| GM Composite Rank | GM Name | GM Team | GM Tenure | Homegrown Picks Percentage | Winning Percentage | Winning Efficiency | GM Homegrown Picks Rating | GM Efficiency Rating | GM Composite Rating |
|---|---|---|---|---|---|---|---|---|---|
| 1 | Eric DeCosta | Ravens | 6 | 61.6 | 68.0 | 11.9 | 90.0 | 80.0 | 85.0 |
| 2 | Brian Gutekunst | Packers | 7 | 62.2 | 63.4 | 6.8 | 91.5 | 64.7 | 78.1 |
| 3 | Brett Veach | Chiefs | 8 | 52.3 | 75.8 | 25.2 | 49.5 | 97.7 | 73.6 |
| 4 | Kwesi Adofo-Mensah | Vikings | 3 | 53.8 | 66.7 | 15.2 | 58.2 | 87.2 | 72.7 |
| 5 | John Schneider | Seahawks | 11 | 55.9 | 60.8 | 8.1 | 69.3 | 68.8 | 69.1 |
| 6 | Jerry Jones | Cowboys | 11 | 62.8 | 58.3 | 1.5 | 92.5 | 45.2 | 68.9 |
| 7 | Howie Roseman | Eagles | 10 | 52.9 | 61.3 | 10.3 | 53.1 | 75.7 | 64.4 |
| 8 | Brad Holmes | Lions | 4 | 51.8 | 58.1 | 7.8 | 47.0 | 67.8 | 57.4 |
| 9 | Mike Brown | Bengals | 11 | 64.2 | 49.4 | -8.2 | 94.8 | 15.5 | 55.1 |
| 10 | Brandon Beane | Bills | 8 | 45.5 | 65.6 | 19.1 | 16.6 | 93.1 | 54.8 |
| 11 | Les Snead | Rams | 11 | 54.0 | 53.9 | 2.3 | 59.3 | 48.0 | 53.6 |
| 12 | John Lynch | 49ers | 8 | 52.8 | 53.0 | 2.1 | 52.3 | 47.6 | 50.0 |
| 13 | Mickey Loomis | Saints | 11 | 46.7 | 55.6 | 8.3 | 21.2 | 69.5 | 45.3 |
| 14 | Chris Ballard | Colts | 8 | 55.5 | 47.3 | -5.2 | 67.1 | 23.1 | 45.1 |
| 15 | Jason Licht | Buccaneers | 11 | 54.8 | 47.2 | -4.9 | 63.6 | 23.8 | 43.7 |
| 16 | Chris Grier | Dolphins | 9 | 43.0 | 50.7 | 5.6 | 9.6 | 60.3 | 35.0 |
| 17 | Terry Fontenot | Falcons | 4 | 52.7 | 42.6 | -8.2 | 52.2 | 15.5 | 33.8 |
| 18 | Andrew Berry | Browns | 5 | 46.3 | 47.6 | 0.6 | 19.5 | 41.9 | 30.7 |
| 19 | Joe Schoen | Giants | 3 | 52.6 | 36.3 | -14.5 | 51.2 | 5.9 | 28.5 |
| 20 | George Paton | Broncos | 4 | 47.3 | 44.1 | -3.5 | 23.8 | 27.8 | 25.8 |
| 21 | Nick Caserio | Texans | 4 | 34.9 | 40.4 | 0.2 | 1.0 | 40.3 | 20.7 |
| 22 | Ryan Poles | Bears | 3 | 48.4 | 29.4 | -18.9 | 28.8 | 2.7 | 15.7 |
[Methodology] Use Homegrown Picks to Rank NFL General Managers
Background
A useful application of going hardcore wonk and developing Homegrown Picks are the metrics. We can use these to rank current NFL general managers (GM).
We know several GMs are pin cushions for the decision maker. For our purposes here, though, GM-by-name will wear these metrics. No shame, professional pin cushions are well compensated. Probably the most successful professional pin cushion in the world is Roger Goodell and that guy makes like $60 million a year.
We’ll use two metrics developed in the Homegrown Picks results. First, we have shown the percentage of snaps taken by homegrown draft picks has a direct relationship to more regular season wins, more postseason success, and longer tenures of GMs and head coaches. We assign a single career percentage to each GM.
Second, we formulated a linear model with winning percentage as the response variable to the contribution of homegrown picks. While the linear relationship is statistically significant the predictive ability is low because of outliers. These outliers take the form of 1) teams that have low contribution from homegrown picks but win a lot of games or 2) teams with high contribution from homegrown picks but are cellar dwellers. We describe these respective examples as highly efficient or inefficient, respectively. As above, we assign a single career efficiency percentage to each GM.
Methods
We use the data from Homegrown Picks. It spans the 2014-2024 seasons. Additional specifics can be found in the methods section. Recall in this dataset each row is a single team’s season.
There are some eligibility conditions. First, the GM must be currently employed by the NFL. So we exclude the newly dismissed GM cohort. Second, there is a question of the current roster reflecting the work of the current GM. We require at least three years tenure and so we exlude the nascent GM cohort. Third, the data window covers 11 seasons. We unavoidably exclude some track record from the senior GM cohort.
Ranking Procedure
- Get the history of all \(n\) eligible GM. Each row is a season and has the contribution of homegrown picks and winning percentage. Compute a game-weighted average for these two metrics. This dataset has \(n\) rows.
- For the homegrown picks column, express this column as a \(z\)-score. Don’t be scared the well-known \(z\)-score formula that is below don’t look away:
\[\mathbf{z}_{column} = \frac{x_i - \mu}{s} \tag{1}\]
where \(x_i\) is the row observation, \(\mu\) is the column mean, and \(s\) is the column sample standard deviation.
- Compute the efficiency rating from homegrown picks and winning percentages.
- Train an ordinary least squares (OLS) linear model from the 2014-2024 dataset where the dependent variable is winning percentage and the independent variable is the percentage of snaps taken by homegrown picks.
- Feed the trained model the percentage of homegrown picks values computed in step 1 above. These are the predicted winning percentages.
- Subtract the predicted winning percentage from the actual winning percentage. The conjecture here is this tells us if the contribution from homegrown picks are actually providing value (there are of course other factors at hand.)
\[ \text{Efficiency Rating}_i = wp_i - (\beta_0 + \beta_1x_i) \tag{2}\]
where \(wp_i\) is a GM’s historical winning percentage from step 1, \(\beta_0\) is the intercept from the trained OLS model, \(\beta_1\)is the coefficient from the trained model, and \(x_i\) is the GM’s historical contribution from homegrown picks from step 1.
- As above, express the efficiency rating column as a \(z\)-score, after which we now have two columns of \(z\)-scores.
- We want to compute a single composite rating. We also want to express our rating in graspable units, like between 0 and 1 thus 0% to 100%. And so now express the \(z\)-score columns as cumulative percentage. We do this with a Student’s t-distribution (the difference is quite pedantic).
- To get the composite rating, average the two cumulative percent columns. Thus, our composite score considers two factors—the percentage of snaps taken by homegrown picks and winning efficiency—and weights them evenly. And so the ratings columns in the results table are labeled:
- GM Homegrown Picks Rating
- GM Efficiency Rating
- GM Composite Rating
- Rankings are the composite score sorted in descending order.
2025 Example
Below we rank current GMs between the 2024 and 2025 NFL seasons. Using the eligibility conditions above there are 22 GMs:
- Excluding the newly dismissed GM cohort excludes Trent Baalke of the Jaguars, Ron Carthon of the Titans, Joe Douglas of the Jets, and Tom Telesco of the Raiders. (Interesting that Baalke and Telesco were the GM of two different teams in our data window.)
- The minimum tenure qualification means a front office would’ve drafted in April 2022, 2023, and 2024. Excluding the nascent GM cohort excludes Joe Hortiz from the Chargers, Dan Morgan from the Panthers, Monti Ossenfort from the Cardinals, Adam Peters from the Commanders, and Eliot Wolf from the Patriots. Omar Kahn from the Steelers is also exluded because he was not GM until after the 2022 draft.
- Some portions of track record are excluded from the senior GM cohort. The data allows for a maximum 11-year track record. Mike Brown has been the owner and GM of the Bengals since 1991. Jerry Jones has been the owner and GM of the Cowboys since 1989 (but he didn’t fire Jimmy Johnson until 1994). Additionally, Mickley Loomis has been the GM of the Saints since 2002, John Schnieder has been the GM of the Seahawks since 2010 (but Pete Carrol wasn’t fired until 2024), and Les Snead has been the GM of the Rams since 2012.