Your AI Leadership Starting Point

Before reading further, try a short poll on yourself. On a scale of 1 to 7, how well do you understand how an AI chatbot produces its answer to a question? Note the number. Now explain it, in writing, in three plain steps, as you would to a new board member. Then rate yourself again.

For many people, the second number is lower than the first. In a series of experiments at Yale, Leonid Rozenblit and Frank Keil asked people to rate how well they understood everyday devices, such as a zipper. They then asked them to write a step-by-step explanation of how each one works, and to rate themselves again. The ratings fell. The researchers called the effect the illusion of explanatory depth, and found it much stronger for knowing how something works than for knowing facts, procedures or stories¹.

That matters because much of what an executive needs to know about AI is explanatory: why a fluent answer turns out to be wrong, why a pilot succeeds and the rollout stalls. It is easy to feel you understand these things from headlines. The test is whether you can explain them.

A starting point, not a score

The starting point is a way to find out where you stand before Level 1 begins. It produces three things: the questions you can already answer, the one gap that matters most for your role, and a real problem from your own organization that you will carry through the modules ahead.

It is deliberately not a test. Nobody needs to see your answers, and a low starting point is not a problem. An unknown one is, because you cannot judge whether the program changed your thinking if you never wrote down where that thinking began.

There is also a reason to be careful about how you assess yourself. Ethan Zell and Zlatan Krizan pooled 22 meta-analyses covering self-ratings of academic, professional, medical, athletic and other skills. On average, the correlation between how good people thought they were and how they actually performed was 0.29: real, but modest². Two conditions made self-ratings more accurate. The question was specific to one domain rather than broad, and the task was objective and familiar. Both conditions shape the method that follows.

The third step is the one that is hard to fake. You can feel confident, and even produce a fluent explanation from memory of a briefing, without ever having applied an idea to a real choice. Pointing to a decision, a question you asked, an approval you held back or a number you demanded, turns a belief about yourself into something another person could check. If you cannot find one, that is not a failure; it simply tells you where you start.

Rate evidence, not confidence

A five-point scale invites a feeling. A better scale asks for evidence. For each question in the next section, place yourself at one of three levels, and use the highest level you can demonstrate, not the one you believe.

The three levels describe you, one question at a time. They are not a maturity model for your organization; that is the subject of The AI Maturity Model in Module 09, and it works differently. Most executives will find themselves at different levels on different questions, which is exactly the information you need.

Keep the page. At the end of Level 1, repeat the same test on the same questions, without looking at your first answers, and then compare. Progress shows up as a question moving from “heard of it” to “can explain it”, or as a belief about your organization becoming something you can show. It does not show up as the number of chapters read.

Seven questions a leader should be able to answer

The seven questions below follow the arc of Level 1. Each is specific, which, as the research above suggests, makes an honest rating more likely. Each also names the module that answers it.

Two cautions help. First, be strict about the middle level. “Can explain it” means you could do it now, aloud, to a skeptical colleague, including one limit or exception. If you would want to look something up first, you are at the first level, and that is fine. Second, notice which questions you were tempted to skip. A leader from a technology background might rate the first question high and the economics question low; a finance leader might do the reverse. Neither profile is better. They need different starting points.

Your organization is a self-assessment too

The illusion of explanatory depth applies to what you know about your own organization as much as to what you know about AI. Leaders form a picture of how AI is used, governed and measured from reports, steering committees and the projects that reach them, and much of the use is out of sight by design. In the University of Melbourne and KPMG’s 2025 survey of more than 48,000 people in 47 countries, 57 percent of employees said they had at least sometimes hidden their use of AI or presented AI-generated work as their own³. A leader who relies on what reaches the steering committee is seeing the part of the picture that people chose to show.

So the same rule applies. For each of the five questions below, mark whether you believe the answer or can show it, with a list, a name, a number or a document. These are not a readiness audit. A full assessment of readiness is the work of Assessing Enterprise AI Readiness, and the scale an organization climbs is the subject of The AI Maturity Model, both in Module 09. These five questions tell you which chapters to read with your own organization in mind.

Moving from “believe” to “can show” is often quick. Ask a team lead which AI tools the team used last week. Find the name on the last AI approval in your area. Ask for the baseline behind the last AI result reported to you. If those requests take more than a day to answer, that is itself a finding about where your organization starts.

Where to start

With seven ratings and five organizational answers in hand, the obvious move is to start with your strongest area, where progress will feel quick. Resist it. Your starting point should be the question your role depends on most and that you can currently answer least well.

“Depends on” should be read over the next twelve months, not in general. A chief financial officer about to approve a large AI budget depends on the economics question now. A chief operating officer whose teams are already using AI tools depends on the governance question now. Pick one, or at most two.

The program’s reading tracks then give you a path. They are named after roles, but choose by gap, not by job title. A finance leader whose thinnest answer is about how the technology fails should take the technology track.

A track is a way in, not a substitute: the full Level 1 sequence is still the most reliable way to answer all seven questions well.

Story: the recaps that launched everywhere at once

In August 2023 Gannett, the owner of USA Today and many local newspapers, began publishing short automated recaps of high school games. They came from LedeAI, a company that turns game scores into news reports, and they ran in several of its local papers⁴. The appeal was plain: local readers want results for games no reporter can attend, and the recaps cost almost nothing to produce.

Within days, readers were sharing the recaps as jokes. One described a football game as “a close encounter of the athletic kind”⁵. A soccer recap went out with its template still showing: the Worthington Christian “[[WINNING_TEAM_MASCOT]]” had defeated the Westerville North “[[LOSING_TEAM_MASCOT]]”. Gannett paused the experiment, saying it would keep evaluating vendors as it refined its processes⁴.

The post-mortem is unusually candid. LedeAI’s co-founder, Jay Allred, later explained that the template text had been written and checked by people. The visible errors came from custom code written for Gannett, which had bugs, and the launch covered six or seven major markets at once⁶. He said the code had been tested less than usual because the company had set itself a deadline, and that if he did it again he would launch on one site and check every piece of content⁷.

Read it through the seven questions. The value and economics questions had strong answers. The understanding question did not: a system that only sees a box score cannot name the player who scored, which is why the recaps read as empty. Nor did the risk question, about what a bad output looks like and who sees it before readers do, or the execution question, about launching on one site or across several markets at once. No one was harmed, and the cost was embarrassment. That is what makes it useful. The decision was strong exactly where the questions were well answered, and it failed where they were not, which is the pattern a starting-point assessment is meant to expose before it costs you something larger.

What this means for leaders

The assessment becomes a plan with two more steps: one gap, and one problem.

Start with one gap your role depends on. Choose a track by that gap. Your strongest area matters too, because it is where you can help colleagues.

Carry one real problem. Choose a problem from your own organization to bring to every module. It should be a business problem, not an AI project: something that matters, that you understand, and where you suspect AI might help but do not yet know how. Choosing which use case to pursue first is a separate decision, and Where Should We Start With AI? in Module 01 covers it. Here, the problem is a learning anchor. Each module will give you a sharper way to look at it, and by the end of Level 1 you should be able to say not just “AI could help here” but why, at what cost, with what risk and on what plan.

Finally, make the commitment specific. Plans of the form “when X happens, I will do Y” make people markedly more likely to follow through than a general intention. Across 94 studies, such if-then plans had a medium-to-large effect on reaching the goal⁸. “I will learn more about AI governance” rarely survives a busy quarter. “When the monthly risk report arrives, I will read one Module 07 chapter before the review” is far more likely to.

Check yourself

1. Feeling that you understand a topic well is good evidence that you do.MythFact
2. Self-ratings are more accurate when the question is specific to one area.MythFact
3. Most of the AI use in an organization shows up in its formal reports.MythFact
4. The best place to start is your strongest area, to build momentum.MythFact
5. The problem you carry through the program should already be an AI project.MythFact
6. Writing a specific if-then plan makes you more likely to follow through.MythFact

Reflection: the end of Level 1

What comes next

Your starting point is written down. The first question to test it on is the largest one: is AI really changing everything, or is that a slogan? Try the explain test on it before you read on. The next chapter, AI Is Changing Everything, opens Module 01 with what the evidence says.