Identify what to automate: the weekly audit

Most people approach AI with the wrong question. They ask "what can it do?" and end up with a tab full of demos and no actual time saved. The better question is quieter and a little uncomfortable: "what am I doing every week that I shouldn't be?"

You can't answer that from memory. Memory flatters you. It remembers the work you're proud of and forgets the 40 minutes you spent reformatting a spreadsheet on Tuesday. Before you automate anything, you need a real list.

The one-week audit

The protocol is simple enough that you can start it tomorrow.

For five working days, every time you start a task, jot it in a single note. Paper, Notes app, the back of an envelope, whatever you'll actually keep open. Title and rough duration. That's it. Don't categorize, don't judge, don't try to remember at the end of the day. Capture in real time or you'll lie to yourself.

The list should feel ugly by Friday. "Reformatted slide deck, 25 min." "Replied to vendor about invoice, 10 min." "Rewrote Slack message three times, 15 min." Good. That's the data.

Friday afternoon, sit down and sort the list into four buckets. Most readers are shocked by what shows up where. The "automate this" pile is usually smaller than expected. The "why am I doing this at all" pile is usually larger.

The four buckets

Keep manual. The work where being human is the point. Customer relationships, hard hiring calls, original thinking, the writing where your taste is the product. If a model could do it as well as you, your edge wasn't taste, and you have a bigger problem than your calendar.

Eliminate. The standup that could be a Slack thread. The weekly report nobody opens. The status update that exists because somebody three years ago didn't trust someone who no longer works there. The recurring meeting you inherited. Killing these beats automating them, every time.

Hand to a person. Cleaner-than-you-do-it work that isn't your highest leverage. Inbox triage, travel booking, light research, basic bookkeeping. A VA, a freelancer, or a teammate who actually likes this kind of work will do it better than you do, and faster.

Hand to AI. Research briefs, draft writing, transcription, summarization, data wrangling, repetitive lookups, formatting, light coding. The tasks where the prospect (or the reader, or the customer) doesn't care that a human did it. That last line is the test. If a human's fingerprints don't change the outcome, the task belongs in this bucket.

The automate-it-anyway trap

The most common mistake is automating work that shouldn't exist.

If a report takes you an hour a week and you build an AI workflow that produces it in two minutes, you feel like a hero. But if nobody actually reads the report, you've just built a faster way to waste two minutes a week, plus the hour you spent setting it up. Killing the report is always cheaper than automating the report.

Before anything moves into the AI bucket, ask: would the world notice if this stopped happening? If the honest answer is no, eliminate. Don't optimize.

A worked example

Here's a Friday-afternoon sort from a real-ish product manager's week.

1. 1:1 with my report (45 min) - Keep manual
2. Wrote weekly update for leadership (60 min) - Hand to AI (summarize from notes and Slack)
3. Reformatted competitor screenshots into slides (40 min) - Hand to a person
4. Recurring "ops sync" meeting (30 min) - Eliminate (could be a doc)
5. Drafted PRD for new feature (90 min) - Keep manual
6. Hunted through Linear for tickets to mention in the update (25 min) - Hand to AI
7. Replied to 14 internal questions in Slack (50 min, scattered) - Keep manual (mostly)
8. Wrote three versions of a tricky message to a frustrated customer (35 min) - Keep manual
9. Pulled retention numbers from three dashboards into one sheet (40 min) - Hand to AI or eliminate (does anyone use the combined sheet?)
10. "Quick" status email to a stakeholder who never replies (15 min) - Eliminate

Tally that up. Two real wins for AI. Two clear eliminations. One handoff. The rest stays. The total time saved is bigger than the AI bucket alone, because the eliminations save you time forever, not just this week.

Closer

Run the audit. Be honest on Friday. Resist the urge to automate everything that ends up in the AI bucket on Monday morning.

For most people, the first item that should actually move is email. It's high frequency, low signal, and the patterns repeat. The next guide, "The email triage playbook," walks through how to set that one up cleanly so it doesn't turn into another report nobody reads.