From AMP to JB’s Game Predictor: I Guess I’ve Been Doing This Longer Than I Thought
I’ve been working on JB’s Game Predictor, a system that uses analytics, matchup evaluation and game simulation to predict college football games.
The more I worked on it, though, the more I realized something.
I’ve actually done this before.
Years ago, when I was training and evaluating football players, I developed a system I called AMP — Agility, Motion and Power.
At the time, I was collecting a lot of testing data on athletes: height, weight, 40-yard dash, shuttle, L-cone, bench press, vertical jump and a Tendo rating that measured the power an athlete generated during the vertical jump.
I had a lot of numbers.
The question was what to do with them.
So I worked with a math professor who helped me develop a formula that took all of those individual measurements and produced what I called an AMP Rating.
That’s when it became really useful.
I had testing data from previous years on some of the top football prospects in the country. That gave me something to compare these athletes against.
Instead of simply saying, “This kid runs a 4.5 forty,” I could look at his complete athletic profile.
How did his combination of speed, change of direction, explosiveness, size and power compare to players who had already gone on to play at the next level?
It even helped me with positional projection.
A kid might be listed as a cornerback, for example, but his AMP profile might compare more favorably to the safeties in my database. Another athlete might be playing one position in high school while his numbers suggested his physical profile projected much better somewhere else at the college level.
The individual number wasn’t the important part. It was what happened when you put all the numbers together.
And that’s exactly what we’re doing now with JB’s Game Predictor.
Instead of evaluating an athlete, we’re evaluating a football game.
We’re feeding the simulator information on team strength, strength of schedule, offensive and defensive efficiency, quarterback play, line-of-scrimmage performance, explosive plays, turnovers, personnel, special teams, injuries and home-field advantage.
But just like AMP, I don’t want this to become a spreadsheet that ignores football.
Numbers need context.
That’s why I’ve added what I’m calling the JB Matchup Adjustment.
If one team’s biggest weakness happens to line up with something its opponent does extremely well, that matters. Personnel matters. Scheme matters. Injuries matter. Matchups matter.
Football isn’t played on a spreadsheet. But a spreadsheet can sometimes tell you where to look.
Then we’re going to take all of that information and simulate the game thousands of times.
That gives us more than a predicted score. We’ll be able to look at win probability, average score, likely scoring ranges and how often different types of games develop.
Then comes the most important part.
We’re going to keep score on ourselves.
After every game, we’ll compare what the model predicted against what actually happened. We’ll look at where we were right, where we were wrong and which variables were most predictive.
As the season progresses, we’ll replace more and more of last year’s information with current-season data. Eventually, we’ll begin building our own historical database.
That’s when this could become really interesting.
Years ago, AMP helped me answer the question:
How does this athlete project at the next level?
Now JB’s Game Predictor is going to try to answer another one:
Given everything we know about these two football teams, what is most likely to happen when they play?
Different application. Better technology. A whole lot more data.
But when I really think about it, it’s the same thing I’ve always enjoyed doing.
Collect the information. Study it. Find the patterns. Trust what the numbers tell you—and then apply what you know about football.
The simulator is already running South Carolina–Kent State. I’ll reveal what it says during game week.
