What should I automate first in my business?

Four properties, and a strong candidate has all of them:

1. It repeats. Weekly or better. Frequency is what pays back the setup effort, because a workflow that fires forty times a year returns its documentation cost in the first month.
2. It runs the same shape every time. Same inputs, same structure, same destination. Sameness is what makes the method writable, and a written method is the thing the AI actually runs on.
3. A miss is cheap. Early drafts will be imperfect while the corrections accumulate, so the first candidate should be work where an imperfect draft costs you a two-minute edit, not a client relationship.
4. You can describe it. If you cannot explain how you do the task, step by step, to a capable assistant, no tool can run it, and the gap is a documentation problem wearing an automation costume.

Score your recurring tasks against the four and the winner is usually unglamorous: the recap, the brief, the report. That is the right answer, not a consolation prize. The point of the first automation is to prove the pattern in low-stakes conditions, and the boring candidates are exactly where those conditions live.

The same shortlist surfaces across almost every established practice, ranked here by how often it wins:

1. The meeting recap and follow-up. A call ends, and the summary, action items, and follow-up email draft themselves for your review. Repeated several times a week, identical in shape, and the single most common leak in client experience.
2. The pre-call prep brief. Where things stand, what changed, what comes next, assembled before every client conversation. Retires the ten-minute scramble and makes you the most current person in every meeting.
3. The content derivative pipeline. One real piece of work, the client call, the workshop, the newsletter, fanned into its byproducts: the post, the follow-up, the case note.
4. The weekly report. Metrics, project status, pipeline, gathered on a schedule instead of by memory.
5. Inbox triage with drafted replies. Sorted by what needs you, with routine responses pre-written for your approval.

Notice what unites the list: every item is preparation for or aftermath of a human moment, never the moment itself. Your calls, your judgment, your relationships stay yours. The machinery around them is what stops needing you.

The disqualified list matters as much as the shortlist, because the wrong first pick is how AI projects lose their mandate. MIT research found roughly 95% of generative-AI pilots producing no measurable return, and the anatomy of failure is consistent: tools aimed at ambitious targets with no documented process underneath.

Hold off on:

1. Anything requiring judgment under stakes. Sales conversations, pricing decisions, the difficult client moment. The system can brief these; it should not conduct them.
2. Anything novel each time. Automation runs on sameness. Work that reinvents its shape per instance has nothing for a method to grip.
3. Anything broken. Automating a process that frustrates clients produces faster frustration. Fix the workflow first, then automate the fixed version.
4. Unreviewed outbound anything. Nothing ships under your name without your eyes until a workflow has earned lighter review over months, not weeks.
5. Anything you cannot yet describe. Undocumented intuition is real expertise, and it is also unautomatable until it becomes words.

The pattern behind all five: automation amplifies what exists. Aim it at documented, stable, low-stakes machinery and it compounds. Aim it at judgment, chaos, or aspiration and it amplifies those instead.

The sequence takes an afternoon of setup and two weeks of proving:

1. Document the method. Do the task once while narrating: what you look at, in what order, what good looks like, what you never do. Transcribe it and shape it into a one-page method the AI will run on. This step is the whole game; everything after is plumbing.
2. Run it assisted. For two weeks, the AI drafts from your method and you review every output. You are not checking whether the tool works. You are finding where your documentation was incomplete.
3. Fold in every correction. Each fix goes into the method document, not just the draft, so misses become permanent rules instead of recurring edits.
4. Wire the trigger. Once the drafts arrive clean, connect the workflow to its natural starting event, the call ending, the Monday morning, the form submission, so it fires without you remembering.
5. Keep the review gate. The workflow now runs itself to your approval, which is the finish line for a first automation. Full autonomy is a later promotion, earned by a track record.

Most owners are surprised by where the time goes: ten percent tooling, ninety percent turning what they do into words.

Four measurements, none of them a feeling:

1. It fires without you remembering. The trigger works, the drafts arrive, and the workflow no longer occupies a slot in your head. The vanished mental slot is the first real return.
2. Review time is shrinking. Week one's five-minute edits become week four's ninety-second approvals as the corrections accumulate in the method document.
3. Corrections are getting rarer. You are fixing new edge cases, not the same miss twice. A repeated correction means the fix landed in the draft instead of the document.
4. The hours actually left. Check the calendar, not the impression. METR's research on AI and productivity found felt time savings and measured time savings diverging sharply, so measure rather than trust the feeling.

When all four hold for a few weeks, promote the next candidate from your shortlist, and expect acceleration: the second automation reuses the loop, the documents, and the confidence the first one built. Owners who prove one workflow typically have four running within the quarter.

Standing up the foundation the whole sequence runs on, your methods captured and loaded into an AI on your own machine, is exactly what our AI Native Activation is for.