Most SME “AI projects” fail the same way: a chatbot demo, no owner for the workflow, and a spreadsheet that still runs the business.
At Dotkod Solutions I take the opposite bet. Start from a painful, repeating ops job. Automate the glue. Add GenAI only where draft-quality text or classification beats a human on speed — then hand the result back into a system someone already trusts.
The pattern that works
- Name the job — “chase unpaid invoices,” “triage inbound leads,” “summarise support threads into a CRM note.”
- Map the triggers — email, webhook, form, schedule, spreadsheet row.
- Automate the path in n8n (or similar) — fetch → transform → decide → write back.
- Drop GenAI on one step — draft reply, extract fields, classify intent — with a human or rule gate when money or reputation is on the line.
- Log failures — retries, dead letters, a Slack/email ping. Silent automation is worse than no automation.
You’re not building a platform. You’re removing hours from a week.
Where GenAI earns its seat
GenAI is useful when the input is messy language and the output is a draft or structured guess:
- Lead emails → CRM fields + suggested next action
- Support threads → short internal summary
- Product blurbs / FAQ drafts from a brief
- Review or feedback clustering for a weekly digest
It’s a liability when you need deterministic money movement, exact tax figures, or “always send this template or don’t send at all.” For those, use rules, APIs, and official channels — same lesson we learned the hard way on WhatsApp elsewhere.
Where n8n (and friends) earn their seat
The value is usually the plumbing:
- Webhooks from payment gateways or forms
- Nightly syncs between tools that refuse to talk
- Approvals: draft → human click → send
- Retries when a third-party API flakes at 2am
SMEs don’t need another dashboard. They need the existing ones to stay in sync without copy-paste.
What I refuse to automate first
- Anything without a clear owner when it breaks
- Anything where a wrong send costs more than a wrong skip
- “AI for everything” roadmaps before one workflow is live for 30 days
Ship one boring win. Measure hours saved. Then expand.
Takeaway
GenAI without workflow design is theatre. Workflow design without failure handling is a future outage. For startups and SMEs, the win is narrow: n8n (or equivalent) for glue, GenAI for drafts and extraction, humans for judgment.
If you’re scoping an automation or custom web build, hire me or email [email protected].