How NFL Teams Use Analytics to Make Decisions
Analytics in the NFL is less about robots calling plays and more about better questions: when to go for it, whom to draft, how to spend a cap and how to keep players healthy. Here is what it does and where it stops.
Say a team trails by four points with a fourth down and two yards to go near midfield. The head coach has a choice between punting and going for it, and in most stadiums he would have been criticized for the second option a generation ago. Today he has a staff member in his ear with an estimate of how often teams convert from that spot and how the game is likely to change in each case. Whether he follows that estimate is still his call, but the existence of the estimate is what people mean when they say a team "uses analytics."
The word is used loosely, and it carries baggage from both sides: fans who think analytics means a spreadsheet is calling plays, and fans who think it is a gimmick that cannot see what the film shows. The truth is more ordinary and more useful. This guide explains the kinds of questions NFL front offices and coaching staffs ask with data, the tools they use, and the limits that keep analytics as one input rather than the answer.
What analytics actually means in football
In an NFL building, analytics is a set of methods for turning information into better decisions. The information comes from several places: the play-by-play record of every snap, charting done by staff who watch film and tag each play with details, and tracking data that follows the movement of players and the ball during games. The methods range from simple counting to statistical models that estimate the chance of winning at any moment.
What separates analytics from ordinary stat keeping is that it asks a decision question. A box score tells you what happened. An analytics group tries to tell you what to do next time, and how confident it is. That makes it a support function, much like the video staff or the medical staff, working for the people who make the final decisions.
A good analytics department also spends a lot of time on a less glamorous job: checking data quality, cleaning it up, and explaining it in plain language to coaches who have limited time. A model that no one trusts or understands changes nothing.
Fourth downs and other in-game choices
The most visible use of analytics is in the decision to go for it on fourth down. The reasoning is straightforward. A team can estimate the probability of converting from a given yard line and distance, and can estimate what the game looks like if it succeeds or fails and what it looks like if it punts or kicks a field goal. Weighing those outcomes gives an expected result for each option. Many teams now go for it more often than coaches did a generation ago because the numbers suggested the old habit was too cautious.
Other in-game decisions use the same logic:
- Two-point conversions. Teams use charts that suggest when it makes sense to go for two instead of kicking the extra point, based on the score and time remaining.
- Timeouts. A model can estimate how much a timeout is worth at a given moment, which helps coaches decide whether to save one or use it.
- Play calling tendencies. Staff study how often an opponent runs or passes on a given down and distance, and use that to guide the plan.
- Clock management. Late-game situations, such as whether to run the ball and drain the clock or keep passing, can be modeled.
The best-known tool behind many of these is the win probability model, which estimates the chance a team will win based on the score, time left, field position, down and distance. A related idea, expected points, estimates how many points a team can expect to score from a given down, distance and field position. Both give a common language for comparing choices. For the underlying measures, see our primer on success rate, expected points and efficiency and our guide to fourth-down decision making.
A worked example, kept hypothetical
Imagine a team facing fourth-and-two at its own 45 with a lead of three points in the third quarter. The coach has two main options: punt or go for it.
Suppose the staff estimates, in this hypothetical, that the offense will convert about half the time. If it converts, it keeps the ball and has a strong chance to build the lead. If it fails, the opponent takes over at midfield with a short field. If the team punts, the opponent starts deeper in its own territory, but the offense gives up a possession.
Analytics does not decide this. It frames it. The staff will say, "On average, teams in this spot gain more by going than by punting, but our offensive line has been struggling against this defensive front." The coach then weighs the data against what he has seen in practice and in the game. The numbers are inputs to a judgment, not substitutes for one.
Draft and scouting models
Analytics also shapes how teams find players. Scouting has always combined film, testing and background work. Data adds layers.
- Athletic profiles. Teams combine measurements such as height, weight, arm length and combine results with college production to estimate how a prospect compares to those who came before at the same position. These comparisons give a starting point.
- Production models. Staff adjust a player's college statistics for the quality of competition, the role he played and the system he was in, to estimate what part of that production belongs to the player.
- Position values. Analysts study how much a position contributes to winning and compare it to its cost, which informs how teams think about spending draft picks and salary.
- Pick values. Teams use charts that put a value on each draft slot to evaluate trades. See the draft trade value chart.
Yet the draft is a place where data is least decisive, because players are young, circumstances change and sample sizes are small. A model might say that a prospect's profile resembles successful players, but a team still needs scouts to tell it whether he loves the game, handles pressure and learns quickly. See the NFL Scouting Combine for the standardized measurements that feed these models.
Roster building and the salary cap
A front office has two limited resources: roster spots and money under the salary cap. Analytics helps spend both. Staff track how much a player is paid compared to what his production is worth, how contracts age as players get older, and how much cap flexibility the team keeps for future seasons.
One use is evaluating how players age at each position. Some positions decline earlier than others, and a team deciding whether to sign a veteran to a long contract can compare that player's age to historical patterns. Another use is comparing the value of a free agent signing with the alternative of promoting a younger, cheaper player.
Cap analysis is also about structure. Contracts can be written in different ways with different guarantees and cap hits each year. Analysts model how a contract would affect future flexibility. For the rules these models rest on, see the salary cap explained and how NFL contracts are structured.
Game planning and opponent research
During the week, analysts and coaches study the opponent. Staff chart every play from past games: the formation, the personnel on the field, the direction of the play, the coverage the defense used, whether it blitzed and what the result was. Over many plays, patterns emerge. An offense may favor a certain play from a certain formation, or a defense may play a particular coverage on third down.
The staff also studies the matchup. If a team's pass rushers win more often against a certain kind of blocking, and the opponent uses that blocking, the plan will account for it. If the opposing quarterback struggles under pressure from a certain angle, the defense will try to bring pressure from that angle. This work overlaps with film study, and the best staffs treat the two as complementary. See building a game plan, week to week and film study: what coaches and players look for.
Tracking data and player health
Many NFL teams now have access to tracking data collected with sensors that record the position and speed of players and the ball many times per second. That data can tell a team how fast a player ran, how far he traveled, how quickly he accelerated, and how much separation a receiver created from a defender.
Teams use this information for performance analysis, but also for health and workload. Sports science staff try to understand how much load a player's body has absorbed in practice and games, and to adjust practice plans so players are fresh when it matters. This is an area where analysts are cautious, because the connection between workload and injury is complicated, and teams use it to inform conversations with medical staff, not to replace them.
Medical decisions always rest with doctors and athletic trainers. For how the league treats injuries, see the NFL concussion protocol, and for how the roster handles them, see the active roster, injured reserve and the rules for returning players.
The limits of analytics
Analytics has real limits, and good analysts are the first to say so.
- Small samples. An NFL regular season is short, under twenty games for each team, so many patterns rest on only a handful of observations. A tendency that looks strong over a few games may fade.
- Context. A play's result depends on the matchup, the call, the weather, the injuries and luck. A model that does not include these things will be wrong in situations where they matter.
- Changing rules and styles. The game evolves. Rule changes, such as those affecting kickoffs or contact, alter the underlying numbers. A model built on past data may not fit a new environment.
- The unmeasured. Leadership, effort, communication and the way a player responds to coaching are hard to quantify. A scout who knows a prospect well may see something no spreadsheet captures.
- Human factors. Decisions are made by people who live with the consequences. A coach who follows a model and loses faces more criticism than one who follows tradition and loses. That pressure shapes behavior, and analytics cannot remove it.
Common misconceptions
Several myths follow analytics around.
- "Analytics means going for it on every fourth down." It does not. It means weighing the odds for each situation, and the answer is often to punt or kick.
- "Analytics replaces scouts and coaches." In practice teams use both. The best departments combine data, film and human judgment.
- "If the model was right, the call must have worked." A good decision can fail, and a bad one can succeed. Judging a decision by its result is a common mistake. A 60 percent chance means failing four times in ten.
- "Analytics is only for big teams." The tools are widely available, and even teams with small departments use them. What differs is how well they are integrated into decisions.
How to use analytics as a fan
You do not need a model to benefit from the same ideas. When you watch a game, ask whether a decision looks right given the situation, not just the result. If a team goes for it on fourth down and fails, ask what the alternative would have looked like. Check the standings and team pages for context on how a team is performing across a season rather than in one game, and keep in mind that a single result is a small sample.
If you play fantasy football, similar thinking applies to your own choices: weigh the odds, accept variance, and avoid judging decisions by outcomes alone. Our fantasy section and leaders pages provide the data to practice with.
The short version
- Analytics in the NFL is a set of tools for answering decision questions, such as when to go for it, whom to draft and how to use the cap.
- Win probability and expected points are common frameworks for in-game choices.
- Scouting models combine measurements, college production and position value, but they supplement film and interviews rather than replace them.
- Tracking data supports performance and workload analysis; medical decisions belong to doctors and trainers.
- Small samples, context, rule changes and unmeasured traits limit what any model can say.
- A good decision can lose and a bad one can win; judge process, not just outcomes.