The Blunt Instrument
Our prediction model applies a -4.0 PPI adjustment to every external coaching hire. Every one. Alabama replacing Nick Saban with Kalen DeBoer? Minus four. Indiana replacing Tom Allen with Curt Cignetti? Minus four. North Carolina replacing Mack Brown with Bill Belichick? Minus four.
Internal promotions get -1.5. The logic: a new external coach means new scheme, new terminology, new player roles, and year-one efficiency declines are structural regardless of the coach's quality. A promotion means continuity, so the penalty is smaller.
This is our single biggest modeling limitation, and the 2024-25 coaching carousel proved it. Here is every coaching change we tracked over the last two cycles, the adjustment we applied, and what actually happened.
The 2024 Coaching Changes: What We Got Wrong
| Team | Old Coach | New Coach | Type | PPI Adj | Actual Result |
|---|---|---|---|---|---|
| Alabama | Nick Saban | Kalen DeBoer | External | -4.0 | 9-4 (missed CFP) |
| Indiana | Tom Allen | Curt Cignetti | External | -4.0 | 11-2 (CFP) |
| Washington | Kalen DeBoer | Jedd Fisch | External | -4.0 | 6-7 |
| Michigan | Jim Harbaugh | Sherrone Moore | Internal | -1.5 | 8-5 |
| Texas A&M | Jimbo Fisher | Mike Elko | External | -4.0 | 8-5 |
| Duke | Mike Elko | Manny Diaz | External | -4.0 | 9-4 |
| Michigan State | Mel Tucker | Jonathan Smith | External | -4.0 | 5-7 |
| Arizona | Jedd Fisch | Brent Brennan | External | -4.0 | 4-8 |
| UCLA | Chip Kelly | DeShaun Foster | Internal | -1.5 | 4-8 |
| Mississippi State | Zach Arnett | Jeff Lebby | External | -4.0 | 2-10 |
| Boston College | Jeff Hafley | Bill O'Brien | External | -4.0 | 7-6 |
| Syracuse | Dino Babers | Fran Brown | External | -4.0 | 10-3 |
Two results break the model's assumption that all external hires are equal.
Indiana under Cignetti went 11-2 and reached the College Football Playoff in year one, then went 16-0 and won the national championship in year two. The -4.0 adjustment assumed Cignetti would face the typical year-one efficiency decline. Instead, he brought a fully formed system from James Madison, used the transfer portal to import players who fit that system immediately, and produced the best single-season improvement of any coaching change in the dataset. The penalty was not just wrong, it was directionally wrong. Cignetti's arrival made Indiana better than the model projected even without the adjustment.
Syracuse under Fran Brown went 10-3 after replacing Dino Babers (who went 37-50 over seven seasons). A -4.0 adjustment to a program that had been trending downward for years penalized the wrong side of the change. When the outgoing coach was already underperforming the program's talent level, a new hire is more likely to produce an upgrade than a decline. The model treats all departures as a loss of competitive advantage. For bottom-tier programs, the departure is the competitive advantage.
Alabama under DeBoer is the opposite failure. The -4.0 adjustment was insufficient. Saban's retirement wasn't a normal coaching change, it was the removal of a generational competitive advantage. Alabama went 9-4 and missed the CFP for the first time since 2019. The model's -4.0 penalty captured some of the transition cost, but it couldn't capture the NIL-driven talent exodus (James Smith and Qua Russaw leaving for Ohio State in the next cycle), the recruiting identity crisis, or the fundamental difference between replacing a capable coach and replacing the most successful coach in college football history.
The 2025 Coaching Changes: Even Stranger
| Team | Old Coach | New Coach | Type | PPI Adj | Actual Result |
|---|---|---|---|---|---|
| North Carolina | Mack Brown | Bill Belichick | External | -4.0 | 4-8 |
| Florida State | Mike Norvell | DJ Durkin | External | -4.0 | 3-9 |
| Stanford | Troy Taylor | Frank Reich | Interim | -2.5 | 2-10 |
Belichick at North Carolina is the most fascinating data point. A six-time Super Bowl champion who had never coached a college game, never recruited a teenager, and never navigated the transfer portal received the same -4.0 adjustment as every other external hire. The model had no mechanism to evaluate whether NFL coaching excellence translates to college, or whether the translation loss is even larger than -4.0. North Carolina went 4-8, suggesting the adjustment was too generous.
The NFL-to-college transition is structurally different from the college-to-college transition in ways the model doesn't capture. In the NFL, players are contractually obligated to learn the playbook. In college, they can leave for the portal. In the NFL, rosters are built through a draft with salary cap constraints. In college, rosters are built through recruiting and NIL. Belichick's tactical genius is real, but tactical genius without the ability to recruit, retain, and motivate 18-year-olds is a different skill set than the one that won in Foxborough.
What the Data Actually Shows
Across both cycles, a pattern emerges that the blanket -4.0 adjustment obscures:
When a new coach replaces an underperformer, year-one results tend to exceed projections. Cignetti replacing Allen (Indiana), Fran Brown replacing Babers (Syracuse), and Manny Diaz replacing Elko (Duke, who left voluntarily) all outperformed the model. The outgoing coach was already leaving value on the table. A competent replacement captures that latent value immediately, especially if they use the portal aggressively to import players who fit the new system.
When a new coach replaces an elite performer, year-one results tend to fall short of projections. DeBoer replacing Saban, Fisch replacing DeBoer (who had just been to the CFP at Washington), these transitions consistently underperform because the penalty underestimates how much institutional knowledge walks out the door with an elite coach.
NFL-to-college transitions carry a larger adjustment than college-to-college. Belichick's 4-8 and Reich's 2-10 aren't definitive (small sample), but the structural differences between the two levels suggest the translation loss is larger than -4.0 for NFL coaches.
Internal promotions are more variable than the -1.5 suggests. Sherrone Moore's 8-5 at Michigan was solid but included a significant talent drop-off from Harbaugh's last roster. DeShaun Foster's 4-8 at UCLA was a disaster. The -1.5 adjustment assumes continuity, but an internal promotion after a head coach departure often coincides with roster attrition through the portal, reducing the continuity advantage.
What a Better Model Would Look Like
The blanket -4.0 was always a compromise. We knew it was too blunt. We used it because coaching changes are infrequent enough (15-20 per year across 136 FBS teams) that training a model on coaching-change-specific features produces unstable estimates. Small sample sizes and heterogeneous situations make it difficult to build a reliable regression.
But the 2024-25 data gives us enough signal to propose a more nuanced framework. Instead of one adjustment, we're testing a system that considers three variables:
1. The gap between outgoing coach performance and program talent level. If the outgoing coach was significantly underperforming the program's recruiting and talent rankings (like Tom Allen at Indiana, where the talent was better than the results), the adjustment should be smaller, or even positive. If the outgoing coach was overperforming (like Saban at Alabama, where the results consistently exceeded even the elite talent level), the adjustment should be larger.
2. The new coach's track record at similar programs. Cignetti had a 52-9 record at James Madison and ran a system that translated directly to Indiana's roster needs. Belichick had zero college coaching experience. These are not equivalent transitions, and the model shouldn't treat them equivalently.
3. Portal activity in the transition. A coaching change now triggers immediate portal movement. The number of departures, the quality of departures, and the incoming portal class all provide real-time signal about how the transition is going, signal that's available before the season starts and should be incorporated into preseason projections.
We don't have enough data to fully validate this framework yet. Two cycles of coaching changes (roughly 30 data points) isn't enough to build stable coefficients. But the directional evidence is clear: a graduated adjustment that accounts for the outgoing coach's performance gap, the new coach's relevant experience, and the portal activity during the transition would have produced better projections for Indiana (less penalty), Alabama (more penalty), and North Carolina (different penalty category entirely) in both 2024 and 2025.
The 2026 Coaching Changes to Watch
The framework above suggests three 2026 transitions worth tracking:
Penn State under Matt Campbell is the most interesting test case. Campbell was an elite coach at Iowa State (multiple top-25 finishes, consistent overperformance relative to talent), he's moving to a program with significantly more resources, and he imported ~24 of his own players through the portal. Every indicator in our proposed framework points to a smaller-than-normal transition penalty. If Penn State outperforms preseason projections, it validates the "coach upgrading to better resources with his own players" category.
LSU under Lane Kiffin is the opposite case. Kiffin is an elite offensive coach (Ole Miss finished top-3 in SP+ offense in 2025), but he's installing a radically different system on a roster built for Brian Kelly's scheme. The portal class is strong (#1 in 247Sports rankings), but scheme translation takes time. The framework suggests a standard -4.0 is appropriate here, Kiffin is elite, but the scheme mismatch creates a genuine year-one efficiency cost.
Oklahoma State under Eric Morris is a controlled demolition. After going 1-11 and losing 64 players to the portal, Morris brought in 50+ transfers for a full roster overhaul. The blanket -4.0 almost certainly understates the transition cost here, because the adjustment is designed for a program replacing a coach, not a program replacing its entire roster. Oklahoma State is less a coaching change than a program reboot, and the model has no category for that.
The Honest Conclusion
The -4.0 adjustment was never good enough. We knew that when we built it, and the data from the last two cycles confirms it. The structural reality is that coaching changes in college football are too heterogeneous for a single number to capture, a Division II coach arriving with a portal-built roster is categorically different from an NFL legend who has never recruited a teenager.
We are building a better system. It won't be perfect,15-20 coaching changes per year is still a small-sample problem, and every transition has idiosyncratic factors no model can anticipate. But a graduated adjustment that accounts for the performance gap, the coach's relevant track record, and the portal activity around the transition will be directionally better than what we had.
The programs that prove the model wrong are usually the ones teaching us the most. Indiana's 2024-25 run didn't just win a championship, it showed us exactly where our coaching-change assumptions break down. That's the kind of failure we can learn from.
For our full prediction methodology and season-long track record, see the track record page. For the AI model autopsy that first identified the coaching-change problem, see Why Predicting College Football Is So Hard: A 748-Pick Season Autopsy.