Why we chose n8n + Postgres over Zapier for African SMBs

A short field note on building automation infrastructure that survives African price points, intermittent connectivity, and the people who built it.

Every SMB owner we meet has the same problem with their automation stack: it works until it doesn't, and when it doesn't, nobody knows why. Zapier is the obvious default for most teams reaching for automation tooling for the first time, but for businesses operating at African price points and African connectivity, it quietly fails in ways that don't show up in the marketing.

We've now deployed self-hosted n8n on top of Postgres for three different SMB clients, and I want to set down, while it's still fresh, why we keep arriving at this stack and what we've learned about running it in production. This is not an n8n tutorial. There are plenty of those. This is the architecture decisions, the costs, and the trade-offs that nobody writes about until they've made the wrong call first.

What Zapier costs you in practice

Zapier's pricing is opaque on purpose. The headline number, the per-month plan cost, is rarely what you actually pay. The real cost is the per-task pricing, and it scales with success: the more your business grows, the more the tooling charges you for the privilege.

For a typical small business with around 3,000 orders a month, multi-step workflows (which is most of them in practice: receive WhatsApp message, look up customer, log order, send confirmation, update inventory) can rack up tens of thousands of tasks per month. At enterprise plan pricing, you're paying $599 a month before you've even started building anything custom.

The architecture we keep arriving at

Three services. Docker Compose. A single Hetzner CX22 box at €4.51 a month. The whole stack looks like this:

# minimum viable automation stack
services:
  postgres:
    image: postgres:16-alpine
    environment:
      POSTGRES_DB: n8n
      POSTGRES_USER: ${PG_USER}
      POSTGRES_PASSWORD: ${PG_PASS}
    volumes:
      - ./data/postgres:/var/lib/postgresql/data

  n8n:
    image: n8nio/n8n:latest
    environment:
      DB_TYPE: postgresdb
      DB_POSTGRESDB_HOST: postgres
      N8N_HOST: ${DOMAIN}
    depends_on: [postgres]

  caddy:
    image: caddy:2-alpine
    ports: ["80:80", "443:443"]

That's it. The total monthly cost, including domain, Caddy-managed TLS, backup storage on Backblaze B2, and the box itself, comes in under $15 USD for the kind of workload that would cost a client $200 to $600 a month on Zapier.

Why Postgres and not the bundled SQLite

n8n ships with SQLite by default. For a single user playing around, this is fine. For anything production-bound, it isn't. SQLite locks the whole database on write, which means concurrent webhook bursts (think 30 WhatsApp messages arriving in the same second after a marketing push) will start dropping or delaying executions. Postgres handles this without flinching.

The trade-offs nobody admits

Self-hosting is not free. It's just cheaper in money and more expensive in attention. Here's the honest comparison after running this stack at three different clients for the last 10 months:

That last row is the real cost. Self-hosting means you are the SRE. For the first month, something will break that didn't break before, and you'll need to look at logs. After that, things stabilize.

What this means for African SMBs specifically

Two things particular to the African SMB context make this trade especially worthwhile:

• Margins are thin. A coffee shop in Abuja running 2,000 transactions a month cannot rationalize ₦300,000 a month for automation tooling. They can rationalize ₦10,000 a month for a VPS.
• Currency exposure is real. Zapier bills in USD. A USD-billed SaaS line item is a hedge against your business that you don't get to participate in.
• Data sovereignty matters more here, not less. Customer data is increasingly subject to local regulation. Keeping it on a box you control, in a jurisdiction you understand, is becoming a feature.

What would change my mind

If Zapier dropped task pricing entirely. If they added meaningful self-hosted options. If they lowered the per-execution price by 10 times. None of these feel imminent, but I want to be honest about the conditions under which our default would flip.

Until then, for any SMB serious enough about their operations to pay for automation but cost-conscious enough to feel it on the bottom line, the self-hosted n8n + Postgres stack is the answer we keep arriving at. It's not always the easiest answer. It is consistently the right one.