Find the few questions doing most of the asking

Jay WestOut of My Head · about 12 min

By the end of this step you have:

Every question in your log named by kind — discounts, scope, scheduling, approvals, where-is-it
Your top three kinds, with a count of how much of your day they take
A count of everything waiting on you right now, the honest size of the bottleneck
Who brings you the most — the person, client or location doing the most asking, named

The idea

The questions feel different every time because the client, the amount and the day change. Strip those away and most of them are the same few kinds. Once you can see the kinds, you can answer each kind once, in writing, instead of every question one at a time.

Each answer costs you more than the minute it takes — the switch away from what you were doing, and the getting back. The small answers spend your day before the big work starts.

And when everything waits on one person, the fix is not that person working faster. It is the system no longer needing them for the parts that repeat.

An example

A managed IT owner has three days of log, thirty-one questions. Sorted, nine are "can we do this for the client without a change order", seven are "who covers this ticket", and five are "can I order this part". That is twenty-one of thirty-one in three kinds.

Then he counts what is waiting on him that minute — four quotes to approve, two escalations, one hire decision. Seven things. He had been telling himself it was a busy week. It is a system that routes seven things through one person.

Do this now

Ten minutes. In the log below, give every question a kind in two or three words — reuse a kind when it fits, the box suggests the ones you already named. Three kinds are worth naming when you see them: "where is it" (a link or a fact, not a decision), "fire" (a problem someone dropped on you), and "unasked" (a decision you made that nobody brought you). Then look at the counts at the top: those are your top kinds. Then read who asks most, beside the top kinds, and any kind that keeps coming from the same person or unit. Last, write on a sticky note or in your phone the number of things waiting on you right now — approvals, replies, decisions anyone is blocked on.

Name the kind of question each one is.

Your Question Log: day 1 of 5 · 0 questions reached you · 0 rules written

This step works on 2 or more days of log, and yours has none yet. Log today, come back in 2 days, and do it then. Or fill the log now: paste a week of your messages with “Paste rows instead”. The steps after this one work without it, so you can carry on and come back.

If your log is a few big one-off calls, not repeats, and you want them challenged

Your log is big calls, not repeats — a hire, a price change, a new market, a partner? Those do not become rule cards. Go to "The short list of decisions that stay with you" and name the ones that stay yours, then keep each in the Big calls section from "Keep a record of your big calls, so you learn from them honestly". At each monthly Big calls review, run the board prompt from "Let the AI argue with you before you commit" on the next big call on your desk, so every new entry is argued with before it goes in.

If your log is thin: nobody asks you, you do the work yourself, you work alone, you mostly look things up, or your bottleneck is fixed

A thin log is a finding, not a failure: questions from other people are not where your time goes. Sort what you do have instead — the decisions you made that nobody asked for, the loops from your brain dump, the approvals or versions your track turned up. Your time goes to doing, not deciding — every job yourself since the team got smaller? "What would stop if you were gone for two weeks" splits the two. Paste a week of your own notes or journal with this — "List every decision I made or put off in these notes, one per line, as: date | me | the decision | what I chose." Then sort those. The kinds that repeat in your own decisions are the rules you write next. Work with agents? Each stop, correction or override is a row too — who is the agent, what it was about to do, what you told it instead.

Working alone? Sort your customers' questions, and the top kind becomes the first reply your AI drafts for you. Your time goes to finding your own material instead? Log the things you went looking for (me | what I looked for | where I found it) and sort those; the top kind is what to load into your notebook first.

Fixed your own bottleneck already? Ask each lead to keep the log for five days. Whoever has the longest waits is the next bottleneck, and the same sort works on them.

Test it

Ask the AI test question again, but pick it from your top kind. If the answer is right, that kind is nearly ready to hand over. If it guesses, the next step writes the rule that fixes it.

This week

Keep logging. By day five the top kinds settle, and the count of what waits on you is the number to watch fall.

Where it goes wrong

Making kinds so broad they mean nothing ("client stuff") or so narrow each has one question in it. A good kind is one you could write a single rule for.

This step is written for this course from the research in the full curriculum. The research behind it, on howtothink.ai (L-0442, L-0441, L-0951)

Your first step · step 1 is next

Next: step 1, which the rest build on →