x
Sort it before you build anything. Two columns, one page, written once and revisited when something changes.
Most automation advice starts with the tool — here's what it can do, now find something to point it at. That sequence produces impressive automations of things that didn't need automating, and leaves the genuinely draining work untouched because it looked harder. Start with the list instead. The tool question becomes trivial once the sorting is done.
Run each candidate task through these in order. The first "no" decides it.
1. Is it reversible? If the automation gets it wrong, can you undo it in minutes? Drafting, filing, tagging, preparing — reversible. Sending, deleting, paying, publishing — not. Reversibility is the single strongest predictor of whether unattended automation is safe, and it's the question most people skip because it feels obvious in hindsight.
2. Is "correct" checkable without redoing the work? If verifying the output takes as long as producing it, you've moved effort rather than saved it. A drafted invoice you can scan in ten seconds passes. A market analysis you'd have to research to verify does not.
3. Does it commit you to anything? Prices, dates, scope, availability. A task that creates an obligation in your name is a task that needs a human, because an obligation is a promise and promises are what you're actually selling.
4. Would a wrong answer be obvious or invisible? Obvious wrongness is survivable — you'd catch it. Invisible wrongness is the dangerous kind: the plausible figure, the confident summary of a document nobody re-reads. If wrong looks exactly like right, keep it human or add a verification step.
Safe to automate — routine replies to questions you've answered many times · filing, tagging, renaming · first drafts of anything you'll review anyway · data reshaping (screenshot to spreadsheet, export to summary) · scheduled reminders and prompts · retrieval (sending the document that always goes at that stage).
Keep human — pricing and quoting · anything in a complaint thread · scope and deadline commitments · final send on client-facing communication · hiring and firing · anything where being wrong is invisible · the decision about what to automate next.
That last one matters more than it reads. The sorting itself is judgement work, and handing that over is how people end up with a business running on rules nobody chose.
Three reasons, and the third is the one people underestimate.
It stops the drift. Automation expands quietly — a system handling drafts starts handling sends because the sending was tacked on and nobody re-sorted. A written list is a boundary you can check against, which is exactly what an escalation rule enforces at the individual level.
It makes delegation possible. The list is the first thing a future contractor, VA or bench recruit needs. Without it, you'd have to be consulted on everything, which is the opposite of the point.
It survives you forgetting. Here's the checkable claim: an undocumented boundary isn't a boundary — it's a memory, and memories degrade under time pressure. You'll relax the line on a busy Thursday and not notice you've relaxed it. Written down, the same moment becomes a deliberate decision, which is worth logging in the decision log. Deliberate is fine. Drift isn't.
In your handover pack, next to the automation map. Same page ideally: what runs, and what's allowed to run. Revisit when you add a tool, take on a client type you haven't handled before, or find yourself surprised by something an automation did.
That last trigger is the useful one. Surprise means the list and reality have separated, and the list is what needs updating — before the next surprise is expensive.
One page. Two columns. Twenty minutes. Then check your existing automations against it — most people find at least one thing running in the wrong column, and finding it on a Saturday is considerably better than finding it in a client's inbox.
For the audit that produces the first draft of this list, the free 5-min read is here: The 10-Hour Audit.
Related Decodes
TL;DR: Sort tasks before choosing tools. Four questions decide it: is it reversible, is "correct" checkable without redoing the work, does it commit you to anything, and would a wrong answer be obvious or invisible? Safe-to-automate covers routine replies, filing, first drafts, data reshaping and retrieval; keep-human covers pricing, complaints, commitments, final sends and anything where wrong looks like right. Write it down — an undocumented boundary is just a memory, and memories relax under pressure.
