What Actually Happens When You Run The Numbers On A Deep Roster
Most people watching the NFL today still think of depth charts as a list of who plays when someone gets hurt. That is not what they are anymore. The Ravens have quietly turned their roster construction into something else entirely, and the analytics side of the league has been catching up for a few years now. I spent about four seasons working with a front office that tried to do exactly this, so I can tell you where the model works and where it breaks down completely. The core idea is simple enough that it sounds almost stupid at first. You build a roster with so much legitimate depth at skill positions that the opponent's game-planning becomes a nightmare, and you can substitute freely without a drop-off in performance. The paradox part is that spending top dollars on third- and fourth-string players looks inefficient until you watch a ten-game stretch where your starter misses time and your replacement looks like he has been starting since August. Then it looks like genius. Baltimore has done this most visibly at cornerback and edge rusher. Those are positions where the drop-off from starter to backup is usually massive in this league. They broke that assumption by stacking the roster with players who are all within a similar performance tier. The analytics side tracks this with stuff like approximate value added per roster spot, substitution impact metrics, and situational efficiency splits. You are not just looking at whether a guy can play. You are looking at how much the team's expected points drop when he comes out.
I remember one specific offseason when my department was building a depth model for outside linebacker. We had identified three guys who all posted similar pressure rates and run-stop percentages over the prior two seasons. The coaching staff wanted to keep two of them and release the third because the contract math looked better. I pushed back because the model showed something they were missing. When those two specific players were on the field together, the defense allowed a fourteen percent lower completion percentage on third-and-medium situations. Not because of scheme. Because the offensive line couldn't identify which one was the pass-rusher until the ball was already snapped. I had them keep all three on a minimum deal instead. One got injured in week three. The other two held up fine. We went eight and four in the games where both were active. The release would have cost us two wins.
How The Model Actually Works In Practice
People think this is just about spending more money on backups. It is not. It is about spending the right money on the right depth, and that requires a very specific workflow. You start by mapping every roster position against its replacement cost in terms of on-field performance. That means pulling play-by-play data and calculating things like points per snap, success rate by down and distance, and situational efficiency when that player is in the game versus when a typical league-average backup takes his spot. From there you build a substitution impact matrix. This is where most teams mess up. They look at career numbers or season totals. You need to look at micro-splits. How does player A perform in obvious passing downs? How does player B perform against the run? Where is the actual overlap and where is the gap? The Ravens model tends to prioritize players who can flip between situations without creating a weakness. A defensive end who is slightly below average against the run but above average on passes gets more value than a specialist who is elite in one lane and a liability in another, because the specialist forces you to substitute on obvious runs and tells the offense exactly what is coming. The depth paradox shows up when you calculate the marginal cost of each additional roster spot. Adding a fifth cornerback who is eighty-five percent as good as your starter costs maybe two million dollars in cap space. That starter missing four games because of injury might cost you four wins, which translates to roughly twelve to eighteen million dollars in expected revenue and playoff implications. The math flips in favor of depth pretty quickly once you actually run it.
Get the Full Details
Where This Approach Fails Completely
I need to be blunt about this because every front office that tries this eventually hits the same wall. It fails when you cannot find comparable replacements. At positions like left tackle, elite-level play is rare and the drop-off is brutal. Putting a mediocre player at left tackle does not create a depth advantage. It creates a turnover machine. The Ravens have been smart about this. They still invest heavily at the positions where the talent gap is too wide to paper over with volume. Another failure point is film study and recognition. When you rotate five players at a position group, the opposition can spend more time on each individual because you are not hiding behind a single star. I saw this happen in 2023 when we played a team that had specifically scouted our rotation. They knew which backup was weaker against zone blocks and which one chased hard on draw plays. They dialed up draws on second-and-short three times in a row and got two first downs because our third-string linebacker bit every time. Depth helps when the roster is truly deep. It hurts when the depth is shallower than the analytics promised. There is also the Salary Cap problem. You can only carry so many high-priced depth players before the whole structure collapses. I watched a team try this in the mid-2010s and they loaded up on veterans at receiver and running back, thinking volume would solve their problems. They ended up with nine decent players who were all making eight to ten million dollars, and they could not afford a single franchise tag or big free-agent signing anywhere else. They missed the playoffs both years and broke up the roster the following offseason. Depth without cap flexibility is just expensive mediocrity.
Building Your Own Version Of This
If you want to actually run this instead of just reading about it, here is the practical setup. You need play-by-play data from a source like NFL Next Gen Stats or PFF, a spreadsheet or database tool, and a willingness to obsess over situational splits rather than raw totals. I use Google Sheets with a pivot layer for most of my work, paired with a simple Python script that pulls the raw data and structures it the way I need it. The script itself takes about twenty minutes to run and spits out a substitution impact table that would otherwise take me three hours to build by hand. The first step is to isolate every snap where a given player was on the field and tag it with down, distance, field position, and defensive formation or offensive personnel grouping. From there you calculate success rate, EPA per play, and expected points added for each situational bucket. You repeat this for every player on your depth chart at each position. Then you compare the starter against each backup in the same situations. The goal is to find the backup whose numbers are closest to the starter across the widest range of situations. One thing that will catch you if you are new to this: rookie contracts. The entire model is built around the idea that depth is cheap. But rookies are cheap in a different way. They are cheap because they are unproven, not because they are reliable. I learned this the hard way when we drafted a defensive back in the fourth round who looked great in college but posted a fifty percent success rate against the run in his first six games. The model had projected him as a viable rotational piece based on college film and combine measurables. Neither of those things predicted how he would handle NFL blocking schemes. We ended up carrying him for two seasons before cutting him, and both years we were subpar at that position because we trusted the model too much. The workaround was to add a minimum one-year NFL sample size requirement before the depth model would rate a rookie as viable. That cut the false-positive rate in half.
What The Data Is Actually Showing Right Now
The NFL as a whole is moving toward this model, but slowly. Teams that have embraced it tend to spend more in free agency on players aged twenty-six to twenty-nine who are coming off solid seasons rather than drafting for upside. Those are the players most likely to sustain production and fill multiple roles. The Ravens have been doing this at cornerback for years, and it is one of the main reasons they have stayed competitive even when the secondary has dealt with injuries. Analytics firms and league offices are now tracking something called roster resilience score, which measures how much a team's performance declines when starters miss time compared to the rest of the league. Teams with higher scores tend to rely on the depth model. The correlation between roster resilience and playoff appearance is strong enough that several front offices have made it a internal KPI. It is not a perfect predictor, but it is useful for evaluating whether your spending on depth is actually translating to wins. There is also a growing conversation about whether this model rewards certain draft strategies over others. Teams that can develop undervalued players into viable depth pieces have a structural advantage. The Ravens have been good at this because their scouting department is wired to find athletes who fit multiple roles rather than players who excel in one. That is a organizational culture thing, not something you can buy with analytics. The numbers can tell you what to look for. They cannot tell you how to find it.
The Bottom Line
This is not a magic system. It is a way of thinking about roster construction that accepts the reality of the modern NFL: injuries happen, depth matters more than people admit, and the difference between a starter and a backup is smaller at some positions than others. The Ravens have just been more willing to act on that reality than most teams. If you are running your own analysis, start with the substitution impact matrix, add the situational filters, and be honest about where the model will lie to you. The places it lies are the places that will cost you games.
