Harvey Elliott scored in the 87th minute at the Parc des Princes on Tuesday night, Liverpool 1-0 PSG on aggregate, and the Champions League Round of 16 produced its first major tactical inversion of the 2024-25 knockout phase. The expected goals for the tie sat at 4.1 for PSG and 1.8 for Liverpool across the two legs. The team that lost the underlying numbers by more than 2 xG won the tie by a single late goal. This is exactly the kind of result that public xG models are bad at celebrating because the models do not measure what actually happened, only what the chances suggested would happen. Liverpool advanced. PSG went home. The xG numbers do not care.
The tactical story underneath the result is the one that travels. Arne Slot’s Liverpool entered the tie as the better-rested side and the better-formed side but the structurally weaker matchup against a PSG team that had spent the league phase humiliating opponents. Liverpool’s adjustment — sitting deeper than they normally would in the first leg at Anfield, then absorbing PSG’s possession in the second leg and playing for one moment — was the tactical inversion of how Klopp’s Liverpool would have played the same matchup. Slot has been rebuilding the press into something more selective. The PSG tie was the first knockout test of whether the rebuild has actually produced a team that can win the games the underlying numbers say they should lose.
What follows is what Slot’s tactical adjustment actually looked like in the numbers, where Luis Enrique’s PSG side ran into the structural limit of its own brilliance, and what the result reveals about how the league-phase format produces matchups the knockout round cannot validate.
The tactical adjustment Slot actually made
Liverpool under Jürgen Klopp pressed at a PPDA of approximately 8.5 across the prime Champions League years. Under Slot in 2024-25, the regular-season PPDA had compressed to 11.2 — meaningfully less aggressive, with the recovery work shifted from high turnovers to organized mid-block. The PSG tie compressed it further: 13.4 PPDA across the two legs. Liverpool were not pressing. They were defending in a structured 4-2-3-1 that conceded possession in the middle third and disrupted the final third.
The structural reason this worked against PSG specifically is that Luis Enrique’s side had been beating teams by combining high possession with quick vertical attacks once the press broke down. A team that does not press is not vulnerable to vertical attacks in the same way. PSG’s offensive structure relies on opponent disorganization in transition. Liverpool refused to disorganize. They conceded the possession battle by enormous margins — PSG had 64% possession across the two legs — and they accepted that the xG-against would be high. The bet was that PSG’s xG-for would convert at less than the average rate, and the bet landed.
Across the two legs, PSG produced 18 shots with combined xG of 4.1. They scored zero goals. Liverpool produced 8 shots with combined xG of 1.8. They scored one goal. The finishing variance ran entirely in Liverpool’s favor and entirely against PSG. That is not a tactical victory in the conventional sense. It is a tactical setup that accepted the structural xG disadvantage in exchange for a higher conversion rate on the chances it did produce. Slot read the matchup correctly. The bet was high-variance and it paid out.
Where Luis Enrique’s PSG ran into the structural limit
| Metric (two legs combined) | PSG | Liverpool |
|---|---|---|
| xG | 4.1 | 1.8 |
| Shots | 18 | 8 |
| Goals | 0 | 1 |
| Possession | 64% | 36% |
| Big chances missed | 5 | 1 |
| PPDA | 9.2 | 13.4 |
The five big chances missed by PSG is the part the tactical community has been writing about all week. Donnarumma made one save that exceeded the xG threshold; the other four big-chance failures were finishing variance from forwards who had been converting at above-average rates during the league phase. The bet against PSG was a bet against finishing regression, and finishing regression arrived in the most-leveraged single match of their tournament.
Our expected goals primer covers why this kind of variance happens. A team that produces 4.1 xG against a quality goalkeeper should score, on average, between 3 and 4 goals across 90 minutes of football. Scoring zero on 4.1 xG is a roughly 3% outcome. It happened. The tournament knockout format produces these outcomes specifically because the sample size of two legs is too short to wash out the finishing variance.
Where this gets weird
The clean “Liverpool tactical masterclass” reading misses three things that complicate the celebration.
The first is that the same Liverpool tactical setup against the same PSG side, run a hundred times in simulation, would produce a Liverpool advance in roughly 12% of the runs. The xG margin is that large. Slot’s setup was tactically correct in expectation but the result that actually happened was a low-probability outcome from the model’s perspective. Smart tactical decisions still lose to better teams most of the time. Liverpool ran the smart play and got the lucky outcome. Both things are true.
The second is that PSG’s bracket got harder, not easier, from this point. The tie they avoided losing — a quarterfinal against Aston Villa, then a semifinal that would likely have been Real Madrid or Bayern — would have been winnable based on the underlying numbers. The bracket geometry that comes from the league-phase format actually favored PSG more than the public conversation acknowledged. Liverpool’s win sent PSG home but also opened a quarter of the bracket that has since produced an Arsenal that’s playing the highest-quality football of the competition.
The third is that the league-phase format itself made this matchup possible. Under the old group-stage format, Liverpool and PSG might have been seeded apart in the bracket because the group draw would have created different seeding lines. The league-phase format pairs teams on a different geometry, and the early-knockout matchups have produced more high-variance ties than the old format would have. The structural reason is that the league-phase rewards teams that performed well across all eight matches, but the knockout phase reset all those advantages and started over. PSG’s league-phase dominance translated into nothing.
What the rest of the bracket actually tells us
- Arsenal looks like the model favorite now. Their league-phase xG profile was already the best in the competition; the PSG departure opens a path that the Gunners can navigate against Real Madrid and a likely semifinal opponent without facing the team that had been dominating the underlying numbers.
- Inter advanced past Bayern. The Italian side beat the Germans 4-3 on aggregate in a tie that the underlying numbers actually had slightly in Bayern’s favor. Two ties in this round have now produced model-incorrect results. The knockout variance is on full display.
- Aston Villa’s run continues. The Premier League side beat Club Brugge 3-1 on aggregate and now plays whichever side comes out of the Liverpool-PSG side of the bracket. Villa’s xG profile is the weakest of the remaining sides but their finishing has been above expectation all tournament.
- Barcelona vs Dortmund is the most evenly matched tie still alive. Both teams’ xG profiles sit within 0.1 of each other across the league phase. This is the tie that will likely be decided by a single moment, in either direction.
The callback
That 87th-minute Harvey Elliott goal at the Parc des Princes, the one that sent PSG home with the better underlying numbers and Liverpool through with the higher finishing variance, is the cleanest example the 2024-25 Champions League has produced of why knockout football is structurally different from league-phase football. The teams that win the league phase produce the best underlying numbers across eight matches. The teams that win the knockout produce the best finishing in two-leg samples. Both skills are valuable. They are not the same skill. PSG was the best team across eight league matches. They are not the best team in the knockout, where the sample is too short for their structural advantage to assert itself. Liverpool ran the smart tactical setup and got the lucky outcome. The lucky outcome is the only one the bracket cares about. The possession trap piece covers an adjacent version of the same problem. Slot’s adjustment was the tactical story. The finishing variance was the actual story. Both are true. The bracket continues, and the model favorite (now Arsenal) starts to look more inevitable with PSG out of the way.
Match data via UEFA; xG context via FBref; tactical breakdown via public broadcast tracking.
