An automation consultant who tells you what not to automate

Don't automate yet if

• Nobody can describe the job. If each person does it differently, automation picks one version at random.
• It happens once a quarter. You won't get enough practice runs to trust it.
• A mistake is expensive and hard to undo. Keep a person in charge, and let automation prepare the work instead.
• The inputs are a mess. If two sources disagree and nobody knows which wins, fix that first.

Automate when

• It happens every week or more.
• One person does it well, and you'd want their version copied.
• You can tell a good result from a bad one by looking at it.
• The information it needs lives somewhere a tool can reach.

Why I use AI for it

Older automation tools are great at moving data from A to B. They struggle with anything that needs reading or writing: summarising a call, drafting a reply, deciding which person an email is about. AI handles that part, as long as the process tells it how.

So I combine the two. Plain automation where the steps are fixed. AI where the work needs judgement, inside rules that say what it may and may not do.

Proof

For a software startup, the AI-run pipeline handles signups, reply matching, bookings, no-shows, call notes and follow-up timers three times a day. Its rules: drafts only, never email the same person twice about the same thing, ask instead of guessing, and never mark a lead lost without a reason.

One honest lesson from that build: links in the drafted emails were expiring before people clicked them, and use of the pipeline was uneven until we found it. The founder applied the fix herself from a plain-language guide. Automation needs someone watching the results, not just the runs.

How we'd start

Tell me the job you most want gone. I'll tell you honestly whether it's ready to automate, or what needs fixing first.