The Math of Excellence

This summer I taught one of my last professional military education classes to my Marines. I wanted to leave them with something they would still use after I was gone, so I wrote one line across the whiteboard:

Quality = Knowledge^a × Practice^b × Talent^c

The idea is borrowed. It comes from Patrick Winston, the MIT professor whose lecture "How to Speak" became a campus tradition for more than forty years. Winston argued that the quality of your communication depends on how much you know, how much you practice with what you know, and your inherent talent, and he wrote the T small on purpose. "What really matters," he told his students, "is what you know." He made the point with a story about spotting Olympic gymnast Mary Lou Retton on a ski slope. He was the better skier, he decided, because he had the knowledge and the practice, and all she had was the talent.

Winston was talking about speaking. Watching the lecture, I kept thinking it described everything my Marines do: fixing a vehicle, running a radio net, leading a fire team. It also describes the work I do now, helping organizations adopt AI.

Two changes to Winston's formula

Winston wrote quality as a function of the three and left it there. I wrote it as a product with exponents, because that is the version that matched what I had seen in twenty years of watching Marines succeed and fail.

The first was multiplication. When the terms multiply, a zero anywhere makes the whole product zero. A Marine with talent and hours of practice but no tactical knowledge is a fast-moving liability. A Marine who knows the manual cold but has never done the task in the dark, tired, with someone shouting, freezes when friction arrives. And the naturally gifted Marine who never works at it gets passed by people who do. None of the three can carry the other two.

The second was exponents. They stand for leverage: how much a gain in each variable is worth. In a Marine unit, knowledge and practice carry the big exponents. They compound, because every repetition builds on the last and every manual you read makes the next one easier. Talent is the starting point. It gets you through recruit training. It does not win the fight.

The exponents are a way of thinking. Nobody can measure them, and nobody can tell you that b equals 2.3. What the model gives you is a question: which variable is near zero right now, and which one returns the most for the next hour of effort?

I closed the class with this: when it is time to execute, you will not rise to the level of your talent. You will default to the level of your practice and your knowledge.

The same equation runs your AI rollout

Most organizations adopting AI spend nearly all of their money and attention on the smallest variable.

Map the equation onto an AI program and the variables line up.

Talent is the tool. The model, the license, the raw capability. It is the variable organizations buy, compare, and announce. It is also the small t. Your competitors can buy the same model next quarter, and a better model will not fix a team that does not know what to do with the one it has.

Knowledge is what your people know about the work and the tool together. Which tasks the tool handles well in this specific workflow. What a good output looks like, and what a confident wrong one looks like. Where the data comes from, and when to go back to the source. That knowledge lives in your domain. No vendor demo can supply it.

Practice is repetition inside the real workflow. Real data, real deadlines, the same tasks people do every week, done the new way until the new way is the default.

Now apply the multiplication.

A company that buys licenses and stops there has talent times zero. Logins rise for a month, then the work quietly goes back to the way it was done before. That is why license counts tell you so little about adoption.

A company that adds training but no practice has knowledge that decays. The course was good and the slides were clear, and three weeks later, under a deadline, everyone defaults to the old workflow, because that is the one they have practiced. The line from my class holds in an office too.

A company that pushes hard on practice without knowledge gets the fast-moving liability: people producing AI output quickly that they cannot evaluate, and shipping it. This is the most dangerous of the three, because it looks like success on every usage dashboard.

What to do with it

Buy talent last. Choose a capable tool, but do not let the model comparison eat the budget and the calendar. It is the variable with the smallest exponent.

Build knowledge by role, inside the workflow. A general AI-literacy course is a start. The knowledge that matters is specific: what this tool does well for this job, where it fails, and how to check it. Write it down where people do the work.

Schedule practice the way a unit schedules rehearsals. Make it frequent and make it realistic. Put the new way inside the existing workflow instead of in a separate tab people have to remember to open. The Marine Corps does not expect a skill to hold under stress if it was practiced once in a classroom, and neither should you.

Measure each variable separately. A product hides which term is zero. Track proficiency, meaning whether people can tell a good output from a bad one, alongside active use in a named workflow, and compare both against a baseline taken before launch. If one is near zero, you know exactly where the next effort goes.

Multiply the team. The last takeaway in my class was to share what you know and lead practice with your peers, because a fire team is only as good as its weakest member's equation. In an organization those people are your champions: the users who get good first and pull their coworkers along. Find them early and give them the time to do it.

The equation I hold RIG to

I run Russell Innovation Group on the same equation, and it explains why the company builds things.

Start with talent. I have a natural aptitude for mathematics and computers. That has never, on its own, produced a quality result for anyone. It is the small t in my own equation, and I treat it that way.

Knowledge, for RIG, means staying current on two things that never hold still: the technology, and what the workforce actually needs from it. Models, tools, and rules change faster than any manual. The needs of the people doing the work change more slowly, but they are the half most AI advice skips.

Practice means building. RIG is not a software engineering company, and most of my AI projects are not products in search of customers. Each one is a repetition: pick a real pain point, build something that goes after it, and find out whether AI removed the pain and created efficiency, or whether it could not. Some of those answers are yes. Some are no. Both are what I am really after, because knowing what AI can and cannot accomplish in a given workflow is the knowledge any organization adopting AI needs, and I do not know a way to get it without doing the work.

Advice about AI from someone who has never built with it is knowledge times zero practice. Building without staying current is practice aimed at last year's problem. I try to keep both terms above zero.

Own the equation

Talent is the only variable you can buy. Knowledge and practice have to be built, by your people, over time, inside the work. That is slower than a procurement decision, and it is the part that decides whether the purchase ever pays off.

I ended the class with a question, and it works just as well in a conference room: which variable is your team's weakest link right now, and what will you do this week to raise it?

About the author

Brandon Russell is the founder of Russell Innovation Group. He has spent twenty years in the United States Marine Corps leading equipment, data, and process change, and now helps organizations adopt AI through change management. The views expressed here are his own and do not represent the Marine Corps or the Department of War.

Sources

Patrick Henry Winston. How to Speak. MIT OpenCourseWare, RES.TLL-005, January IAP 2018. https://ocw.mit.edu/courses/res-tll-005-how-to-speak-january-iap-2018/