Find players you didn’t know you needed.
Match turns a one-line brief into a ranked shortlist of targets — explained, sourced, and re-ranked nightly. So your scouts spend their hours judging players instead of compiling them.
In one sentence
You write the brief in your own words.
Match returns the names you should already know.
Match is a ranking engine, not a database query. It interprets your brief — “left-footed CB, ≤ €15M, ≤ 23, agile in possession, top-5 leagues only” — and returns the players who fit, ordered by a defensible match score.
Every result carries an explanation: the features driving the score, the stylistic siblings the player resembles, the transfer probability layered on top, and the contract clauses worth knowing about.
Six features that close the loop.
Each one earned its place by removing a step a sporting director shouldn’t still be doing in 2026.
01
Plain-language briefs
Type the player you want in your own words. Match parses the brief into structured filters — and asks you to confirm before it runs.
02
Explainable match scores
Every player carries a 0–100 score with feature attributions. You see why a player ranks 91 and the player below ranks 88.
03
Transfer probability layer
A second model estimates the chance a player moves in the next 6 months. Stop scouting players who won’t leave.
04
Stylistic peers
Match clusters players by movement, decision-making, and on-ball style — not just position. Useful when “a left-back” isn’t the same shape twice.
05
Contract & clause radar
Release clauses, expiring contracts, loan-back options — surfaced inline so you don’t learn about them from a journalist.
06
Saved briefs
Save a brief once. Match re-runs it every night and emails the diff — new entrants, score changes, contract movements.
No black boxes. No unfounded confidence.
Architecture
- 1
Brief parser (LLaMA 3, fine-tuned)
Maps natural language into a structured query — leagues, age, market value, foot, traits — with a confirmation step before any inference runs.
- 2
Feature pipeline (StatsBomb · FBref · open data)
200+ player features computed nightly. Possession quality, progressive carries, defensive zone activity, set-piece involvement, role-adjusted xG/xA.
- 3
Ensemble ranker (XGBoost · CatBoost · Random Forest)
Three independently-trained models. Their disagreement is itself a signal — large variance flags low-confidence rankings.
- 4
Transfer-probability head
A separate model trained on contract length, recent rumours, club financial signals, agent history, and comparable transfers.
- 5
Explainability layer (SHAP)
Per-player feature attributions. The result: every score arrives with the three features that pushed it up and the three that pulled it down.
- Train on your private club data without explicit, written consent.
- Generate match scores without showing the features behind them.
- Hide the leagues a brief covered or didn’t cover.
- Pretend a 60-confidence ranking is a 95.
If a brief returns weak signal, Match says so — and recommends widening the filter or running Vision over candidates manually.
How a sporting director uses Match in a window.
Send us your hardest brief. We’ll send the shortlist back.
Email your one-line target description and the leagues you want covered. We’ll return Match’s live top-10 with explanations, the same day, signed by a founder.