“What should I automate first?” sounds like a question about tools. It is really a question about work.
Find something that repeats. Make its inputs and decisions visible. Decide how you would recognise a good result. Only then do you have a useful automation candidate.
Why the tool is not the starting point
A tool-first search begins with capability: learn what the product can do, then hunt for somewhere to use it. That gives the business no standard for deciding which process matters most.
Defining the job first creates criteria. The business can compare products against the inputs, output, security requirements, integration needs and review standard of that process.
Narrow the question
Do not rank the whole business from memory. Look at a representative period and record repeated searches, transfers, drafting, chasing and corrections.
Choose one process from that record. The task that felt frustrating may be a useful lead, but check the feeling against what actually happened.
Three checks for a first candidate
It repeats. There are enough real examples to compare and test.
It follows a recognisable shape. The stable steps can be separated from the exceptions.
Its inputs can be made explicit. The source, owner and current version can be named.
If a task misses one of those checks, investigate before automating it. The checks narrow the field. They do not prove the task should be handed over.
What that looked like in the steel engagement
A steel-trading client's complex quotes were taking her 2 to 4 hours each. Run that against the three properties.
It repeated. This was how every complex quote got built, not a one-off.
It followed a shape. The same assembly each time: prices gathered from three suppliers, costs compiled by hand, margin applied, client document written.
Its inputs already existed. They were sitting in her email and her WhatsApp, just scattered across both.
Same quotes now take her 20 to 40 minutes. She still signs off every offer before it reaches a client.
Choose the tool after the job
Write down the job before comparing products:
- the accepted inputs;
- the required output;
- the steps and exceptions;
- the decisions that remain human;
- the data-handling requirements;
- the checks and failure response.
Use that list to compare tools. A feature matters only when it meets a requirement of the job.
Monitor the first build
Record errors, missed triggers, human interventions, elapsed time and maintenance. Decide in advance who notices a failure and what happens next.
A defined process makes quiet failure easier to spot. It does not guarantee that the automation will keep running or remain useful when the work changes.
Begin with a testable candidate
Choose one repeated process, check the three conditions and define the review before building. The first step is complete when the candidate can be tested, not when a tool has been chosen.
If you want a structured first pass, the free AI Workflow Map takes about two minutes and returns a one-page read on where AI may fit.
