typical timeline (20–40 active workflows) | Risk profile |
|---|---|---|
| Zapier → Make | 1–2 weeks | Low — similar mental model, parallel run easy |
| Zapier → n8n Cloud | 1–3 weeks | Low — workflow logic ports cleanly |
| Zapier → n8n self-host | 3–6 weeks | Medium — adds DevOps surface area |
| Zapier → custom code | 4–8 weeks | Medium-high — net new system, but eliminates vendor lock |
Standard playbook on any of them: export the Zap specs to a written doc, rebuild on the target platform, parallel-run for 2–4 weeks in dry-run mode, then cut over workflow-by-workflow. Don't big-bang it.
The honest verdict, by company shape
| Company shape | Recommendation |
|---|---|
| 0–5K ops/mo, non-technical | Zapier Starter ($19.99/mo) |
| 5–30K ops/mo, marketing-led | Make Core or Pro ($9–$16/mo) |
| 30–100K ops/mo, technical operator on staff | n8n self-hosted on Hetzner (~$5/mo) |
| 30–100K ops/mo, no technical capacity | n8n Cloud Pro or Make Teams ($30–$54/mo) |
| 100K+ ops/mo OR workflow doesn't map to drag-and-drop | Custom code on Cloudflare Workers / Hetzner (~$5/mo + build) |
| Regulated workflows (HIPAA, finance) | Zapier or Make on enterprise tiers |
None of these is "the right answer" universally. The right answer is whichever lets you ship the workflow without paying a vendor markup you can't justify, and without taking on operational load you can't sustain. The build-vs-buy matrix walks through the broader version of this decision across the SaaS stack, not just automation.
If your monthly Zapier bill is under $50, the math doesn't favor changing anything. If it's over $300/mo, you're paying a tax that a one-time $4,000 build amortizes in under 14 months. That's the decision worth running the numbers on.
For every monthly cost cited above, the live dataset is at Automation Platform Pricing 2026 — updated quarterly, free under CC BY 4.0. The underlying decision tool is the SaaS Replacement Matrix; the short-form is the 5-Question Build-vs-Buy Rubric.
For more cost teardowns and replacement playbooks across the SaaS stack, see the full /blog archive.
Frequently asked questions
What is the cheapest of the four at 50,000 tasks per month?
Self-hosted n8n on a €4.51/mo Hetzner CX22 wins at $5/month for unlimited executions, followed by a Cloudflare Worker (free tier covers 100K requests/day). Make Pro at $39/month handles 30,000 ops and Make Teams at $29/month covers 10K — you'd need a higher tier to hit 50K. Zapier Professional at $49/mo only covers 2,000 tasks; at 50K you're on Company tier ($103+/mo) or higher. The cost gap between vendor and self-hosted is roughly 10–20× at this volume.
Does n8n self-hosted really replace Zapier at production scale?
For ETL, webhooks, scheduled jobs, and API orchestration: yes, with the caveat that you're now operationally responsible. The workflow engine itself is feature-parity for most Zapier use cases — 1,000+ pre-built integrations, branching, error handling, queueing via Redis. What changes is uptime is yours, version upgrades are yours, secrets management is yours. For an SME with a part-time technical operator, the operational load is modest (1–3 hrs/month). For non-technical teams, n8n Cloud Pro at €50/mo is the sane middle path.
When does Make beat both Zapier and n8n?
Make wins on three axes: cost-per-operation at mid-scale (10K–80K ops/mo), visual workflow design that non-engineers can edit, and complex data transformation without code. The operations pricing model is more honest than Zapier's task-based one — you pay for actual API calls, not abstract 'tasks.' Make's weakness is the same as Zapier's: vendor lock-in, no self-host option, and pricing that scales hard past 100K ops. Best fit: marketing ops teams running 20–80K ops/mo with no in-house engineering.
What's the realistic build cost to replace a Zapier setup with custom code?
A typical replacement is 800–2,000 lines of TypeScript or Python and ships in 20–60 engineering hours ($2,000–$6,000 at senior contractor rates). The work splits into webhook receivers (one day), the workflow logic (two to four days depending on branching), error handling and retries (two days), and a thin operator dashboard if non-technical staff need to manage it (two to four days). Hosting on Cloudflare Workers, Hetzner, or a free Vercel tier runs $0–$10/month. Break-even against Zapier Team ($69/mo) hits month 30–80 depending on build cost.
What about reliability — is custom code as stable as Zapier?
Zapier reports task failure rates around 0.1–0.5% under normal conditions and runs on infrastructure most SMEs can't match for raw uptime. A custom Cloudflare Worker or n8n self-host can match this if you build retry logic, dead-letter queues, and observability into the system — but you have to actually build them. Most failures we see in custom replacements come from missing error handling on third-party API timeouts, not from the platform layer. Budget two days of build time specifically for resilience or stay on managed.
Can I migrate off Zapier without breaking existing workflows?
Yes, with a parallel-run period. Export each Zap's logic to a written spec (Zapier supports JSON export on Team+ tiers), rebuild on the target platform, run both in parallel for two to four weeks with the new system in dry-run mode, then cut over Zap-by-Zap. Most migrations take three to six weeks for a typical SME with 20–40 active Zaps. The risky workflows are anything writing to billing, payroll, or contracts — keep those on the proven path until you have observability matching Zapier's logs.