Project · Ongoing
The Vikings offense, by the numbers
There's no shortage of opinions about Kevin O'Connell's offense. I wanted to know if I could work out what's actually going on from the data, so I've been running it as a real analysis: one question at a time, each hypothesis written down before looking.
Bottom line so far
- The scheme makes a QB's weaknesses cost more. QBs who already process fast have been fine. QBs still learning to read defenses quickly have struggled.
- O'Connell's "QB whisperer" effect is small. His QBs beat projection by +0.02 EPA/dropback, 12th of 27 coaches. Accuracy goes up, and so do sacks.
- The one consistent O'Connell mark is the sack tax. His regime adds the 2nd-most to sack rate of 42 head coaches. His QBs face more pressure here: about 60% from the line and 40% from holding the ball longer than they do elsewhere.
Vikings offense, league percentile by season
Opponent-adjusted EPA per play · 100 = best offense · * 2026 through week 10
Data table
| Season | Coach | Adj. EPA/play | Rank | Percentile |
|---|---|---|---|---|
| 2015 | Zimmer | +0.034 | 9 / 32 | 74 |
| 2016 | Zimmer | −0.051 | 24 / 32 | 26 |
| 2017 | Zimmer | +0.031 | 9 / 32 | 74 |
| 2018 | Zimmer | −0.075 | 26 / 32 | 19 |
| 2019 | Zimmer | +0.039 | 10 / 32 | 71 |
| 2020 | Zimmer | +0.072 | 11 / 32 | 68 |
| 2021 | Zimmer | +0.014 | 16 / 32 | 52 |
| 2022 | O'Connell | +0.001 | 18 / 32 | 45 |
| 2023 | O'Connell | −0.032 | 19 / 32 | 42 |
| 2024 | O'Connell | +0.040 | 15 / 32 | 55 |
| 2025 | O'Connell | −0.102 | 28 / 32 | 13 |
| 2026* | O'Connell | −0.155 | 31 / 32 | 3 |
Average at best through 2024, then near the bottom of the league in 2025 and 2026. The record hid that: O'Connell's teams won about 8 more games than their point differential predicted, mostly in one-score games, and that doesn't repeat year to year.
Is O'Connell a QB whisperer?
Each QB's EPA per dropback in Minnesota minus his projection
Projection from the QB's prior 3 seasons and age · positive = better than expected
All O'Connell QBs together: +0.022 (95% CI −0.035 to +0.079), 12th of 27 coaches. Darnold kept his gain in Seattle. Cousins matched his Zimmer-era level.
Data table
| Row | Value |
|---|---|
| Wentz 2025 | +0.11 |
| Mullens 2023 | +0.08 |
| Cousins 2023 | +0.08 |
| Darnold 2024 | +0.07 |
| Cousins 2022 | −0.04 |
| Dobbs 2023 | −0.07 |
| Murray 2026* | −0.11 |
Takeaway: a small, uneven lift, middle of the league, and well short of the reputation.
Who drives passing efficiency?
Share of team-to-team differences in passing efficiency
League-wide model, 2015–2026: QB, receivers, head coach, and opponent fit together
Head-coach share is about 0.2% once the QB and receivers are accounted for. The coach effect that does survive shows up in sack rate.
Data table
| Row | Value |
|---|---|
| QB | 30% |
| Receivers | 24% |
| Situation | 5% |
| Opponents | 3% |
| Head coach | 0% |
| Unexplained | 38% |
Deep dives
- Why Vikings Quarterbacks Get Sacked So Much
O'Connell's offenses add more to sack rate than almost any head coach's. His QBs don't turn pressure into sacks more often; they face more of it. About 60% of that is the line, and 40% is QBs holding the ball longer in Minnesota than they do elsewhere.
- Is Kevin O'Connell a QB Whisperer?
Measured against what each QB was expected to do, O'Connell's quarterbacks have played about as expected: more accurate, sacked more often, and in the middle of the league overall. Sam Darnold's 2024 is the best case.
- What Happens When Vikings QBs Go Past the First Read
Murray and McCarthy play at a normal level on first-read throws. Past the first read, especially against the blitz, their results drop sharply. Veterans who knew the system held up.
- J.J. McCarthy's Real Odds of Becoming a Starter
Held to a multi-season standard, McCarthy's odds of becoming a good NFL starter are about 1 in 10. His first Giants games doubled them, from a low base. Here's how first seasons like his usually play out.
How it's built
This is also an experiment in using AI for data science: how much to hand an AI agent, and where a person has to stay in the loop. The setup:
- A question log is the source of truth. Every question gets a hypothesis and "what would change my mind" written down before the data is touched. The finding is logged against it, including when I was wrong.
- I pick the questions and the definitions; the agent writes and runs the analysis (Claude Code, Python/Polars, nflverse play-by-play and charting data).
- Guardrails are written down once: one shared set of play filters, opponent adjustment, intervals that account for plays within a game being related, and a league benchmark on every number.
- Every finding gets robustness checks (neutral game situations, excluding the partial season, alternative thresholds) before it goes in the log.
The clearest lesson so far: the definitions are where I add the most value. When I changed the McCarthy projection to use his actual draft slot and to require a QB to be good for more than one season, his odds of becoming a good starter fell from about 30% to about 10%.
Every question
| # | Question | Short answer |
|---|---|---|
| 1 | How efficient has the offense been, adjusted for opponent and context? | Average at best: 15th–19th in adjusted efficiency in 2022–24, then near the bottom of the league in 2025–26. |
| 2 | Do QBs play better in Minnesota under KOC than they do elsewhere? | QBs beat their projection by a little (+0.02 EPA/dropback, 12th of 27 coaches). Accuracy up, sacks up. |
| 3 | How much of offensive performance is the QB, the play-caller, the supporting cast, and the opponent? | QB and receivers drive passing efficiency; the head coach barely registers. KOC's one clear mark: second-most added sack rate of 42 head coaches. |
| 4 | When the offense struggles, what specifically breaks? | When it struggled in 2022–25, sacks and turnovers broke, not routine plays. 2026 is different: routine plays are failing. |
| 5 | Which "struggle" metrics are stable signal, and which are noise? | How the team operates (timing, pressure, hold) is stable signal. Turnovers and red-zone results are mostly noise. |
| 6 | Does the offense's complexity show up as a measurable operational cost? | Pre-snap penalties run high under KOC. Late snaps are a 2024+ thing, not the whole era. |
| 7 | Why has Justin Jefferson's production dropped in 2025–26? | Jefferson isn't declining. 2025 was the McCarthy connection; 2026 is low volume and an ankle injury. |
| 8 | What separates the Vikings offense from the league's best offenses? | The 2022–24 gap to elite offenses was half sacks and turnovers, double the usual share. |
| 9 | Does KOC's regular-season record overstate the team, and does that explain the playoff results? | KOC's teams won 8 more games than point differential predicted. That luck doesn't carry year to year. |
| 10 | Is the pressure the offensive line, or the ball being held? | Excess pressure is about 60% the line and 40% the QB holding the ball. |
| 11 | Why do Vikings QBs hold the ball? Design or execution? | QBs hold the ball longer in Minnesota than they did elsewhere, and the charted play calls don't explain why. |
| 12 | What can we learn from 2026 so far? | 2026 is a defense-and-close-games team; the offense is 29th. |
| 13 | Does KOC's scheme override his QB's established style, and does that cost anything? | QBs adopt KOC's style about as much as any QB adopts a new coach's. The concept mix didn't cost them. |
| 14 | Is the Vikings' run problem the backs or the blocking? | open |
| 15 | Has the league figured out KOC since the 2024 Rams playoff game? | No copied Rams plan. The decline since follows the QB. |
| 16 | What does Kirk Cousins' return do to the offense? | open |
| 17 | Did Cousins reshape KOC's offense toward his own style in 2022–23? | Cousins didn't reshape KOC's offense; it already fit his style. |
| 18 | Why didn't Kyler Murray work in KOC's offense? | Murray's shortfall is all shotgun: throw value and drops, not accuracy. |
| 19 | Does KOC's passing game put more on the QB, and is that where Murray broke? | Murray is fine on first reads. Past the first read, his throws collapse. |
| 20 | Is KOC's play calling predictable, and has it gotten worse in 2025–26? | Pass/run calls were very readable in 2024–25, with no measurable cost. |
| 21 | Was the offense simplified for McCarthy in 2025, and does simplifying work under KOC? | McCarthy's 2025 offense wasn't really simplified. Wentz's was. |
| 22 | Did the 2025 formation change help McCarthy through progressions, and was moving on from him defensible by base rates? | Few QBs with a rookie year like McCarthy's become average starters. Moving on was defensible, not clear-cut. |
| 23 | Do simple passing offenses have a ceiling? | Simple offenses don't have a ceiling. Minnesota is complex and slow almost every year. |
| 24 | Where is McCarthy likely to land, by QB tier? | McCarthy's odds of becoming a sustained good starter: about 1 in 10. |
| 25 | Do the Vikings lack a "pressure plan" against the blitz? | Since 2025 the blitz costs Minnesota far more than the league, and the cost sits past the first read. |
| 26 | Is McCarthy's Giants move the same "play" that turned Goff and Lawrence around? | A new coach doesn't make a struggling young QB better. Goff's and Lawrence's turnarounds looked nothing alike. |
Data: nflverse play-by-play, Next Gen Stats, participation and Pro Football Reference data (CC-BY 4.0), and FTN charting via nflverse (CC-BY-SA 4.0). Not affiliated with the NFL or the Minnesota Vikings. Fan project.