Start with the work, not the tool
Before we open n8n or write a prompt, we map the process as it actually runs. Who starts it, what triggers it, which systems are touched, where decisions are made and what happens when something goes wrong.
That map usually reveals that the expensive part is not the visible task. It is the waiting, the re-entry of data or the follow-up that nobody owns.
Where automation pays off
Good candidates share a few traits: the work is repetitive, the rules are stable enough to describe and the cost of a mistake is manageable or reversible.
- Routing enquiries to the right owner with the right context.
- Syncing records between a CRM, inbox and internal database.
- Drafting first-pass replies, summaries or research for a human to review.
- Running scheduled checks and escalating only exceptions.
- Handing structured data from one tool to another without copy-paste.
Guardrails are the product
The difference between a demo and a dependable automation is what happens at the edges. We add retries, timeouts, logging and a clear failure path. When an LLM is involved, we constrain its output, validate it and keep a human approval step wherever the consequences are real.
An automation that fails silently is worse than a manual process, because nobody knows it has stopped working.
A practical first step
Pick one workflow that happens every week and has a measurable cost. Document the current steps, automate the narrow happy path and keep the exception path manual at first. Measure the time saved before expanding the scope.
That approach gives the team something useful early and avoids a large, fragile automation that nobody trusts.