Are Tottenham Actually Getting Better? What the Data and AI Metrics Really Show

The most damning thing about Tottenham's recent Premier League seasons isn't a scoreline. It's a passage of play you've watched dozens of times by now: possession held, the ball worked into the final third, a reset backward, and the move quietly dying before anyone in the box has had a touch. By late 2025, the tracking models had stopped filing that under style and started filing it under pattern.

The Chelsea Night That Broke the Model

Start with November 2025, because it's the cleanest single piece of evidence on the table. Tottenham generated somewhere between 0.05 and 0.10 expected goals across a full Premier League match against Chelsea. That's the lowest single-match xG output in the club's recorded Premier League history, per Opta's tracking, which runs back to the 2003–04 season.

Sit with that number for a second, because low xG and low goals aren't the same problem. A side that creates three good chances and misses all of them posts a bad result with a perfectly respectable xG.

A side that posts 0.05 to 0.10 didn't miss anything. It never got near the point of shooting.

Replay the match and the mechanism is legible. Spurs held the ball, circulated it, reached the edge of the final third, and reset.

Chelsea didn't have to press with any urgency, because pressing wasn't necessary. The passing lanes Tottenham wanted didn't exist. The ones they settled for led sideways.

One match proves almost nothing. xG is a probabilistic measure, and over 90 minutes it's noisy by design. The Chelsea night matters because of what it sits inside.

What the Tracking Data Actually Measures

A quick grounding for anyone who reads football coverage without having sat inside an analytics department.

Expected goals assigns a probability to every shot based on where it was taken from, the angle, the body part used, the defensive pressure applied, and the phase of play it emerged from. Add those probabilities across a season and you get an estimate of how many goals a team should have scored from the chances it actually created. It's a model, not a truth, and every provider weights the inputs differently.

• Opta's published methodology describes expected goals as a probability model built on those inputs, summed across a match or season. *(Paraphrased from Opta's public xG methodology documentation.)*

Field tilt works on a different axis. Instead of measuring shots, it measures the share of final-third touches and passes a team controls relative to its opponent. A side with 60% possession but a 50% field tilt is spending its time with the ball in places that don't threaten anyone.

Expected points takes the shot values from both ends and simulates the match repeatedly to estimate what a typical result would have been. It strips out finishing luck and goalkeeping variance. It does not strip out bad decisions.

That distinction matters. It's the fault line running through this entire debate.

The Trend Line, Not the Outlier

This is where the single-match caveat stops protecting Spurs.

The xG collapse is the headline. Dropping from 69.21 to somewhere between 42.53 and 48.2 isn't a wobble. It's the loss of roughly a third of a season's chance creation, depending on which end of that range you take.

The shot volume is the number that should worry people most. Spurs averaged 9.5 shots per game in 2025–26.

That's the club's lowest offensive output since Opta tracking began in 2003–04, a span that covers some genuinely poor Tottenham teams, including sides that finished mid-table with far less of the ball. It's one thing to be bad. It's another to be bad in a way that has no precedent in the dataset.

There's a piece of this that doesn't fit neatly, and it's worth flagging rather than smoothing over. During Tottenham's best sustained stretches, rolling xG per 90 sat between 1.8 and 2.1. A season total of 42.53 to 48.2 across 38 matches works out to roughly 1.1 to 1.3 per 90. Those two figures describe very different teams.

The gap suggests the peak windows were short and the floor was long, which is a different diagnosis than simple decline. It's one the aggregate numbers alone can't fully resolve.

The underlying metric trajectory across xG, xGA and xPts is where the full picture sits. It's less flattering than any single match.

Possession Without Position

Tottenham averaged over 60% possession in 2025–26. They also averaged their fewest shots in recorded club history.

Read those two sentences together and the tactical problem becomes fairly specific. Possession here isn't a platform. It's a holding pattern, the ball recycled through low-risk areas while the opposition organizes, compresses the space between the lines, and waits for the eventual backward pass.

The field tilt drop from roughly 62% to 50.4% is the measurable version of that. When your possession percentage holds steady while your final-third control falls by double digits, you aren't dominating territory anymore. You're occupying the ball in zones that don't produce anything.

That connects to a structural issue the trade-offs around a high defensive line have exposed. Opponents have worked out that sitting in a compact mid-block and inviting Spurs forward is a low-cost strategy.

It removes the space Tottenham's build-up wants, forces the ball wide, and leaves a high line exposed to direct counterattacks. Press efficiency drops. Chance quality drops with it.

Early 2026–27 data showed expected goals against running at 2.59 per match. That figure comes from a limited sample, and I'd treat it cautiously. Early-season defensive numbers in any model are volatile and often settle. But it points the same direction as everything else.

The Counter-Argument: Cups Are Not Nothing

Plenty of data-led commentary overreaches here, so it's worth pushing back on.

The standard critique holds that cup runs and trophies are being used as evidence of squad development when the domestic metrics say otherwise. Fair enough. But that framing dismisses something real.

Knockout football is a different sport in terms of how games are prepared and managed, and winning tight, low-margin matches is a genuine skill. A cup run is not a rounding error.

There's a second, sharper counter-view buried in the expected points number. Spurs posted 50.1 expected points against 41 actual points in 2025–26, finishing around 13th. That's a gap of roughly nine points, and it's the kind of gap finishing variance and goalkeeping swings can create.

Attribute that divergence to luck and the season starts to look less like a collapse and more like a mid-table side having a bad year in front of goal. The underlying numbers don't say relegation candidate. They say mediocre, underperforming, and going nowhere fast.

There's also a scale point that's easy to lose. Opta's global club power rankings dropped Tottenham out of the European top 15 and into the 30–40 band worldwide through late 2025 and into 2026.

That's a real decline. It's also still a top-40 club on the planet. The distance between "no longer elite" and "broken" is wide, and the two get conflated constantly.

Even the pre-season modeling cuts both ways. Opta's Supercomputer put Tottenham's top-four finish probability at under 25%, landing somewhere between 18% and 25% in that window. That's a forecast, not a verdict, and pre-season probabilities are among the least reliable outputs in sports modeling. The mechanics of the Supercomputer and spatial control models are worth understanding before treating those numbers as destiny.

Key Uncertainties and Open Questions

Several things here can't be settled with the evidence available, and it's worth being explicit about which.

The most obvious is sample size on the defensive side. A 2.59 xGA per match figure from early 2026–27 comes from a restricted set of fixtures. Defensive metrics in particular stabilize slowly, and there's no way to know from this window whether that number represents a genuine structural problem or an early-season artifact that regresses toward the mean.

Opta's model weights aren't fully public, which limits how far anyone outside the company can audit these outputs. Two providers can produce meaningfully different xG values from the same match. That doesn't make the numbers unreliable, but it does mean treating any single figure as definitive is a mistake.

Expected points is a derived construct, not a league table. It's useful for separating process from outcome, and it's also a simulation with its own assumptions baked in.

Then there's the question the data genuinely can't answer: whether targeted recruitment or a shift toward a mid-block can arrest the slide without abandoning the tactical principles the squad was built around. Those two goals may be compatible. They may not be. Nothing in the current dataset settles it.

Whether cup performances should count as evidence of progress is also unresolved, and it's less a data question than a question about what the club is actually trying to be good at.

Where the Trend Could Bend

One interpretation is reasonably well supported by the evidence. This looks like a structural problem, not a finishing problem.

A finishing problem shows up as strong shot volume with poor conversion. A structural problem shows up as historically low shot volume, falling field tilt, and collapsing xG against organized defenses. Tottenham's numbers match the second pattern, not the first.

That distinction has practical consequences. Signing a striker doesn't fix a team that isn't generating shots. Changing the finishing coach doesn't help when the issue is that the ball never arrives in dangerous areas. Get the diagnosis wrong and the fix will be wrong too.

What remains genuinely open is whether the slide is reversible from inside the current tactical framework. The honest caveat deserves repeating: the degree to which recruitment or mid-block adaptation can reverse the pattern without dismantling the core principles is unknown.

That may sound like a dodge. It isn't. Football analytics has a habit of retrofitting certainty onto outcomes that were never certain, and the honest position here is that we have a clear trend, a plausible mechanism, and no reliable way to forecast the inflection point.

FAQ

Are Tottenham actually getting better according to underlying data?

No. Opta-tracked metrics show structural regression from 2023–24 through 2026. Despite cup-run highlights, expected goals fell from 69.21 to below 48.2, shots per game dropped to a club-record low of 9.5, and field tilt slid from roughly 62% to 50.4%.

What was Tottenham's lowest xG performance?

The November 2025 Premier League fixture against Chelsea, where Spurs generated between 0.05 and 0.10 expected goals. That's the lowest single-match xG output in the club's Premier League history since Opta tracking began in 2003–04.

What does field tilt tell us that possession doesn't?

Possession counts time on the ball. Field tilt measures where that time is spent. Tottenham held over 60% possession in 2025–26 while their final-third control fell to 50.4%, meaning the ball was being kept in areas that didn't create chances.

Does winning cup matches prove the team is improving?

Not on its own. Cup football rewards different qualities than a 38-game league season, and knockout variance is high. Cup results and domestic underlying metrics measure different things, and the domestic numbers have moved consistently in one direction.

What do expected points say about Tottenham's 2025–26 season?

Spurs recorded 50.1 expected points against 41 actual points, ranking around 13th on the expected measure. That gap suggests underperformance in front of goal, but even the adjusted figure places them mid-table rather than competitive at the top.

Key Takeaways

• The problem is chance creation, not chance conversion. Signing a finisher won't fix a team that isn't generating shots.
• Possession without field tilt is just a slower way to give opponents time to set. Spurs' ball control and final-third control now tell opposite stories.
• The "unlucky rather than broken" case has real evidence behind it, but its best-case number still leaves Spurs mid-table.
• Early defensive figures are provisional. The broader attacking trend is not.
• Decline and collapse are different claims. Tottenham can be worse and still not be broken.

The question worth sitting with isn't whether Spurs can win another knockout tie. It's whether the club's internal definition of progress, the one that gets used in press conferences and season reviews, can survive contact with what the models have been quietly recording for two and a half years. The data don't say Tottenham are broken. They say the club has been measuring the wrong kind of progress, and the next transfer window and tactical reset will show whether anyone upstairs is reading the same numbers.