12 n8n templates that save SA businesses real time (with the gotchas)

n8n is the unsung hero of SA SME automation. It self-hosts on a R150/month Hetzner box, has a visual workflow editor your ops manager can read, and the licence cost is R0. Here are the twelve templates we deploy most often, with the gotchas we wish we'd known the first time.

Sales & lead workflows

1. Multi-channel lead router

Leads from Meta Ads, Google Ads, your website form, and inbound WhatsApp get unified into one queue with channel attribution. Round-robin to the right rep on Slack/Telegram with a 15-minute SLA timer.

Gotcha: Meta's webhook can fire twice for the same lead within 30 seconds. Add an idempotency key based on lead_id and a 60-second deduplication window in Postgres. Otherwise your reps will get double notifications and stop trusting the system.

2. Lead intake → CRM enrichment

Form fill triggers an enrichment step (company size, industry, region from public data), then pushes a clean record to HubSpot/Pipedrive. The rep sees a pre-qualified lead instead of three fields and a name.

Gotcha: Don't put enrichment in the critical path. If the enrichment API is down, you still want the lead created. Use a fan-out pattern: write to CRM first, enrich in a follow-up step.

3. Compliant WhatsApp broadcast

Opt-in segmented broadcasts using approved WhatsApp templates. Throttled, deduplicated, click-tracked, with one-tap opt-out.

Gotcha: WhatsApp's marketing template policy will get you suspended if you treat it like email. Throttle to 80 messages/minute, segment small (under 5,000), and never broadcast outside an active 24-hour window.

Support workflows

4. WhatsApp support triage

Inbound message → intent classifier → answer from FAQ vector store → escalate to human if confidence is low. Logs every conversation for audit.

Gotcha: Set the confidence threshold at 0.75 to start, not 0.9. Too-strict thresholds mean almost every message escalates and you've achieved nothing. Tune downward as your test set improves.

5. Appointment reminders

Pull tomorrow's appointments from Google Calendar / Calendly. WhatsApp 24h-before reminder with confirm/reschedule buttons.

Gotcha: SA timezones. n8n's Cron node runs in UTC by default — set the timezone explicitly or your reminders go out at 02:00.

Finance & ops workflows

6. Invoice follow-up sequence

Read overdue invoices from Xero/Sage. Send polite WhatsApp + email nudges at day 7, 14, 30. Stop on payment. Escalate after 30.

Gotcha: Xero's invoice "paid" webhook sometimes lags. Always re-check status before sending a follow-up — otherwise you'll WhatsApp a paid client demanding payment, which is exactly the experience you were trying to fix.

7. POS daily reconciliation

Yesterday's POS transactions matched against bank settlement. Flag missing or duplicated. Daily 7am summary email.

Gotcha: Bank settlement files arrive at 03:00–05:00 SA time and occasionally not at all. Build a "did we get a file?" check first; otherwise the workflow silently produces an empty report and nobody notices.

8. Document OCR + classification

WhatsApp/email-in a document, OCR it, classify (invoice / quote / proof-of-payment), extract structured fields, file in the right folder + ERP.

Gotcha: SA bank statements have a thousand layouts. Don't try to extract structured data from them with regex. Use an LLM with structured output and a confidence score — and fall back to manual review below 0.85.

Ops & compliance workflows

9. Compliance evidence capture

Floor staff send a photo + GPS via WhatsApp. The pipeline timestamps it, files to S3/MinIO with a SHA256 hash, posts to a compliance log. Auditor-ready.

Gotcha: WhatsApp strips EXIF. Capture GPS from a custom WhatsApp message form (location share) instead of relying on photo metadata.

10. AI report summariser

Weekly: pull metrics from your warehouse, send an LLM a structured prompt, generate a stakeholder-ready summary, post to Slack and email.

Gotcha: Stakeholders get bored of LLM summaries that all sound the same. Constrain the prompt to surface 1-2 anomalies (week-over-week movements > 1.5 std deviations) instead of generic recaps.

Infra workflows

11. Cloud cost anomaly alerting

Daily: pull AWS / Azure / Hetzner billing. Compute moving baseline. Alert on >15% deviation with the top three SKUs driving it.

Gotcha: AWS Cost & Usage Reports lag by 24 hours. Build the workflow with that lag in mind — alerting on "today" will always look broken.

12. IoT sensor anomaly workflow

Stream sensor data to TimescaleDB. Rolling z-score per device. On anomaly: WhatsApp/Telegram alert with last-hour chart attached.

Gotcha: Z-scores assume normality. For sensors with bimodal distributions (e.g. cold-room compressors that cycle on/off), use a different baseline — or you'll get a flood of false positives every time the compressor kicks in.

Three rules for production-grade n8n

• Use Postgres queue mode. Memory mode is fine for testing; in production it loses workflows on restart.
• Set timeouts on every HTTP request. The default is "wait forever," which means one bad upstream API can lock your entire instance.
• Tag every workflow with an owner. When 40 workflows are running, "who maintains this one?" is the most expensive question you'll ask.

If you want any of these built, customised, or audited — see the full template library or our n8n build service.