Who makes this, and how it works
The short version
FPL Analytics answers the questions an FPL manager asks before every deadline — who to captain, who to bring in, whether a hit is worth it, which differential is worth the risk. It started as the thing I wanted in front of me before a deadline and turned into a statistical model I keep building in public. Free, no account, no paywall, no ads. Part-time data analyst. Full-time FPL manager.
How the projections work
FPL’s scoring rules are published and exact — a goal from a midfielder is five points, always. So there is nothing to learn about how points are awarded. All the uncertainty is in how many things happen: goals, assists, clean sheets, defensive contributions, minutes, bonus.
The model estimates each club’s attacking and defensive strength, turns that into the expected goals for a specific fixture, splits those goals among the players likely to be on the pitch, and then simulates the gameweek thousands of times. What comes out is not one number but a whole distribution — which is why this site can tell you a player’s chance of a big score rather than only his average one. An average cannot be captained.
It is tested the way it would actually be used: train on the gameweeks already played, predict the next one, never look forward. Splitting a season randomly would let a later gameweek teach the model that predicts an earlier one, which flatters every score it produces.
- Rank correlation
- 0.36Against 0.27 for points-per-game and 0.25 for last-five form — all three on the same players, in the same gameweeks.
- Gameweeks it beat points-per-game
- 31/31Every gameweek tested, not the average across them. Against last-five form it wins 29 of 31.
- Calibration error
- 1.6%When it says 40%, it happens about 40% of the time — off by this much.
- Points of headroom left
- ≈0Per player per gameweek. It has reached its own noise floor — what is left is football being random.
Measured on the 2025/26 season, across all 31 gameweeks it was tested on. Last re-measured 2026-08-13.
What it cannot do
Most of the error left in these projections is noise, not ignorance. Measured against the model’s own distributions, it has already reached the floor: the error a perfectforecaster would still make on the same gameweeks is no lower than the error this one makes. A deflection, a substitution on 60 minutes, a penalty given or not given: none of that is knowable in advance by anyone. A model that claimed otherwise would be fitting last week’s luck and calling it insight.
That is also why 2 of the questions a manager asks every week still have no model behind them here. Rather than answer them badly, they are named:
- Who is actually going to play, and who do I bench? One decision, not two: the eleven you start. Minutes risk is the missing input, and until it exists this says so.
- When do I play each chip? A squad-level question this model does not yet answer. The Squad Planner does, for a specific gameweek.
Every one of those is on the build list, and each will appear here the week it can be answered honestly — not before. The blog records what was believed, what was measured, and what was done about it, including the attempts that failed.
Where the numbers come from
Player, fixture and points data comes from the official Fantasy Premier League API — the same service the game itself runs on, operated by Pulselive on behalf of the Premier League. Expected goals and assists originate with Opta and reach us through that same feed. This data is refreshed daily through the season, so prices, ownership and injury flags stay current rather than being a weekly snapshot.
Everything built on top of it — the projection model, the data visualisations, the Squad Planner — is FPL Analytics’s own.
Source: the Premier League — www.premierleague.com. This site is not affiliated with, endorsed by, or connected to the Premier League or Pulselive.
How the Premier League’s data is used, and the lines that are held
FPL Analytics takes the Premier League’s terms of use seriously. Those terms restrict commercial use of Premier League content and the building of databases from it. FPL Analytics’s position is deliberately conservative:
- Free and non-commercial. There is no subscription, no paywall and no advertising on this site. It is not run to make money.
- No bulk redistribution. We do not publish the data as a feed, a download or an API for anyone else to consume. It is shown to managers inside the tools, alongside our own analysis.
- Attribution everywhere it belongs. The Premier League is credited as the source of the underlying data, here and in the FAQ.
- We stop if asked. If the Premier League or Pulselive want anything here changed or taken down, get in touch and we will do it — no argument, no delay.
Nothing here is an attempt to take advantage of a free service or to trade on someone else’s rights. It is a tool for people who play the game, made by someone who plays it.