FPL Expected Points Model 2026/27: How I Predict Every Player

TL;DR

I rebuilt my expected points model from the ground up for 2026/27. Every player's projection is now built from six scoring components (goals, assists, clean sheets, saves, bonus, minutes), each predicted separately and summed, so you can see exactly where every point comes from. In testing against last season, the new model beat the old one on 18 of 20 evaluation sets. See the live xPts table or run it on your own squad.

What Are FPL Expected Points?

Expected points (xPts) forecast how many FPL points a player will score in upcoming gameweeks. Projected points, predicted points, xPts: different names, same idea.

Form tells you what already happened. Expected points tell you what happens next, because they are built from the things that cause points: chances created, minutes played, opponent strength, and fixtures. That is the whole edge. If you want the basics first, start with Expected Points Explained.

What's New in the 2026/27 Model

The old model worked. The new one is better where it matters most: the top players you actually captain and build around.

• Component predictions. Goals, assists, clean sheets, saves, bonus, and minutes are each modelled separately, then converted to FPL points and summed. No black box: open any player in the xPts table and the breakdown adds up to their total, point by point.
• Live team strength. Every club's attacking and defensive strength updates match by match and projects forward, so a team that sharpens in March is not still rated on August form.
• Smarter minutes. A player who plays 60 minutes scores differently from one who plays 90. The minutes model blends confirmed team news, press reports, and post-match data, and it updates through the week.
• Honest testing. Every change has to win on frozen historical data with no future information leaking in. If a tweak does not improve real forward-looking accuracy, it does not ship.

I tested the new model against the old one across two full seasons of history. It won 18 of 20 evaluation sets. The other two were ties I am still annoyed about.

How the Model Predicts Each Part of a Player's Score

Every projection starts with the question: how many points does each scoring rule pay this player, this fixture?

• Goals: shot volume and chance quality (xG), scaled by the opponent's defensive strength and home or away.
• Assists: chance creation (xA), set-piece and penalty duties included.
• Clean sheets: the full defence's probability of a shutout given the opponent's attack.
• Saves and bonus: modelled from workload and match patterns, because keepers and ball-winning defenders quietly bank these every week.
• Minutes: the multiplier on everything above. A benched premium scores nothing.

Double gameweeks double the exposure automatically, and blanks zero it. That is why chip timing falls straight out of the same numbers on the chip strategies page.

How Accurate Is It?

Accuracy claims in FPL are cheap, so here is how I actually measure it. Every candidate model replays historical seasons with its knowledge frozen at each gameweek, no peeking at the future, and gets scored on the decisions that win you rank: did it find the haulers, did it rank the right captains, was it right about the top players and not just the average ones.

The 2026/27 rebuild beat the previous model on 18 of 20 of those evaluation sets. I also benchmarked it against bookmaker player odds, and on average across players it came out ahead of the betting market's implied expectations. The market prices in margin. I do not.

How to Use Expected Points

Three decisions, one number:

• Captaincy: sort next gameweek's xPts, pick from the top. Vice-captain the second name.
• Transfers: sort by cumulative xPts over your planning horizon, not last week's haul.
• Chips and planning: let the solver do it. It reads the same projections and finds the best transfer and chip path for your exact squad, up to 8 gameweeks ahead.