AI and Data Can Put a Number on Handicapping "Guesses"

 The NFL announced today a few new features in their data and stats offerings. 

One - tackle probability - I found to be pretty Jetsons. 

Traditional statistics like solo and assisted tackles have long been used to measure defensive performance, but they often fail to capture the nuances of the game. Enter Tackle Probability, a revolutionary AI-driven metric developed in collaboration with the Amazon Web Services Professional Services team and trained on AWS SageMaker. 

Tackle Probability leverages a tree-based machine learning modeling architecture to process millions of data points per game, incorporating 20 different features for each of the 11 defenders every tenth of a second to estimate the chances of a tackle. By predicting the likelihood of a successful tackle after a handoff or catch, the model converts these probabilities into detailed metrics like tackle opportunities, missed tackles, group tackles and more.

So, we're now looking at player movements as a real modelable data point. 

We know the NFL shoves mega-bucks into these things, whereas in horse racing we're wondering if the first-time gelding data is real or memorex, but it theoretically portends some amazing stuff for this sport, right?

We often watch, say, a completed turf race and wonder (and argue on twitter) what a faster pace would've meant to our losing closer. Imagine being able to punch in our own fractions and having an AI spit out the new finish. 

We all have the horse racing disease where results bias elicits the feels. When a rider makes a winning move it's a great move, a losing one is boneheaded. But what are the true probabilities of the move? Was Borel and Street Sense going up the wood a probabilistic losing move, not a winning one, based on his speed and margin at the finish? Could Borel have been dumb for risking being shut off, when he could've won without incident going wide?

Let's take something remarkably simple. Imagine if we had a personal AI model that spit out accurate fair odds after the race based on each horse's position, movements, pace, etc. A back-marker 40-1 shot could've raced like a 6-1 shot with this buried line - this happens all the time - and we'd actually have that quantified for ourselves to use for next time. 

I'm sure many of you as sharp players could think of a hundred things you'd want to see with this data. 

Right now, smart players and sophisticated tools can predict with accuracy some of these things. But we're truly guessing. We're guessing if our horse would've won by four if the jock went up the wood, just like we're guessing that safety Kyle Hamilton would've made that tackle at a 47% higher rate than Josh Metellus. 

The NFL and Amazon AWS and AI tools are taking some of the guesswork out of it for a very popular sport. At some point we'd have to figure horse racing tries to join that party. 

Have a nice Thursday everyone.