How Baseball Led Me to Marketing Analytics

My dad took me to my first Yankee game in 1959. Patrolling centerfield, number 7, Mickey Mantle…the king of the tape measure home run and the fastest runner ever to first base (my love of stats starting…)

In high school, I discovered Strat-O-Matic baseball, a remarkable game that purported to accurately replicate the statistics of every hitter and pitcher. I had to see for myself! One summer I created an 8-team league (some constructed to be pitching heavy, some batting average heavy, some based on home run hitters…like test vs. control of marketing tactics!), and I set up a schedule where each team played others the same number of games (a good experiment!). 

I keep stats on every hitter and pitcher.  There were no calculators back then, just handwritten calculations on dozens of sheets of paper! Strat-O-Matic did replicate player performance remarkably well, but with some variations… as you might expect with anything probabilistic.

Boom, there I was at the age of 16, entering the realm of statistics, probability theory, causality, and Monte Carlo simulation.

In grad school at the University of Chicago, a stats professor introduced me to “Percentage Baseball” by Earnshaw Cook. This book was the first of its kind…using Bayesian math and decision trees to discover some very unconventional truths, like bunting a runner to second by non-pitchers was a dumb thing to do except situationally in late innings.

Motivated by Cook’s work and fueled by instincts from my Strat-O-Matic summer, 20 years before Bill James and Sabermetrics, I wrote a term paper for a stat class at U of C, “Predicting baseball outcomes”. I build a two-stage regression model…I predicted won-loss percentage as a function of run differential (runs scored-runs allowed) (Basically what Bill James discovered much later with his Pythagorean formula for predicting wins and losses). 

Then I predicted runs scored with various batting statistics and I predicted runs allowed with pitching and fielding statistics. The model had an amazingly good fit to the data.

Big insights came out of my model:

1. On base percentage (OBP) was clearly a superior predictor vs. batting average…a finding that was documented many years later in Moneyball.

2. For pitchers, I created a WHIP statistic (walks plus hits per 9 innings), figuring it was analogous to OBP. WHIP was highly correlated with Earned Run Average.

3. I could simulate who was worth more, Mickey Mantle or Sandy Koufax at their best.  I did this by simulating the difference in runs scored or runs allowed by substituting either of them for the average player’s performance at their position. Then, holding the rest of the team constant, I predicted changes in run differential which allowed me to predict a change in won-loss percentage.  (BTW, Mickey’s “partial derivative” was a few games more than Sandy.)

Basically, I had created WAR (wins above replacement), the most important statistic today for player evaluation in any sport.

Since 2015, baseball geeks like me have a new candy store called StatCast.  Now, they measure things like spin rates and movement on breaking pitches, batting average against on different pitches (guides pitch selection), exit velocity/velo (how hard hitters hit), spray charts, etc. These are the predictors of performance statistics like batting average, home runs and earned run average.

So what does all of this have to do with marketing analytics?

In my consulting, I regularly think about Moneyballing marketing…using math rather than judgement to engineer better outcomes. That is where the idea for the Movable Middle Growth Framework came from.

The best predictors of home runs are exit velo and barrel rate and the best predictor of a consumer’s responsiveness to your ad is their baseline probability of purchase towards your brand (discovered via calculus and proven via case studies.)

Continuing the metaphor…

 

Moneyballing Baseball	Moneyballing Marketing and media to engineer success
Batting average by pitch type and location.	Don’t swing at pitches you can’t hit and don’t waste advertising on consumers who don’t give a damn about your brand.
Splits (e.g. batting average vs. left-handed, right-handed pitchers)	Match creative to target and context.  AI and machine learning can double creative effectiveness. These will be ubiquitous practices in the coming years.
Not letting most pitchers face a lineup for the third time through the order	Find the telltale signs that advertising effectiveness is beginning to wear-out.
A pitcher needs a “plus fastball”	What is your personal fastball in a new age of marketing?  What is your brand’s fastball? (For Amazon, I’d guess it is “shopping simplification”).

When the Yankees first put numbers on players’ jerseys in 1929, numbers matched their everyday slot in the lineup.  Now baseball teams might have a different, optimized, lineup every day often guided by massive simulations that predict performance for different lineups against specific teams, parks, and pitchers. 

Marketing can have optimizers too! In fact, I built an audience optimizer where I have found that optimal advertising allocation across audiences can vary by 20X or more per ID/consumer (optimal means that the last dollar of advertising against each audience is expected to produce the same ROAS result…a marginal utility approach.) We area simulating that 10-90% increases in ROAS are possible through better audience-based ad allocation alone.

The focus of stats in baseball used to be retrospective, “back of the baseball card” stuff. Now it is about engineering success…making different player decisions and adjusting mechanics to create more successful outcomes.

Isn’t it time to Moneyball marketing, going beyond measuring success to making success happen?