← → navigate · P to print
gaffer■

The behavioral
intelligence layer
for football

Predict fit before you sign
0 style [ built ]
→
1 fit [ next ]
→
2 outcome [ north star ]
[ 01 · Problem ]

A perfect match at first sight.
Costing millions in the end.

In 2011, strikers Andy Carroll and Luis Suárez arrived at Liverpool FC. Carroll, their most expensive transfer in history, didn't succeed. Suárez became one of their most influential players.

Both were excellent players at their previous clubs. The difference wasn't quality. It was fit.

Carroll is a tall target man thriving from crosses, but Liverpool did not play that way, and did not have the squad to accommodate his style. The more mobile Suárez gelled much better with the players surrounding him on the pitch.

Not a standalone case of bad luck. In the past 10 seasons in the Belgian first league alone, unsuccessful transfers added up to a total loss of €750M.
A recruitment decision with serious impact
ANDY CARROLL
  • 58 games / 11 goals
  • Bought for €35M
  • Sold at a £20M loss
LUIS SUÁREZ
  • 133 games / 82 goals
  • Bought for €22.7M
  • Sold at a £75M profit
[ 02 · The gap ]

Fifteen years later and vastly more data later,
clubs still evaluate players in isolation.

Physical data, event data and tracking data are everywhere. None of it answers the only question that decides a transfer.

Distance covered
How much a player ran.
Individual
Passes completed
What he did with the ball.
Individual
Expected goals
How well he finished.
Individual
“How will this player actually play alongside the ten I already have?”
Interaction is where value is made or lost. And it's the one thing the data leaves out.
[ 03 · Solution ]

Predict fit before you sign.

Gaffer scores how players combine, using the event data clubs already license — computable for players who have never shared a pitch.

Input · data you already license
target
your #10
your #9
The target's style fingerprint, set next to every player already in your squad.
The engine
Pairwise chemistry
Scores every squad player against the target — positive and negative. Who lifts whom, who clashes.
Output · illustrative
Fit vs squadTarget · LW, 22
Playmaker #10
+0.8
Left-back #3
+0.6
Striker #9
+0.3
Holding mid #6
−0.2
A fit score for each partnership — and each one opens to the attributes that produced it.
Closes the loop Gap analysis→Search→Shortlist→Fit→Player log
[ 04 · Based on published research ]

Not a new idea. A published one.

The mechanism behind the chemistry score is peer-reviewed. Gaffer is the first to turn it into a product.

1

Joint impact

Every two-player interaction on the pitch — a pass into a take-on, a pass into a shot — is valued and summed per pair, per 90.

Measured
2

Predictive model

A model trained on pair features estimates chemistry for players who have never played together.

The mechanism
3

Fit score

Pairwise chemistry, rolled up against a club's own definition of fit — the number a scout acts on.

The product
Bransen & Van Haaren (2020), peer-reviewed, presented at MIT Sloan. SciSports co-authored it and never shipped it. A co-author now leads data at Club Brugge. Nobody has run it on club-owned behavioural data.
[ 05 · Landscape ]

Everyone captures player attributes,
no one models the pair.

InteractionalIndividual
GAFFER
Pairwise scores · fit predictions · chemistry models
Not recordable
How two players interacted can only be inferred — and the pair a club is deciding on has never played together at all.
SciSports
Contributed to the 2020 research, never productised
Wyscout · Hudl · StatsBomb · Opta · Impect · SkillCorner · Track160 · Catapult · STATSports · PlayerData + 20 moreEvent · physical · tracking data
Transferlab · Smarterscout · Comparisonator · DataMB · Marquee · Twelve · Gradient Sports · HiddenKickAI + 15 moreMetric builders · LLM wrappers · comparison tools
Data captureData generation

“We'd do that for coaches too — if someone clashes with our DNA, we don't want them either. I get the concept.”

Thomas Rypens
Head of Innovation, Club Brugge
on predictive-fit modelling
[ 06 · Product ]

Three models. Each built on the one before.

Same underlying data. Increasing context: one player → a partnership → the whole match.

Model 0

Playing style

What players are like this player?
Style fingerprint from how he actually plays, not his stat line. Surfaces lookalikes across leagues and positions.
Built & validated
Complete
“Find a left-winger like Doku, under 23, under €10M.”
→
Model 1

Fit

How does he fit with the others?
Sets his fingerprint next to the squad and reads every partnership — who lifts whom, who clashes.
This round
In 6 to 12 months [ post pre-seed ]
“Would this target combine with our front three?”
→
Model 2

Outcome

What happens, in full context?
Drops player and fit into a real team, opponent and scoreline. The “what if” engine, in two tenses.
North star
In ~3 years [ post-seed ]
“Start him against Marseille last season — do we score more?”
[ 07 · Potential ]

An issue bigger than football

We start in football, where our network is strongest and the leagues are wealthiest. Every roster sport has the same gap: quality is measured, fit is guessed.

Eligible clubs — already licensing data Gaffer sits on
Football
~1,500
Of 3,986 professional clubs (FIFA), those already licensing event, physical or tracking data.
Basketball
~450
30 NBA + ~365 NCAA Division I + ~60 EuroLeague and top-division European clubs.
American Football
~300
32 NFL + ~135 FBS + ~130 FCS programs. Near-universal PFF and Catapult adoption.
Hockey
~170
32 NHL + 32 AHL + ~100 top-division European clubs. Sportlogiq and Stathletes normalised.
Rugby
~130
~90 professional union clubs across six leagues + ~30 rugby league clubs.
Baseball
~80
30 MLB organisations + NPB, KBO, LMB, CPBL. Fewest clubs, deepest analytics maturity.
~2,630
eligible clubs worldwide
Range 2,260 – 3,160. Bottom-up from vendor client counts and per-tier penetration.
€150M – €250M
spent every year on third-party recruitment data
Event data €35k–€100k+ · video and physical €28k–€85k · tracking €20k–€60k, per club per year.
Gaffer takes share of a budget line that already exists. We are not asking clubs to fund a new category.
[ 08 · Traction ]
Model 0
Built & validated

Player similarity proven on limited data — the proof case for Model 1.

6 clubs
In active conversation
GentAntwerpClub BruggeOH LeuvenMechelenKortrijk

“Our new signing only speaks Spanish. He's left to fend for himself — that costs real time before he can perform.”

Peter CatteeuwHead of Performance, Royal Antwerp

“Group dynamics are hard to quantify. But we think about it — nationality, language, dressing-room composition.”

Nathan De RidderLead Recruitment Analyst, KV Mechelen

“We're working on that, but it's still in its infancy.” — on squad modelling and character fit

Jan BodenHead of Data & IT, OH Leuven
[ 09 · Team ]

Built by lovers of the game.

Paired since IntelliProve. Full-time on Gaffer from September 2026.

Seppe De Langhe
SEPPE DE LANGHE

Technical co-founder

AI engineer. Data thesis on Computer Vision for Analytics in Football. Currently building AI at IntelliProve (health tech). Football data geek.

Senne Van Kerkhove
SENNE VAN KERKHOVE

Commercial co-founder

Commercial and operational roles across tech start-ups. Currently Head of Growth at IntelliProve, former Chief of Staff at Daltix & intuo. Football since childhood, still playing.

Assembling advisory team (conversations pending)
R&D

Data leadership at a top-flight club. Academic background in football ML.

Go-to-market

Built and exited a football-analytics company.

Product

Leads innovation at a top-flight club. Former product co-founder in football tech.

[ 10 · Raising ]
€300k
Pre-seed · 12 months
30%
Data acquisition
Event-data licensing (V0 → V1)
30%
Team
Two founders, full-time from September
15%
Product & infrastructure
Fit engine, compute, tooling
15%
Go-to-market
Events, branding, identity, travel
10%
Operations
Legal, accounting, admin, buffer