Day 11. Introduction to Workflow Automation

What makes automation hard is not the concept. It is honesty about your process.

A trigger fires, one or more actions run, that is the entire model. The difficulty is the parts of the process nobody wrote down: every step, every decision, every place a person judges something. Tools today: Zapier and Power Automate.

Not every repetitive task deserves a workflow. Score candidates on frequency, determinism and the cost of getting it wrong. The best first automation is high frequency, fully deterministic, and cheaply reversible. The worst is the interesting one where the edge cases outnumber the routine cases.

The written lesson. Read it slowly, it saves you later.

Map it on paper first

Write the process as it truly runs, not as the manual describes it. You will find three things. Steps that exist only because two tools do not talk to each other, automate these. Steps that exist because someone must decide, leave these alone or route them to a human. Steps nobody can justify, delete these, often the largest saving in the exercise, and it requires no software at all. Automating a bad process gives you a fast bad process.

Data does not arrive clean

The trigger delivers a payload, and it will be messier than you expect: inconsistent capitalisation, dates in three formats, empty fields, names with trailing spaces, duplicates. Put the formatting and validation steps in from the start rather than discovering them in production. Assume every field may be empty and decide what happens when it is.

Build the failure path before the happy path

The professional habit, and the one beginners skip entirely. What happens when the trigger fires twice for the same event, when a required field is missing, when the downstream service is down, and above all, how will you find out that it broke. A manual process fails loudly because a person notices. An automation fails silently and keeps failing, at scale. Build a notification into the error branch on day one, and check the run history deliberately, not only when someone complains.

AI steps inside automations

Classification, extraction, summarisation and drafting inside a workflow are genuinely powerful: route this enquiry by intent, pull the invoice number out of this email, summarise this ticket for the handover. Two disciplines. Lock the prompt with a fixed output format, because downstream steps break on free prose. And guard the output, because the model will occasionally return something unexpected and nobody is watching. Validate before you write to anything that matters.

Two tracks. Pick yours, produce something.