How the billing models differ

Platform	Unit billed	Implication
Zapier	Task (per action step)	Multi-step Zaps multiply cost fast
Make	Operation/credit	Loops multiply credits
n8n Cloud	Execution (full workflow run)	Steps inside one run = one execution
n8n self-hosted	Server cost only	Unlimited executions; you ops the box

Example monthly cost at scale

Assume 10,000 workflow runs/month, average 5 steps each:

• Zapier: ~50,000 tasks → often $200–400+/mo on Team tiers
• Make: ~50,000 operations → $50–150/mo depending on plan
• n8n self-hosted on $12–24 VPS → ~$12–24/mo + your time

At low volume (<500 runs), all three can sit on free tiers — self-hosting is not worth the ops overhead.

Self-hosting is not free

Budget for:

• VPS ($6–30/mo) or homelab power
• Backups (restic/S3)
• SSL + reverse proxy (Caddy/Nginx)
• Upgrade testing when n8n releases breaking changes
• Monitoring when workflows fail at 2 a.m.

I run self-hosted n8n for content pipelines, SEO automation, and client integrations — the break-even vs Zapier happened around 2,000 executions/month.

When to choose each

• Zapier — fastest setup, largest app directory, low volume
• Make — visual complex flows, moderate volume
• n8n self-host — high volume, multi-step flows, data sovereignty
• n8n Cloud — n8n without ops, per-execution pricing

Production setup (docker-compose outline)

Typical stack on a $12–24/mo VPS (2 vCPU, 4GB RAM):

• Docker + docker-compose
• n8n + PostgreSQL (not SQLite at scale)
• Caddy or Nginx reverse proxy with Let’s Encrypt
• Daily restic backup to S3-compatible storage
• Uptime monitoring (Uptime Kuma or external ping)

First-time setup: 4–8 hours for someone comfortable with Linux. Ongoing: 1–3 hours/month for updates and failed workflow triage.

Five workflows we run in production

1. Lead routing — form submit → CRM + Slack + email
2. Content pipeline — draft notification → review webhook → publish hook
3. SEO monitoring — rank drop alert → task in project tool
4. Client reporting — GA4 API → Looker Studio refresh trigger
5. Backup verification — nightly check → alert if stale

Export workflows as JSON for disaster recovery.

Break-even calculator

Monthly executions	Zapier est.	Make est.	Self-hosted VPS
500	Free–$20	Free–$10	Overkill
2,000	$50–100	$20–50	~$18
10,000	$200–400	$50–150	~$18–30
50,000	$500+	$200+	~$30–50 (+ ops time)

Add $50–150/hour for your ops time when comparing.

Failure modes (honest)

• n8n upgrade breaks credential encryption → test in staging
• Webhook timeouts on cheap VPS → increase timeout, queue jobs
• OAuth token expiry on Google nodes → refresh automation
• No monitoring → silent failures for days

When NOT to self-host

• Team has zero Linux admin capacity
• Compliance requires vendor SOC2 and you cannot attest infra
• Volume under 1,000 runs/month

Architecture: production stack explained

Self-hosting is not “install and forget.” Here is the stack we run for Gurkha Technology client automation and internal SEO pipelines.

Internet → Cloudflare (DNS/proxy) → Caddy/Nginx (TLS) → n8n (Docker)
                                                      ↘ PostgreSQL
Backups: restic → S3-compatible bucket (daily)
Monitoring: Uptime Kuma + webhook alert to Slack/WhatsApp

Why PostgreSQL, not SQLite

SQLite works for experiments. Production workflows with concurrent executions, webhook bursts, and 50+ active workflows corrupt or lock SQLite under load. PostgreSQL adds ~512MB RAM overhead — worth it.

VPS sizing guide

Workload profile	vCPU	RAM	Disk	Est. USD/mo
Light (<2k runs/mo)	1–2	2GB	40GB	6–12
Agency default (2k–20k)	2	4GB	80GB	12–24
Heavy (20k+, many HTTP nodes)	4	8GB	160GB	24–48

Providers we have used: Hetzner, DigitalOcean, Vultr — pick Kathmandu/Singapore region for lower latency to Nepali APIs where available.

docker-compose essentials

• Pin n8n image version — do not track latest in production
• Separate .env for secrets; never commit credentials
• Set N8N_ENCRYPTION_KEY before first run — loss means re-entering all credentials
• Configure WEBHOOK_URL to public HTTPS domain
• Limit execution data retention (prune after 7–14 days) to control DB size

See GitHub Actions CI/CD for how we deploy static sites that receive n8n webhooks.

Detailed cost model (year 1)

Cost line	Self-hosted n8n	Zapier (mid volume)	Make (mid volume)
Platform fee (annual)	$0	$240 – $4,800	$120 – $1,800
VPS (annual)	$144 – $576	—	—
Backup storage	$12 – $60	Included	Included
Domain for webhooks	$0 – $15	—	—
Setup labor (one-time)	4–8 hrs	1–2 hrs	2–4 hrs
Ops labor (monthly)	1–3 hrs	~0	~0

Hidden Zapier tax: each extra step in a Zap is a billable task. A 5-step workflow at 10,000 runs = 50,000 tasks. n8n counts 10,000 executions.

Task math example

Scenario: Lead form → validate email → CRM create → Slack notify → Google Sheet log → conditional branch to assign owner (6 steps)

Runs/mo	Zapier tasks	Make ops (approx.)	n8n executions
1,000	6,000	6,000	1,000
5,000	30,000	30,000	5,000
20,000	120,000	120,000	20,000

At 20,000 runs, Zapier Team/Company tiers dominate TCO unless ops time is valued at zero.

Step-by-step self-host deployment

1. Provision VPS — Ubuntu LTS, enable firewall (22, 80, 443 only)
2. Install Docker + Compose
3. Create PostgreSQL + n8n services in compose file
4. Configure Caddy automatic HTTPS for n8n.yourdomain.com
5. Set encryption key + admin password — store in password manager
6. Create non-admin users per client or per workflow group (RBAC)
7. Import test workflow — HTTP webhook → Set node → Slack
8. Configure backups — nightly DB dump + restic
9. Set up monitoring — external ping every 5 min
10. Document runbook — upgrade procedure, restore procedure

First production workflow should be low-risk (internal notification), not client-facing payment sync.

Workflow patterns for marketing agencies

Client lead routing

Webhook from website form → dedupe by email → HubSpot/Sheets CRM → assign by city → Slack channel → delay → email auto-reply

Why n8n: branching and error handling without per-branch task billing.

SEO rank drop alert

Scheduled cron → pull rank API → filter drops >3 positions → create Asana task → optional LLM summary for client email draft (human sends)

Connect to automated SEO pipeline and AI automation stack.

E-commerce order ops (Nepal)

WooCommerce webhook → WhatsApp template to warehouse → courier API → update Sheet → failed payment alert

Local wallet callbacks sometimes duplicate — idempotency node required.

Reporting automation

Monthly cron → GA4 Data API → populate Looker Studio source sheet → email account manager “report ready”

Cheaper than Supermetrics at scale; more setup than Zapier’s GA connector.

Security hardening checklist

• HTTPS only; HSTS via Cloudflare or Caddy
• Basic auth or SSO in front of n8n UI
• Separate credentials per client workflow where possible
• Webhook paths use unguessable tokens
• Disable unused nodes/community nodes you do not audit
• Patch VPS monthly; test n8n upgrades on staging clone first
• Backup restore tested quarterly (not just “backups exist”)

Upgrade and migration runbook

1. Snapshot VPS
2. Export all workflows JSON to Git
3. Pull new n8n image on staging
4. Run test workflows including OAuth (Google, Meta)
5. Production upgrade in maintenance window
6. Monitor error queue for 24 hours

OAuth tokens break most often — budget re-auth time for Google Sheets and Ads nodes.

n8n Cloud vs self-host vs Zapier (decision matrix)

Criterion	Zapier	Make	n8n Cloud	n8n self-host
Time to first workflow	Minutes	Minutes	Minutes	Hours
Multi-step cost scaling	Poor	Moderate	Good	Excellent
Data residency control	Low	Low	Medium	High
Custom code (JS/Python)	Limited	Limited	Strong	Strong
Nepali SME budget fit	Low volume	Mid volume	Mid volume	High volume

Honest ops time journal (sample month)

Activity	Hours
Failed workflow triage	1.5
n8n patch + test	0.5
VPS security updates	0.5
New client workflow	2.0
Backup verify	0.25
Total	~4.75

At NPR 2,000/hour equivalent, add NPR 9,500/mo to true TCO — still beats Zapier at 10k+ multi-step runs.

Community nodes vs core nodes

Community nodes extend n8n (scrapers, niche APIs) but add supply-chain risk:

• Audit GitHub repo activity before install
• Pin version; disable auto-update on community pack
• Prefer HTTP Request node + documented API over obscure community node
• For client data, default to core nodes only unless security-reviewed

High-availability (when clients depend on it)

Single VPS is fine for most agency automation. Consider HA when:

• Lead routing for NPR 500k+/month ad spend client
• Payment-adjacent workflows (inventory sync, not card data)
• SLA in contract with financial penalty

Pattern: primary VPS + warm standby + DB replication — doubles cost, rarely needed before 15+ retainer clients.

Comparison with Make for visual builders

Make excels when non-developers build flows. Our team uses:

• Make: one-off client automations, marketing ops staff can edit
• n8n self-host: high-volume, branching, code nodes, cost-sensitive

Document who owns each platform — client offboarding nightmare when automations live in personal Make accounts.

Legal and client contract notes

• State in MSA who hosts automation server
• Data processing: lead PII flows through n8n — secure VPS, NDPA awareness
• Export JSON on contract end — same as GA4/GTM ownership in pricing guide

Staging instance (non-optional at scale)

Run n8n-staging. subdomain on smaller VPS or same box with separate DB:

• Test OAuth reconnect after Google password policy change
• Clone production workflow before editing live lead router
• Upgrade n8n image here first

Cost: +$6–12/mo or shared resources — cheaper than one hour of missed leads.

Docker-compose reference (minimal production)

Conceptual structure — adapt paths and secrets:

services:
  postgres:
    image: postgres:15
    environment:
      POSTGRES_PASSWORD: ${DB_PASSWORD}
    volumes:
      - pg_data:/var/lib/postgresql/data

  n8n:
    image: n8nio/n8n:1.72.0 # pin version
    environment:
      DB_TYPE: postgresdb
      DB_POSTGRESDB_HOST: postgres
      N8N_ENCRYPTION_KEY: ${N8N_ENCRYPTION_KEY}
      WEBHOOK_URL: https://n8n.example.com.np/
    depends_on:
      - postgres
    ports:
      - '127.0.0.1:5678:5678'

  caddy:
    image: caddy:2
    ports:
      - '443:443'
    volumes:
      - ./Caddyfile:/etc/caddy/Caddyfile

Run Postgres on same VPS until 20k+ executions/month; then consider managed DB.

Webhook reliability patterns

Nepali ISPs and mobile networks drop connections:

1. Retry with exponential backoff on HTTP Request nodes
2. Queue heavy jobs — split webhook ack from processing
3. Idempotency key in header or body — store in Postgres
4. Dead letter Slack channel for workflows that fail 3×

Wallet callbacks from eSewa/Khalti are the highest-stakes webhooks — log raw payload before parsing.

Monitoring and alerting stack

Tool	Purpose	Cost
Uptime Kuma	HTTP ping n8n + public webhook	Free self-host
n8n error workflow	Slack on execution failed	Free
Grafana + Loki	Log aggregation at scale	Free–low
PagerDuty	Client SLA with on-call	Paid

Minimum viable: Uptime Kuma + error workflow — 30 min setup.

Credential management

• One credential set per client for OAuth where possible
• Rotate after staff offboarding
• Never share production n8n login — use RBAC user per team member
• Store .env in password manager, not Slack

Client billing for automation work

Agencies undercharge automation setup. Reference digital marketing pricing:

Deliverable	NPR one-time
Lead router workflow	15,000 – 30,000
Report automation	20,000 – 40,000
E-commerce alert pack	25,000 – 50,000
Monthly monitoring retainer	5,000 – 15,000/mo

Explain ROI in hours saved — finance clients understand.

Zapier migration checklist

Moving client from Zapier to n8n:

1. Export Zap list with step counts (task math)
2. Rebuild top 3 Zaps by volume first
3. Run parallel 2 weeks — compare outputs
4. Cut Zapier subscription after validation
5. Document JSON exports in client Git repo

Typical savings NPR 15,000–40,000/month equivalent at scale — see break-even table above.

Homelab vs cloud VPS

Homelab on NTC fiber saves USD but risks:

• Power cuts (UPS required)
• Dynamic IP (Cloudflare tunnel helps)
• No SLA for client contracts

Use homelab for personal automation; client production on VPS with backup.

n8n Cloud pricing snapshot (2026)

When self-host ops is not viable:

n8n Cloud tier	Executions/mo	Approx USD	vs self-host
Starter	2,500	~24	Similar VPS cost, less ops
Pro	10,000	~60	Cheaper than Zapier tasks
Enterprise	Custom	Custom	SOC2, support

Break-even: self-host wins above ~5k executions unless you bill ops time at zero.

Make.com vs n8n feature trade-offs

Feature	Make	n8n self-host
Visual builder	Excellent	Good
Code nodes	Limited	JavaScript/Python strong
Self-host option	No	Yes
Nepal latency	EU/US servers	Choose APAC/SG VPS
Client handoff	Easy for non-devs	Needs technical handover doc

We keep Make accounts for two legacy clients who edit own flows; everything new defaults n8n.

Execution data retention policy

n8n stores execution logs in Postgres — grows unbounded:

• Set EXECUTIONS_DATA_MAX_AGE (e.g., 168 hours)
• Prune success executions aggressively
• Keep error executions longer for debug
• Monitor disk weekly on small VPS

Disk full at Dashain campaign peak = missed leads — automate disk alert.

Integration with GitHub Actions CI/CD

Pattern: post-deploy webhook → n8n → Slack “site live” + IndexNow + cache purge request to Cloudflare API.

Same orchestrator handles marketing site and client static properties — one VPS, many webhooks.

Disaster recovery drill (annual)

1. Restore Postgres backup to staging
2. Import workflow JSON from Git
3. Reconnect OAuth credentials (expect re-auth)
4. Fire test webhook end-to-end
5. Document minutes to recovery

Target RTO under 2 hours for agency SLA credibility.

Real workflow JSON export discipline

Every production workflow:

1. Export JSON after change
2. Commit to private Git with message workflow: lead-router v1.3
3. Tag release before n8n upgrade
4. README in repo: credentials mapping (not secrets), owner, client

Client offboarding = Git handover + credential rotation checklist.

Scaling beyond single VPS

When executions exceed 50k/month:

• Split read-heavy workflows to second worker
• Move Postgres to managed service
• Consider n8n queue mode with Redis
• Rate-limit outbound API calls to avoid Google/Meta throttling

Most Nepali agencies never need this tier — single 4GB VPS handles 20k runs comfortably if DB pruned.

Cost comparison worksheet (copy for clients)

Item	Zapier annual	n8n self-host annual
Platform	$2,400	$0
VPS	—	$216
Backups	—	$36
Setup (one-time)	$0	$400 (8hr @ $50)
Ops (monthly × 12)	$0	$600 (1hr @ $50)
Total year 1	$2,400	$1,252

At 10k multi-step runs Zapier often exceeds $3,600 — worksheet closes sales for automation upsell tied to agency stack.

Troubleshooting production incidents (runbook)

Symptom	First action
All workflows failed	Check VPS disk, RAM, Docker daemon
Single workflow failed	Open execution log, test credential
Webhook 504 timeout	Increase proxy timeout, optimize nodes
Google node auth error	Reconnect OAuth in n8n credentials
Duplicate CRM rows	Add dedupe node on email field

Keep runbook in Notion/Git — junior on-call should recover without calling you at midnight.

Zapier task reduction tactics (before migrating)

If client resists migration short-term:

1. Combine multi-step Zaps into Code step (single task)
2. Use Paths sparingly — each path branch may bill
3. Filter early — drop noise before actions
4. Schedule batch runs instead of per-row triggers

Still temporary — break-even table above shows when migration pays.

Observability: what to log

Minimum logging for production n8n:

• Workflow name + execution ID on failure
• Webhook payload hash (not full PII) for dedupe debug
• Outbound API status codes
• Execution duration — flag workflows >30s

Ship logs to Loki or even structured JSON files with logrotate — debugging without logs wastes more time than logging costs.

Related reading

• Automated SEO pipeline architecture
• AI automation stack for agencies