Coming next: The Second Factory →

Back to Library
Metrics & Evidence⚠️ Pitfall

The 41,000 Hours (Saved From What, Spent Where?)

The Problem

Finance surveyed everyone. How many hours a week does AI save you? They multiplied it out, and now the board slide says it in very large type: 41,000 hours saved. Ask one person where their four hours went last week. There's a long pause, and the pause is the answer. Nothing in anyone's week was designed to catch the time, so the next thing swallowed it. Meanwhile, something happened that no survey asked about. AI took the easy work first: the clean case, the routine email, the five-minute task that used to sit between two hard ones. Every team is finishing more than ever. And by four o'clock everyone's fried, because an eight-hour day is now eight hours of hard.

The Principle

Never skip the middle. Engagement, experience, effectiveness, efficiency: in that order, every time. Most AI programs measure engagement because it's easy, then leap to efficiency because that's what the board wants. Are people using it? Is it a good experience for them? Is it actually working? Only when all three are true does the hours-saved number mean anything. Pair every efficiency number with its counterbalance. Hours saved, next to where they went. Heavy use, next to good use, because some of your heaviest users are agents stuck in a loop. Output, next to the share of AI use that happens after eight at night. And treat tokens like clicks: they tell you how much behavior there was, not whether it was any good.

🎯

4 Steps to Take Action

  1. 1

    Run the four-hour test

    In your next team meeting, ask: roughly how many hours did AI save you last week? Where exactly did they go? Then the question that matters: next week, where do you want them to go? Do it on yourself first.

  2. 2

    Build the four-row dashboard

    One row per E, in order, top to bottom, each with one measure and its counterbalance. If a row is empty, leave it empty and visible. An empty row on a board slide gets filled faster than a missing one.

  3. 3

    Pull the after-hours share

    The share of AI use after eight at night, by team, and whether it's rising. It's the experience number that shows up before the resignation letters do.

  4. 4

    Give some easy work back

    On one team, with HR in the room, send one easy case in five or six back to a person. Throughput dips first, and errors and burnout take longer to move, so don't call it after a week.

💡 Inspired By

Chapters 9 and 10 of The Second Factory, "Twenty One-Person Companies" and "The Middle Two E's." And the Four E's from The Auction Block, learned the time the order was wrong: efficiency came first, handle times went down beautifully, then experience tanked and took everything else with it.

This is what Analytics Pitfalls & Principles is about: people helping people.

In Practice

ClassicMotors' CEO had his next board slide ready. Two numbers, very large: hours saved, 41,000; cycle time, down 32%. Finance had surveyed everyone and multiplied it out. A few days earlier, Rita, who runs seller verification, had stopped Nora by the coffee machine. "I'm getting more done than I ever have in my life. My team is too. So why am I more tired than I've ever been?" Nora sat with Rita's team for an afternoon. The agents now handled almost all the easy cases, so every case left on a human's desk was an estate sale or a lien dispute. "We used to do an estate, then a few easy ones, then another estate," Rita said. "The easy ones were where you caught your breath." At her next session, Nora asked the room how many hours AI had saved them last week. Three, five, six. Then she asked where those hours went. Nobody answered. Finally a woman from Operations said, "I think I just did more stuff." Then Nora asked the data team when people were using it. Fourteen percent of all AI use happened after eight at night. Near the top of the list was her own name.

Take it to work

Give this card to your AI

A skill is a file your AI keeps. Add this one and it knows The 41,000 Hours whenever your work runs into it.

Then ask it: “Does my plan fall into The 41,000 Hours?”

How to add it
  • Claude (a paid plan): Customize, Skills, +, Upload a skill. Pick the file you downloaded.
  • ChatGPT or Gemini: where your plan has Skills, upload the same file.
  • Microsoft 365 Copilot Cowork: Plugins, Add, Upload, then pick the Cowork file.
  • An AI at work that won’t take a skill? Open this card as text and paste it in.

All the AI skills, including ones for the moment before an AI, data or BI strategy, a dashboard or a readout goes out.

More on Metrics & Evidence

From the next book

In The Second Factory, the CEO’s next board slide says 41,000 hours saved. The week before, one of the company’s best people asked Nora why she had never been more tired.

This card is drawn from The Second Factory, Brian Cariveau’s next novel, about a company learning to work with AI. The opening scene is free to read now.

Heard something like this at work? See the pitfalls behind what else you’ve heard this month.Heard at work: free, 1 minute, no account