The Athletic Football Show: A show about the NFL - The next frontiers of tech in football with Prime Vision's Sam Schwartzstein

Episode Date: July 2, 2025

If you listen to a show like this, you already have an idea of how much modern technology is shaping the NFL. There's a decent chance, though, that its reach into what you watch on Sundays...and Mondays...and especially Thursdays is far greater than you suppose. And it's only increasing. Sam Schwartzstein, the analytics expert behind Amazon's Thursday night Prime Vision broadcasts, joins Robert Mays to discuss the future of football and where the marriage between tech and the sport is headed on this episode of The Athletic Football Show.Hosts: Robert Mays and Derrik KlassenWith: Sam SchwartzsteinExecutive Producer: Michael BellerProducer: Michael BellerSubscribe to The Athletic Football Show...⁠Apple⁠⁠Spotify⁠⁠YouTube⁠Follow Robert on Bluesky: @robertmays.bsky.socialFollow Derrik on Bluesky: @qbklass.bsky.socialFollow Robert on X: @robertmaysFollow Derrik on X: @QBKlassFollow Sam on X: @schwartzsteinsTheme song: HauntedWritten by Dylan Slocum, Trevor Dietrich, Ruben Duarte, Kyle McAulay, and Meredith VanWoert / Performed by Spanish Love SongsCourtesy of Pure Noise / By arrangement with Bank Robber Music, LLC Hosted on Acast. See acast.com/privacy for more information.

Welcome to the athletic football show. I'm Robert Mays. Doing something a little bit different today. A couple years ago, my friend Sam Schwartzstein, who does a fantastic job on Thursday Night Football and Prime, he's their analytics expert for Thursday nights. We talked about the state of football analytics in 2023. And so much has changed with just technology in general in those two years. Think about the phrase chat GPT and how ubiquitous that feels now compared to

couple years ago. And the way that it feels like artificial intelligence is just in the conversation in a way that it just wasn't even that recently. So I wanted to revisit sort of a big picture idea with Sam that touches on some analytical aspects, but takes it even a little bit broader than that. What we're going to do today in an hour is try to figure out the future of football to an extent. We're going to bucket this in a bunch of different ways. We talked about artificial intelligence and analytics. That's the majority of this conversation. What is possible? What might be possible in the near future? How much are teams already using AI and just where is this going to go?

But I also wanted to talk to Sam about some schematic tweaks that he thinks might be on the horizon based on all of the research that he does. A couple of things with broadcasting and where they might be able to take Thursday night football over the next couple of years based on some of the things that are going to become available to them. And we also talked about because Sam loves this stuff, a couple rule changes that he'd be interested in that he thinks might help the league. So rules, analytics, schematics, and broadcasting. We spent the next hour talking about the future of football with Sam Schwartzstein. I hope you guys

enjoy it. It's late June. It's summer in the NFL calendar. And this is the time of year where we do stuff like this. Kind of take a step back, do some bigger picture discussions. The last time Sam and I had a conversation like this was about the same time in the calendar. It was July two years ago. And I felt like enough stuff has changed over those couple years that it was time to have my friend Sam Schwartzstein, the analytics expert from Amazon and Thursday Night Football back on to discuss what the near term future of football looks like because that's exactly what you should be talking about when it's late June. Sam, how are you doing, my friend? Great, yeah. And a lot changed. Last time we had this conversation,

chat GPT wasn't public for everyone to use. And that will be a topic of conversation, but it's changed a lot about how teams and LLN's in general approach. analytics. So I'm so glad that you said that because that's exactly my motivation for doing this in this moment. Just the way that we discuss artificial intelligence and how ubiquitous the term is and just how many teams and companies are thinking about it on a big picture level, I think that we have to kind of readdress some of this stuff. There are great things with AI, as you can see what we try and do on Thursday night football. And I think there's people who have apprehensions, but I think there's been kind of a mysticism that's been spread across

the league that's kind of being debunked on what it can and cannot do. And I think there's a lot of teams that are finding ways that's going to help them even just communicate better when people are trying to discuss topics that either side might not understand 100%. I think that's been the best use case I've seen for it so far. So I was going to end with this, the analytics and data side of it. We might as well just start here because it is the impetus for wanting to do the show. So this is the first thing I want to do with this.

I want you to explain to people that are. football fans, what parts of their football enjoyment experience, whether that's the broadcast they watch, some of the stats that they're dealing with, what things that we've just come to understand as normal parts of football discussion, EPA, certain models, what is, how much artificial intelligence is already built into this? How much is artificial intelligence already shaping the way that you are engaging with the sport if you're a person who probably listens to this show? Yeah, I think some of the statistics, like EPA goes back. to the hitting game of football, the 80s, Virgil Carter,

former Cincinnati Bengals quarterback, being kind of the first person to implement something like this, to equate down-distance location on field and the yards you gain into different values, different parts of the field. And so you can go far back as what maybe like the pre-AI boom of like machine learning. A lot of those stats, CPOE from NGS,

they can take a lot of different value data points, location on field, FID tracking data, running through big models, and then have something to spit it out. I think AI currently used today is kind of this generative AI idea where you're able to ask questions through an LLM and then it's creating something for you. I use this a lot in NASCAR where I worked on NASCAR as well. I would not call myself an analytics expert for NASCAR as much as I did for football. But I use an LLM a lot as a tool to teach me about what RPMs and throttle can do to be able to spit out the miles per gallon on a broadcast.

And using AI to get me to the right passageway, it was a huge value add there. And I think that's where you're going to see a lot of things forward. So if we're watching Thursday Night Football and you're seeing defensive alerts or coverage ID or pressure alert, you're going to see those models are built off of likely what's called a transformer model, which is an AI model that's taken. taking data, a lot of information, running through a neural network, and then spitting out a result. And Robert, I'll send you something afterwards. I'll just explain to somebody else.

It's two input, a lot of different outputs or neural network working together, and then an output that comes out of it. And so there's a lot of different things that go into that. These now are going to be more ubiquitous across teams. We saw the big datable submission utilizing, the one that won, was a transformer model to predict coverages, worked really well, but they took a step further, and they were able to identify which player has the largest impact on whether we'll make a man or zone prediction and coverage. And that was a really cool teaching point to go one step further and say, not just going to give

you the coverage, but who should your key re-be? And that's the next level, start thinking, and that's where coaches can now interact with these AI models. And I think that's a big piece of it is how do I make this something that's approachable for a coach? Because when I was, doing the XFL rules, I was like, everything starts with film, where we want to give them data, where we really want to give them. If a guy needs to go to the doctor for an appointment, it should go through the film portal that the coach sees, right? Everything should be film. And that's kind of that where the coach can go is, oh, am I watching film? And how can it integrate? That's why we have the, the or thebubes on defensive alert or the bug pop for coverage ID.

So that's fitting in the video and not just being a number that they could get somewhere else. The big data bowl submission was fascinating. It was, I might get the, their names are on, but Vesak Sanwar was one of the people who submitted that paper that ended up winning. And if you go and read it, I think they do such a great job of laying out where the use case for this sort of technology lies with coaches. And the last time you and I had this discussion, we talked a lot about that being one particular bottleneck of how analytics can be implemented, where even if you have all these

wonderful ideas, if you can't communicate those in a way where a coach thinks he can implement some of those ideas, then they cease to be important. And the way that they laid out this paper, I'm going to read from it directly because, again, I think that the communication in this is so well done. We believe the power of this digital whiteboard, which allows you to move players and kind of change the probability based on where those players are.

So, for example, they lined up an example where I think it was Nassir Adderly. It was a disguised version of Cover 2 with the Chargers playing against the Jaguars. And him bailing out just a little bit at the snap is what actually gave it away as a zone coverage versus a man coverage. So they said, we think that the will, with the digital whiteboard's ability to experiment with leverage players like Adderley, whose pre-stap move is impact coverage odds. For example, if an offense identifies a handful of promising plays against a particular defense, they can recreate that defense's tendencies in the whiteboard to understand the spectrum of pre-stap

looks they might face. From here, the offense can pinpoint which defensive players are important to read and how their specific movements telegraph coverage intentions. And so they take it one step further than that, and they actually give you a few examples with teams, where they'll look at the bills and one specific slot corner for the bills, the way he was shaded or the way that he would move on a particular play, that gives you a 30% chance based on those movements of understanding the coverage. It's not a perfect science, but a 30% chance based on you being able to zero it in on one player is an incredible advantage potentially for an offense. So of all of the use cases and just the examples of this, I've seen beyond some of the stuff you guys do at Thursday night football, this was one of the easiest ones for me to understand how this sort of AI and how these sorts of models can have a direct impact on game planning and preparation for these coaches and players. And I think you brought up so great as their digital whiteboard. It's like going from MS DOS and command prompts to the GUI because the GUI operating system.

Explain that in a way that a moron like me would understand. So before the GUI operating system was actually developed by Xerox and then Apple and Microsoft created it, but it's now your ability to click Chrome to open up your Chrome browser. You can click an icon that'll do it for you. You used to have to enter it in as a command, like old school rogue-like video games, the first video game rogue, or one of the first video games, Rogue, where you're just playing a quest and it's just typing out what you should do, and you type back what you're supposed to do.

Now we can point and click. That's what this next level of an LLM where you can have a conversation with a model where you can then make your own sequel queries and you don't even know you're making SQL queries because you're just typing in, hey, can you pull up every single time they were, Taryn Johnson was in the slot, but in to the boundary instead of to the field, which was a huge man in Decatur for the for the bills. And so you're now saying, oh, I can pull, I can pull that up versus you used to have to do a lot of different sequel queries or you're doing a lot of different filtering.

Now you just do using free text. So many filters. And so you're now, and it's just, and you're just, and you. might be able to ask the question much easier than understanding the system that you're trying to pull that information from. And the same goes for what their digital whiteboard is, is it's okay for me to say, hey, if he's shaded inside, shade it outside, get that information. But now someone can do it themselves. You know, I was talking to a couple teams this offseason, and they were asking, hey, how can we use coverage ID to teach me coverage? I'm like, how do you need someone to teach you

coverages, but the model is able to find things that a human's not currently caught up to do and predict those different features. Okay, now you can see the aspect of it is, where do I have my conventional knowledge, which is a lot of what we do in the offseason when we're building new models, I'll be a baseline and write something out, and then we have the model trying to see who's going to do better in the results. It's shame to say the model often beats me in some things. And so that's our kind of benchmark. Now teams can do that too. when you're able to go a step further of just not giving you, okay, you lost to the model,

now going, here's why you lost, here's what you can do, and then we get to the part that AI can't do. The AI can't drop the plays or coach the players. That's not going to be what it's going to do. Like when we got on the field and we were practicing reach blocks when I was in center at Stanford, we practiced shade versus two-eye

and what techniques do I have to use. The model can help deliver that kind of information, but at the end of the day, you have to get in the field and do the work to now know how different it is to reach a shade. versus two eye. An example that I think, again, is instructed from that paper about where the model can win and the human mind and eyes just can't.

There was an example that they used of the Bengals having a nickel corner foul a player in motion, which for our like football brains on a very basic level, that gives you an indication that this is man coverage, just based on how the players moved to that motion. But the way the nickel lined up, it was just like a shade inside. and then the linebackers bumped just a tiny bit. And to a quarterback, even one that's very smart, like Aaron Rogers or some of these advanced veteran guys, that's almost imperceptible to the human eye. But the model can see that stuff.

And so the fact that the models are getting this nuanced where it's going beyond rudimentary rules as we would understand them and it's actually digging a layer deeper than that, that's where this stuff becomes really interesting, really fast, because the models are going to be better than you are at identifying. this sort of stuff, especially in retrospect and sometimes in real time. Yeah, I think the way I look at it is we started out doing a lot of if-then statement models, right, rules-based models, business logic. And you're going to get really good at finding, at getting your own answers right.

If a player has within five yards of a line to gain, they have six or more yards between them and the closest defender, they're likely open, highlight with a green orb, right? Things like that. Well, we're not getting every open player, but it's hard to argue when those players reach those criteria that they're not open. AI can go to the next level and say, I'm going to find the right answers for you instead of you trying to create

the answers to the test early. And that's what's been a nice little development. This past year of different tools. Nothing's perfect yet, you know, I think, but as we get every single year, it's funny now, like more and more and more football people are asking about it because either their kids are telling them about chat, GPT, or that they can just now interact

with data in a much different way using that free text model. My understanding, after having some conversations about it this week, to try to get approaching your level to have this discussion, is that the large language model application in the NFL is kind of like what you were talking about, where if I go, and this is on a very basic level, you have more layers to it on something like PFF Ultimate or some of the other databases that exist. But let's say I want to go to like next gen stats, okay? And I want to plug into next gen stats. I want the numbers for what Patrick Mahomes did against a two high shell on third and five or more with, you know, the other team and nickel personnel. I literally have to go in and click every single filter in order to get that information. The hope is with the implication of large language model technology, I could literally say, I want Patrick Mahomes on third down to get this and this and this and then it can spit it all out for me.

would you say that is the best and most realistic application of large language technology for where the NFL is going to be going in the next two years? Yes, that's the top one right now. Again, everything goes back to film. How do I make the film study easier for the coach? I think what I want to see from coaches next is what do you recommend or what's the best plays? And then that's where you can sneak in success rate in EPA.

And you're servicing those plays, right? because, you know, the more conversation I had about EPA with football people this year goes, it really is the football guy stat. It's telling you it's great to convert on third and seven plus, right? It's telling you the value of yards and certain. High leverage downs, baby. It's the situational football stat. Why do we not care?

We love explosives, you know, like, why are we not even more passionate about it? It's everything you guys say. But, like, I think there's going to be a way to where you can start trusting the models to be like this all. seen I and they're going to be able to see things kind of like the way chess players use engines like stockfish where the chess engine will always beat a human right now and there I think we'll get to that point in football where there be people who trust it I'm not saying that's going to happen but there will be a trust level to the models being able to spit out to them

provide you recommendations right hey give me these plays that where the team was successful now what plays do you think we can do based off our team makeup right because that's another part of it is how do we know team strength? A lot of times when we're evaluating decision making, we're saying which team is stronger. Probably the best way to do that right now is looking at the spread. And how do we get the spread? Well, it's the sports books and they don't go out of business. But, you know, cowboy games, Notre Dame games, like, there's betting swings that shift the spread. So, like, are we going to really include that in team strength? Well, there might be better ways to find team strength when they can know the roster, the opponent's

roster, what they've done previously. So I think right now, doing queries and being able to come up with filters that maybe NGS or PFF don't have built in. I was doing a lot of PFF filters before we came in and NGS filters before we had this meeting. And I was like, I don't know where any of this stuff is. And I don't, oh, that's not what I really want. I'll work around it. Hey, there's going to be some things now that teams will be able to look, okay, get the data pulled before a person would even know to put that filter there. There are a few things that I'm curious about, like, where this goes from here in terms of the inputs we're putting into some of the models.

Right. So what we're working with right now with tracking data is you can look at the vector directionally of how a player is moving. But this is on a two dimensional level. Literally there are dots on a page. There's a reason we talk about it in that format. Now that we're moving to the Hawkeye technology with where the cameras are going to be, it seems like we're going to be in a place very soon in the NFL where we're going to have the pose 3D data that's something like the NBA has, which feels like a pretty big game changer for like the type of things that are going to be capable. of. So as you think about how a lot of this information and a lot of this stuff is going to be three dimensional soon, what sort of avenues does that open up for teams and even for somebody like yourself? The teams is technique, right? Being able to know when a guy has a tight base first not type base and all that charted data could be, hey, you know, we're trying to teach an alignment, you know, keep his feet underneath himself. Don't get outside your own body. And we can't, that has to be charted by somebody, what kind of technique they're using. and you can be able to create a technique aspect to it.

Injuries is another amazing spot for a lot of these things. How can we avoid injuries by using pose data to know how players got in certain body positions that then can go into teaching opportunities? Right now we'll see the hip-drop tackle and go, okay, we know this is hurting players, but it's 15 guys, yes, it's a bad injury. But hey, there's a lot of other stuff

that we could kind of do from an injury prevention type of thing. Return to play protocols is a pink thing I'm passionate about. Maybe many of your listeners might not know what the FMS test is, the functional movement screen, but it's a basic test where if someone watch you do about, I think it's seven different, yeah, seven different movements, and then someone goes, okay, you got a one, two, or three. Well, we can actually create an individual player's baseline using pose data, and then as we're doing a return to play, using LIDAR or 3D cameras, you can now be able to get that guy's actual return to play for that specific player. Are they

ready? Are they stable? I think for TV purposes, it's going to be amazing, right? the Odell Beckham catch being able to see the improbability NBA did a great job this year with their dunk score being able to see who had the best dunks and I passed the smell test I think last year

not this past season the year before Anthony Davis is or not Anthony Davis Anthony Edwards a huge dunk and it was the number one dunk on the dunk score it's like oh okay that makes sense we should have Anthony Edwards

the number one dunk passes the smell test I think there's going to be a lot of cool things catch probability being able to understand how amazing these human bodies are in the NFL on what body positions they're able to put themselves in and make big plays.

And then, you know, right now with defensive alert, we don't know if a guy's light-handed, how much weight he has on his front hand or not, if he's going to be a zone dropper. Sometimes we still find zone droppers, which is crazy to me, power of AI, but there's still a lot more we can do by understanding what kind of stances players are in. We don't know if a D-N is in an up stance or a down stance right now.

But knowing that information changes a lot, I'm talking to, you know, mild quarterback, Andrew Luck, GM of Stanford football, buy your season tickets. He makes me say that every time I do a podcast. He provided a good life for me. But he's like, I never run a bootleg into a stand-up d-end. So if we get ever into a play recommendation world, hey, you know, we want to run bootlegs in a stand-up d-ends. That posed a, even something as simple as that would help us see the game a lot more like quarterbacks. I'm so interested.

And you and I talked about this a little bit last time we had this discussion about like where some of the limitations for this exist. And the first thing I want to ask you, it seems like broadcasting in a lot of ways is a good use case for this stuff for multiple different reasons. One, this is cool, is an acceptable answer on a broadcast.

It's not an acceptable answer for how an NFL team operates. And I think that has given you a lot more leeway than somebody within a building might have because I think you feel good about what your models are doing. But if you're wrong,

the downside is not nearly as big, as it is for an NFL coach. And so I think that's why you've gotten to play around in the sandbox a little bit. But getting back to this idea of communicating these ideas, talking to somebody this week who works in this field and AI specifically as it relates to football analytics, I was asking him like how a certain model works. And he was saying to me that as these things get more intricate and more complicated, that question is going to be harder to answer.

And why that matters is that the coaches care about the inputs. and if you can't prove to them the quality of the inputs, they're going to be less attuned to accepting this stuff. So as the models get more intricate and better, but harder to explain, how do you think that is going to complicate coaches and front offices being willing to accept this sort of stuff? It's very funny.

You say this because I've said this a lot this offseason, but the statement I hate the most in analytics is numbers never lie. Because every analytic you see, I want you to see. Just know that. No, if I bring up a number, it's to prove me right. Right? Like, there's a hundred other numbers I could find,

and it's lies, damn lies, and statistics. Like, that is the truth. I think what's important for people to realize is these analytics people on teams are going to have to build trust. That's a big part of Amazon culture. Our stuff is, how are you building trust? Either you're working the hardest,

or you've been right enough times that I trust you what you're spitting out, right? because we had all these different fourth down models. I've talked to teams, how do they make fourth down decisions? We don't use spread. We build our own individual player strength or team strength. We don't use individual players we do. Even on the salary cap stuff, we don't build in the void year information.

We don't add when we do picks average first round pick likelihood to be a running back versus a wide receiver O'Lyman. Right. Like there are teams that do incorporate all that information. And so at the end of the day, you have to be a lot of. to trust the person telling you that information because you can get a number 100 different ways I think Michael Lopez wrote about this a few years ago all fourth down engines are wrong right because they're all spitting out different things there's not consensus among the top 11 fourth down

models so they're all wrong well that's not necessarily case but you have to trust the person that's building or communicating to you so that's the biggest thing that's what I try and do our job is there are going to be things about football that Ryan Fitzpatrick knows that I will never know. I have never thrown a football. I can barely throw overhand. I'm starting physical therapy next week to try and figure out how to do that. And so, like, there's other things that we can't do. We all have to do our 11th and, you know, including the analytics people.

And I think that is just, how are you going to build trust? Because, again, explaining all the high-level stuff to people gets really, really hard. When I did the kickoff, I was going through and explaining every analytic to people about here's why we're doing. this kickoff where the ball is moving and no players are. And those numbers did not matter to the coaches until I got them on the field and showed them this is how it's going to work. And then they still didn't really like it. So, you know, it's, it just takes time and time and time and just got to trust that person.

Ultimately, I built enough trust with them to where they did it. We're going to take our first quick break here. And then I want to come back and talk about how some of these ideas and how some of this efficiency is actually affecting some of the schematics that we're seeing on the field. So I want to talk to you about a couple trends that you think might be coming to the NFL this year practically. But first I want to ask you a big picture question about how the last topic filters into this one. If we're going to do all of this work to make coaching more efficient and make the transfer of information more efficient across the football landscape, it's undeniably become that, right? Like the football world informationally is flat now.

If a team runs something that is effective, thinking about like the sharp motion the dolphins are doing a couple years ago, Sean McVeigh, can pull up every single instance of that that led to an explosive play on his computer the next day in 30 seconds. That is not something that was available to football coaches 15 years ago. So as that process has become more efficient, what elements and what spots on the cycle where offensive innovation, widespread acceptance, and then defensive response, and then how it all goes back around again, what spot on that circle is most interesting to you and who do you think has the hardest job on that circle? I lean towards the fourth down calls.

We're seeing teams go for it more and more than ever before. I did the numbers before this, so it's teams attempted fourth and zero to one. And we used to do fourth and two as short yardage, but now because of the QB sneaks proliferation, it's fourth and one because that's when teams sneak. So zero to one, it was 67% attempt, 73% conversion. Go back to 2015, 41% and a 66% conversion. So we're doing it more and they're converting more. And it's because of the innovations around what teams are doing with play innovation,

specifically offensive play design for situational football. I think that's where this is going to be a huge thing because the tush push is still in. The owner talked about how this is like the coolest thing in the world for him that he has this play. So what is every other owner going to do to their coach? I want my tush push. Make my play. I want my play too. right and I think Stephen Rees showed that it was actually Kansas State years ago that started the tush push but you know he created it in his head I think that's where a lot of this is going to go is specific play designs trick play is used more because teams are going to want to have that information and even before being able to pull up these sharp motions you know it's the number one way to find what trick plays are going to be run did they run them in college football or if you type in coolest trick plays on YouTube those plays often matured

trickling to the top, the YouTube algorithm, because that's what coaches are searching. We have now coaches, offensive coordinators, that grew up in the YouTube age. And that's what they're going to be searching and looking for. So even as basic things like that, but I think every owner is going to ask their team, their coordinator, where's my version of the touch push? I want my play that I get to hang my hat on. I'm thinking of an example. I'm going to struggle to remember the details of this.

So stick with me for a second. It was in like 2012 or 2013. and the chiefs were playing on the road. And I think it was Syracuse, was playing a game on Saturday night while they were in the hotel. And it was that middle shovel pass to Travis Kelsey after going in motion

that worked for, I believe, Syracuse in the Red Zone. And so the chiefs implemented that. And that's exactly what you're talking about here. We're like, that used to be the way that this would happen, where you would see it happen somewhere and then you would be able to implement it.

So now that all this stuff happens so quickly, do you think that the more important distinction is going to be between the offensive coaches who are quick to accept and understand how that information transfers, or do you think it's going to be between the defensive coaches who do or do not do it? Where do you think the biggest gaps are going to start to exist between people who are fluent in how this stuff goes and people who struggle to be fluent in it? Defense is so reactive, in my opinion, and I'm going to have some defensive guys that tell me, oh, no, you're wrong.

But I just feel like you have to react to what the offense is doing more. and it's you know right now it's just building to stop the pass or how do we build throwing to or beat throwing to spots where spot coverage has come back in the NFL versus match coverage we see cover to invert being so much more valuable now because you could play that that three high look but then come down and play too high or the one high look and come down like there's a lot of stuff that defense just have to react versus play design is going to come from offense and i'm I'm interested to see how what the transfer portal and the use of spread option, spread offense, where I think spread offense is going to be more and more used across football,

not just because it's easier to get the fast guys on the field and spread the ball out, but it's a lot easier to teach it when you're teaching guys in eight weeks of their offense when they come in for training camp, right? And so I think that what effect is that going to have on way these offenses are, and then how can they be a little bit more unique in their play design to then come up with the new thing to beat the defense that they'll ultimately have to react to. I spent some time with the Stanford Offensive line this past off season and training camp. And we're we're, we're trying to figure out how to beat the bones front, which is what they call the 6-1 from

Fangio, 2018. But now it's making its way to beat all zone teams. So it's one of the only times you see it coming back down in the college football. And so like we're trying to figure out the bones front. And you've got to go and look for all the different bones front six-one stuff you're trying to beat. So I think there are, there's more things about an offense that they'll have to do to make the defense step up versus the defense the other way around. I think that makes sense. And I think fronts is a really interesting example on the other side of it.

And that's why to me, and I, this is kind of revisiting what we talked about before, but I do, it's one thing I wanted to ask you about. In my opinion, as I think about this with no knowledge of what is possible and no knowledge of like how far this stuff can go. To me, my favorite application of the sort of. of player tracking and just like the models that we might be able to use is manipulation of pass rush rules based on how your front is lining up. So if I'm in a 5-0 look and I have a linebacker walk down into the B gap, can I figure out by some of the models and just all the

things that at my at my fingertips here? If I drop player X, how does the offensive line react? I think that's something that feels achievable and it feels like the best situation where the defense can be the ones dictating to the offense rather than the other way around. So do you think that some sort of model where you can establish and figure out past protection rules based on how teams react to which players are coming and which players are dropping is something that might be in the realm of possibility for NFL teams in the near future? Absolutely. I think what's so hard is when we're thinking about those plays could happen six times,

third and seven plus or, you know, if teams start approaching. like and and then it's game plan specific type things i think one of my favorite dumb highlights ever is uh Travis frederick pointing in the double mug to the guy to the right and then sliding to the left yeah just to just to trick them and then the next development is okay we're both rush we'll do a modified green dog where the guy whoever the slide postnap is then they take the drop afterwards but like like i love that i was like for me playing center i was like that's so cool that's the best though like how Travis frederick is playing against mike zimpererner is like the ultimate football nerd nonsense from 2014.

Yeah, I was like, I was like, but like, you can do that? You can point someone and slide the other way. Like, like, I was like, I never even thought of that. Wow. I did a dummy call, but I didn't point that. And so it's like those stuff exists. I think those minute scenarios about who to get where. I think showing if a team has a tendency to slide to the field,

to slide to a certain player, are they going to slide to my star, right? If teams need to start, you know, it's so funny going from Aaron Donald and now teams might have to slide to verse, right? Like the Rams, just the riches that they get at the D-Line spot. And so it's like, do I have to slide to an individual player? I think that those tendencies and then trying to predict what the team's going to do, I think that post-snap stuff of being able to manipulate what's going to happen post-snap to where, oh, if a team is going to do a half-slide scenario, there is a weakness if your guys are fast enough to take a step forward and then bail out. so those into your linebackers. But I do have a note here. You ask me for schematic changes.

When's a team just going to run QB power into double mug? That's all I want. They can't run it anymore. I don't care if it's third and eight. Just show them that you're willing to do it. That's been my thing is like I hated double mug look, but at least at Stanford we would check into a run just to like just to let you know we're still here.

Like it's still possible. Well, now that we're in a world where Fort Downs are just, it's a much more acceptable way to live. it just seems like third runs on third and seven are just going to become more prevalent in the NFL and should as we move into a four down world. You're starting to see that already. Like, you know, the Lions do a ton of that. I feel like the Eagles ran the ball in third down a bunch last year.

Arizona was number one. Yeah. Not surprising at all. So let's get into some of the specifics here. Running quarterback power into double mug and running on third and mediums because you're going to go forward on fourth down anyway. Those are two of them. Give me a couple more schematic trends that you are looking out.

for as we head into the 2025 season. So it's a little schematicy, but the thing I'm interested in is these reclassification of positions. As we become so analytical, oh, they're in 11 on 12 personnel, which when I played, we didn't call it that. We called it Tiger or U, or we had the word for it. And what you could do with that is if we called Tiger, that meant normal two tight ends. But if we wanted to put Kobe Fleener at the Y and Zach Ertz at the U, we'd call that Panther. And so it's, it's, it's, to the defense, they see a different personnel grouping, but we now switched players. And for us, there's a whole variety of plays we could run out of that. And I think that's going to become

a little bit of a trend. And I see it with these tight ends. Brock Bowers, you know, listed as tight end. He's detached from the formation 58% of the time. The average tight end is attached 35% or detached 35%. So is he really a slot? Well, he's only technically in the slot 40% of the time, but then he split out wide. And I think we can see that with a lot of players of when I'm getting these tweeners, how do I manipulate what they're actually classified as? So that way the defense then is going to have to react differently in their personnel packages. So give me an example. Like what's an if then that applies to what you're talking about?

So we're going to line up against, we're a defense. We're going to line up against 12 personnel and base defense every single time. Now, why is that important? Because your nickel is better than your third linebacker now. that just we're paying it's going to be paid differently nearly and your third linebacker is not nearly as good because that's who and he doesn't get the same reps in practice Mike Asicki this is a Keegan Abdu he made this presentation a couple years ago Mike Gisicki he never lines up attached to the formation so he's a slot wide receiver the entire time

but he's listed as a tight end and so if a team's not really paying attention they're going to say we're going to match 12 with base every time now Mike like his tiki has a linebacker on him, right? But you could go in and you could say, well, if Pooka Nakua, if it wasn't for the franchise tag, if he's going to be attached to formation as much he is, or Luke or Ben Skoronic, or no, Ben Skoronic lined up at fullback in half his games, right?

Why can't we list him as a fullback? He runs in, he's wearing the 40s number, and then he's lined up in the slot. And these misful classifications of players and trying to put them in different archetypes, always a tight end, but he's always playing out wide. Zach Ertz, a good friend of mine, they would go one by three with him

out wide and 11 personnel the year they won the Super Bowl. Well, is he really a tight end at that point? Yeah. Right? And are we going to match personnel that way? And how do I try and misclassify or reclassify these players to try and change how

personnel matching from the defensive point of view? Because that's what I've seen a lot of is if you can get a personnel mismatch best off of base versus nickel, then you can win, whether it's a runner pass. And a lot of teams are making, it used to be box count. Original spread was box count.

If they have six in the box, we pass, five in the box we run. Well, now it's personnel match. Did they match personnel or did they not match personnel? You're going to see teams like the Broncos, I feel like, where they switch personnel every single play to try and make you switch every single play to try and get an advantage. And then we can catch you off guard. Yeah, it's so interesting. The Gassiki example is a really good one.

I was going back and rewatching the Bengals this week. And they do a lot of that, where he'll be. He'll be in a one by three is the isolated receiver. And like, what are you going to do with those other three guys on the other side? How do you match that? And to me, the best example of this is when we're talking about dictating how the game is going to go, whereas the Niners and the way that the Niners use 21 personnel is their best players were 21 personnel. But at the same time, it is 21 personnel is the easiest way in the NFL to dictate what the defense is going to do against you.

Because now, if you look at the personnel and the percentage, is in 12, it's like 50% nickel, where 10 years ago it was 75% base. Now, 21 is the quickest way to make sure I know I'm going to get three linebackers on the field. Well, that's what they want you to do. And if you're not as worried about them running the ball against you, should you play into their hands to that extent? And I think that little cat and mouse, as more teams try to adopt those ideas, is really,

really interesting. I mean, even thinking about listing Patrick Ricard as a fullback, and I love doing that you know, he's a sixth alignment. Yeah. Right? He's going out for passes barely. And so it's, you know, we have all this jumbo data in six alignment. Oh, the bill's led.

No, Patrick Card is a sixth alignment. And so, you know, as a data person, I like, oh, I have all this data. Well, if you're classifying it wrong, that's worse than having the data itself or not having the data at all because you would see the 300 pounder out there and you'd have a coach go and flow up or their jumbo formation to the card to everybody. And so even just something like that, like you're now manipulating the team to think, oh, it's just just 21. No, it is not. When he's on the field, it is real 6-0 linement.

Any other schematic trends that as you've kind of dug into stuff this off season, you know that you're going to be looking for when things start up in a couple months. This is super niche. I've shown you a couple clips. Have you listened to this show? The something I'm spending a lot of time in is understanding the O line core. quarterback relationship and how it is. And we have all this, oh, sacks are are a quarterback stat. Pressures are now a quarterback stack in my mind as well, just because of how we classify. You have to have the ball on your hand to be able to be pressured. So if you got rid of the ball,

you got pressure. There's also, if you're watching a bunch of unpressured scramble clips, you'll find guys that avoided a pressure before it could happen. O-line got beat. Yeah. And then he avoided the pressure. So, but he doesn't get, the whole quarterback doesn't get credit for that. Or he holds on the ball too long, and then a guy gets pressure. He's not standing where he should be. Where's the spot? The QB spot. Where do they want to launch the ball from? And I think there's been, I think this happened in the Super Bowl where there was so much Matt Nagy catch it and throw it from the chiefs that the D-Ns for the Eagles didn't run the hoop. And so traditionally you're running up the field and then bending the hoop at about seven and a half

to try and get to the quarterback at about eight yards. Well, they were just taking a straight angle at five yards. And that was turning the hip open for the offensive tackles. which made Mahomes have to scramble faster because they're running so much catch and throw and he's not taking the drop. You'll see a guy like Bryce Young who drops deep just to be able to step back up to his spot and knowing that it's not on just the moment

the quarterback throws the ball, but his pathway to get to his spot and footwork matters a ton. I think teams are going to take a longer, harder look at how quarterbacks are dropping now that the three-step drop doesn't exist anymore because it's just everyone's starting in gun. And is that going to affect how,

quarter or de lineman rush is a lot more bull rush a lot more just straight angles versus the traditional up the field bend the hoop that that's been my unique one that i'm like do i have something here i don't know if i have anything but that's what i think i'm thinking about right now that to me going back to again this beginning of this conversation being able to properly communicate the idea of pocket health and how it affects offenses is something that i'm fascinated by because i'm I think that all of these places, like next gen, PFF, they're doing the best they can to try to articulate and explain to you how much pressure is happening and how it's affecting the quarterback. But there isn't a perfect way to do it because of the way that we classify pressures and because of what tracks as a pressure from the tracking data. So this idea of how does the pocket health get maintained and do we have a metric that goes along with that?

I think it would be so important to understand why some teams are successful and why some teams are not. You can get there to an extent. You can look at the pressure rates for the 2024 Texans and look at the amount of unblocked pressures and etc. And you can figure out, all right, this is why they were bad. But I still think there's another layer to this because pocket health and just the comfort level of the quarterback in the pocket, to me is the biggest indicator and dictator of whether teams are going to win or lose in the NFL right now. The best example over the last few years, in my opinion, is like the Rams. If you look at what the Rams pocket health looked like in 2023 and even last, last year, you can explain the difference in what their offenses felt like.

And so us getting to a better understanding of which teams are maintaining the pocket and which teams are puncturing the pocket in the most effective ways, I truly think that is the quickest route A to B to understanding why some offenses are better than you think and why some offenses are worse than you think. And it's not totally explained by pressure rate. Yeah. And I think something I was probably a little too hard on Bo Nix at the end of the year, going through a lot of the plays and rewatching all of their unpressured plays and I was saying oh he needs to get rid of the football he's escaping

healthy pockets so often but in reality nobody was open you know they were spending the least amount of money on skill position players last year and it showed when teams have played man coverage and just shut them all down and O-line can be as fantastic what possible and so I think connecting the two together is where I think the holy grail is knowing is anyone open at these timing intervals in which the quarterback should have been healthy.

And then we get you to defensive vulnerability. Where should your eyes have been as a quarterback? I think that's probably the most fun I have on our show is we're trying, we're learning football as we use these AI tools that teach us new things, but also hopefully we're able to teach other people new things about the game of football because we are able to do the fun stuff like you said, the on-screen visualizations that help you now see the game the way the quarterback does. And I think that's part of the problem with like somebody in my position

when we're trying to talk to football fans, quarterback situation and being able to properly explain quarterback situation is incredibly hard because we don't have catch-all metrics for it. So what you're saying here where, okay, when the quarterback gets to the back, the top of his drop, what does the protection look like and who's open? Well, when you watch people that are explaining film or going through film, that's where they pause the tape.

But that's truly one example. So unless you're watching the tape with that person and trying to extrapolate that example, there's no way to properly explain to someone, this is a good quarterback situation, this is a bad quarterback situation. So trying to use some of the information at our disposal to better explain what is good and what is bad

that these quarterbacks are being dropped into, I just think it would be the best possible way for the general fan to understand why the NFL season is playing out the way that it is. Yeah, I 100% agree. And then quarterback impact, attribution, all that stuff can be extremely, useful. And I think

being able to do it and help teams understand where that attribution could come from or as teams start wanting to engage more, I think that's extremely helpful because I think this is just going to get more and more interesting. The players coming into the NFL now have grown up with people tweeting go for it on fourth down. The ages of people, every player's played Madden. They probably still play Madden, right?

And so I think it's just going to make it even more exciting for teams to be able to leverage data in these conversations. All right, before we move on, we're going to take one more quick break. You just mentioned the idea of what fun stuff you guys can do on Prime Vision and on Thursday night. You've been doing this for two years now. You've had two full seasons. I'm sure you understand the contours of what is possible and where the football broadcasting world might be going here in the near future. As you think about what that direction looks like and what is achievable and what sort of things are available to you right now that maybe you wouldn't have

even understood a couple years ago. What about the way the football is being presented on TV excites you the most as we look at it moving forward? I really love this kickoff where nobody moves. It's fantastic, really. No, I think there's some really cool ways that we're trying to visualize and work with people on the on-screen visualizations, whether or now it's the camera angle, which we had our first half of a season where we were behind the quarterback, emulating traditional video games and being able to see the play

from the quarterback's point of view. I know there's a lot of people who love the sideline point of view, but we got amazing feedback on this. Now, and the camera's going to be even better and tighter scrambling quarterbacks. We know we have to fix what it looks like on our prime vision when the quarterback scrambles

and we're up above. So we are really excited about what that view helps us visualize. You brought up pose data and the camera, computer vision, Maybe there's some other cool things. And that's why I kind of love about this. It's like there's so many unknown things now.

If we could get pose data, 3D recreations, all these different aspects of the game. I think, you know, a funny one is, you know, typically when you ask people, like, personalization. I can want to plug my fantasy team into your show. And I can have my fantasy players. Well, I feel like we solved that a little bit by if you have Pooka Naku on your fantasy team and he's on the field, we highlight where he is on every play. So I feel like we get there a little bit. It funny goes back to a story. My dad didn't know anything about football.

and my brother was in the stands with him in high school and he's asking my brother every single play hey where's Sammy where's Sammy which I didn't really move that much as an offensive lineman so why is he always asking. So my brother he was talking to the coach like I want to spray Sam with this film this spray and then my dad's going to have polarized lenses and he'll find Sam that way

they're like you're not coming you're not spraying your brother before the game so that your dad can find him so it's like you know that's like the first ever iteration of a player tag I think my brother gets to have credit for that. And so I think personalization is a fun one. A lot of people have tried to do it. I think it's, I think it's definitely a route that people will, will want to see. Because then as streaming works, it's not just a delivery system. It's you have a profile, you have all this information. I think

that's a cool way to ultimately watch the sport. And that makes a lot of sense. I think that as things get more specific, more personalized, more granular, and more people even have rights to do these things, you're going to have more options to have it feel like it's for you. I think that totally tracks. In terms of like the audience response to some of these things, I have a feeling that you're like even more encouraged by people's appetites for this stuff than you probably would have been when you started, right? Like the fact that it seems like everyone is excited about these things, I can't imagine how rewarding that's been for you guys to kind of give you more energy to keep pushing because the response has typically been so positive. Yeah, we had no idea of defensive alert.

and even defensive alert, like there was times where like, let's restrict it. Let's only show it on third and seven plus, you know, our feature to highlight blitzers. And it was just going to be a prime vision only feature. And then there was so much positivity.

Like the play that went viral was a cornerback blitz on a run, which I didn't expect to ever do that. It was a second down, second and short. And I'm like, wow. And then there's been positive feedback. I think what's been great is also learning what doesn't work. You know, there's a lot of stuff that we had year one. When we had the L bar,

I don't know if you guys remember, the L bar. We had stats and we made your screen smaller. We got rid of that. What I love about our show is it's never the same show week to week. We did drive focus last year. We liked it. And then it didn't really work later in the year.

We've done, I had an idea to have a tether between every defensive player, the closest defensive player to the wide receiver I thought they might be matched up on. And we just had these lines all over the screen. They're crossed over. And it's like, no one go back and watch Prime Vision, Buccaneers versus Lamar Jackson. a really good idea to do Tom Brady versus Lamar Jackson, Sam,

to display a new idea that's going to paint the screen. And so, like, there are things that we try and, and what I love about the Prime Vision community is, they're going to give us honest feedback, say they love things, say they don't want things, and we're going to work around it, and we're going to work with the audience, build in public is what we call it,

and we're going to build in public and figure out what works and what does it works, and what's nice is we have a team that wants to get these things to the main broadcast. We have to make it work and prove it out, and that's a lot of work. But we can make stuff end to the main broadcast. And then now defensive alerts is in front of 24 million homes watching a wildcard game,

which is crazy to me. The last thing I want to ask you about, and this is driven in part because I'm honestly shocked when we do mailbags and we solicit questions, how many questions we get about rule changes. We get so many of them where people are like, if you were the czar of football and you got to dictate what sort of rules were changed, what would you do? And I'm just the worst person to ask about that because that's not how my brain works. Like, I am not a disruptor. Like, my entire life has been, how can I live within the status quo and just get where I want to go?

That is not how your brain works. Your brain works in the exact opposite way. So you have carte blanche here. I'm going to ask you the same question our listeners ask me all the time. If you could implement like two rule changes beyond the ones you've already influenced in the NFL, what would you like to see and how the game would be changed? The coach to player communication is what I'm super passionate about. And I'll say this, I had no cutoff switch.

And so for the people listening at home, coach to player, coach to quarterback communication, it's only one to one on offense, one on defense. And it shuts off with 15 seconds left on the play clock. So only have 25 seconds to communicate. I had it in every player's helmet or every skill position player's helmet from a cost perspective. And at the XFL you're talking about. At the XFL, yeah. In 2020, in 2020.

So I don't know what they're doing in the UFL now. But in 2020, I had it in every offense player's helmet. We could also put the play call over the air. So we could teach people football to hear two jet squire. And then Greg Mackor, I can tell you, they're going to throw it at the goalpost on a, on a dig. So like that was like a cool thing that we could do. But I like coached player communication setup because I think it's really good for teaching. And you can now teach and teach.

communicate to everybody, not just the quarterback. I think sideline video, we're at sideline pictures right now, sideline video could be helpful. I know why they don't want to do it because they want to have that skill aspect part of the game, but I think sideline video will help us see more and more creative adjustments throughout a game. And I think that's what people love is when I was doing the XFL, it was, what do you love about football, innovation, and strategy?

And those are the two things that we always leaned on. So those are probably the things I would do. here's the crazy, crazy, crazy one. If they go to a 20-game season, go to 12-minute quarters, and you're going to have 960-minute game minutes versus 1,080 game minutes. So you actually play one in a third fewer games, but this way the wear and tear on the player's body isn't the same.

Oh, man. As it now records, total points scored, all this stuff goes out the window, but we could go faster. Well, it doesn't matter now. Now it's all changed because we're playing more games anyway. So points scored, whatever, points per game. Again, we've already muddied the waters in terms of like what a 2,000 yard season would mean as a rusher. So why does it matter if we go to 12-minute quarters if the player safety thing is helped in that way,

if we do get a little bit of an edge and a leg up there? Yeah, and I think you could even manipulate the end game rules as a funny story. I had inside two minutes, I changed the rules drastically to make it easier for the team losing to come back. I had a 25 second play clock that started on ball spot, but I gave you the play clock didn't start until it was 20 seconds. It was very confusing to the viewer at home, but ultimately you couldn't kneel it out until there was a minute left in the game versus two minutes. That was like the main goal of it. And so I had a coach tell me comeback period. Does that mean I can only run comebacks?

I go, what do you mean is like 18-yard stop routes? I was like, no, it's just, it's, it's, it's from video game logic. It's imagine you're in last place a Mario Kart and you got a blue shell or a star. That's what that's what I tried to do. And I think you could do stuff where you could go to a three-minute warning or try to that end game to try and make it so that there could be closer games if you went to the 12-minute quarters. But I just think that if we go to 20 games, I think that would make a big leeway with the players.

Again, this is why I'm asking you this question because my brain. You could live my life a thousand times, and I would never go to any of these places. So I'm very glad you're here to take us to them. Sam Schwartzstein from Thursday Night Football and Amazon Prime. Sincerely appreciate the time. Always great to chat with you. Let's hopefully do it again soon, man. Absolutely.

All right, guys, that's all we got. Thank you so much to Sam for his time. We're going to have a couple more shows coming your way this week. In the meantime, really appreciate you guys listening. We'll talk to you very soon.