HYPOTHESIS · PRE-MATCH INTELLIGENCE · BETTING EDGE

The market prices form, injuries, and history.
It doesn't price the manager's face.

EchoDepth analyses 44 facial Action Units in pre-match press conferences to generate a Pre-Match Manager Confidence Score — a signal the market does not currently see. We're testing whether it predicts outcomes.

A manager can say "we're fully prepared." The face tells a different story. We measure the face.

Press conferences in backtesting dataset

Facial Action Units per frame

Directional accuracy in high-confidence signals

Request Data Partnership →See the hypothesis

The Opportunity

Alternative data has transformed betting markets. Emotional signal hasn't been tried yet.

Trading desks at major bookmakers and sports betting funds use GPS data, weather models, referee assignment patterns, and social sentiment. None of them are using pre-match manager emotional state — because until now, it wasn't measurable at scale.

Uncorrelated alpha

Manager confidence signal is derived from publicly available press conference footage. It has no correlation with existing physical performance data, injury status, or historical H2H statistics — meaning it adds independent signal to any existing model.

Involuntary signal

A manager briefing the media has no awareness of which specific AU combinations they're producing. Unlike team news, tactics, or press conference language — all of which are carefully managed — the facial signal cannot be optimised for public consumption.

24–48hr pre-match window

Pre-match press conferences typically occur the day before or morning of the match. The signal is available after the main betting markets have opened — creating a potential window between signal generation and market movement.

Scalable to 40+ leagues

Any competition with publicly available press conference video can be scored. Premier League and Championship are the initial focus. Bundesliga, La Liga, Serie A, and Ligue 1 are natural expansions. International tournament coverage possible.

The Hypothesis

When a manager's confidence signal diverges from the odds, something is being missed.

Bookmaker lines primarily reflect team-level performance data and market sentiment. Pre-match manager emotional state — specifically the divergence between performed composure and genuine confidence — is not in the model. If it's predictive, it represents a systematic pricing gap.

High Net Confidence (≥ 0.35)

64%

Win rate in backtesting sample where manager pre-match Net Confidence exceeded 0.35. Vs 48% average win rate for the same teams in the same period.

Low Net Confidence (< −0.10)

29%

Win rate when manager pre-match Net Confidence fell below −0.10. The signal diverges from market-implied probabilities in roughly 30% of these cases.

Important caveat

These figures are from an early-stage backtesting dataset. Statistical significance has not yet been established at publishable confidence levels. This is a live research hypothesis. The data partnership we're building is designed to generate the evidence base — with trading desk partners — before commercial data feed pricing is set.

Sample Analysis

Pre-match signals diverge. The market doesn't always notice.

Same competition. Same matchweek. Two managers. One showing genuine confidence. One performing composure while the face shows something different.

Manager A

Pre-match presser · Friday

High

Genuine Confidence58%

Instability Index22%

Net Confidence+0.36

AU6+AU12 dominant throughout. High Dominance. Low suppression signal. Matches what he's saying. Signal aligned with stated confidence.

Manager B

Pre-match presser · Friday

Suppressed

Genuine Confidence32%

Instability Index41%

Net Confidence−0.09

High AU1+AU4 (brow raise/pull), elevated AU15 (lip corner depression). Suppressed anxiety beneath composed delivery. Face diverging from words — signal flagged.

Data Partnership

Built for trading desks, quant teams, and alternative data buyers

Sports Bookmaker Trading Desks

Your pricing model uses form, injuries, H2H, and market movement. Manager confidence signal is orthogonal to all of these. Even a marginal predictive signal at scale has commercial value.

Betting Exchanges

Exchange matching engines can't see non-priced signals. If the manager confidence divergence predicts sharp money movement, it's actionable pre-movement.

Quant Sports Funds

Alternative data that has no correlation with existing factors is the only kind worth adding. Manager emotional state has zero overlap with standard sports analytics datasets.

Alternative Data Aggregators

If the signal validates, this is an exclusive dataset at launch. Early distribution partnership creates first-mover advantage across your client base.

The Data Product

Pre-Match Manager Score: what you receive and when

Data fields per match

→match_id, competition, date

→manager_a_genuine_confidence (0–1)

→manager_a_instability_index (0–1)

→manager_a_net_confidence (−1 to 1)

→manager_b_genuine_confidence (0–1)

→manager_b_instability_index (0–1)

→manager_b_net_confidence (−1 to 1)

→divergence_flag (boolean)

→market_implied_prob_home, market_implied_prob_away

→confidence_vs_market_gap (delta)

Delivery

FormatJSON via REST API or CSV push

Timing2–6 hours post press conference

Coverage (Phase 1)Premier League + Championship

Coverage (Phase 2)Bundesliga, La Liga, Serie A

Latency SLADelivered before match-day open (target)

History2023/24–present backtesting dataset

ExclusivityNegotiable at launch

Partnership modelAnnual data feed or rev-share on alpha

Questions from trading teams

Is this statistically validated?+

Not yet at publishable significance. We have directional signal from a 312-match backtesting dataset. The data partnership is structured to co-develop the validation — partners get access to the raw scoring data and can run their own backtesting. We set a price when the signal is proven, not before.

Is analysing public press conferences legally compliant?+

EchoDepth analyses publicly broadcast press conference footage under a Legitimate Interests basis (Article 6(1)(f) GDPR), applied to public figures acting in their professional and public capacity. Processing does not produce biometric identification profiles. Cavefish Ltd is ICO registered (ZB915633) and maintains a Legitimate Interests Assessment on file. A full DPIA is available on request.

How is this different from analysing press conference text sentiment?+

NLP/sentiment analysis scores the words — which are carefully selected for media consumption. FACS analysis tracks 44 involuntary facial muscle movements that managers cannot consciously control. The signal we're after is the divergence between what a manager says and what their face shows — that divergence is invisible to any text-based approach.

Can you cover international matches and tournaments?+

Yes — any competition with publicly available press conference video can be scored. International tournaments (World Cup, Euros, Nations League) are high-value targets because the stakes are highest and the market is deepest. Scheduling and resource constraints would determine phasing.

What's the go-to-market if the signal validates?+

Phase 1: annual data feed subscription per competition, structured like standard alternative data licensing. Phase 2: API with per-event pricing for lower-volume buyers. Phase 3: integration with existing sports data platforms (Sportradar, Genius Sports, Stats Perform). Revenue share on demonstrated alpha is also on the table for early partners.

DATA PARTNERSHIP

Get early access to
pre-match manager scores

We're building the validation dataset with a small group of trading desk partners. If the signal proves out, early partners get preferential pricing and exclusivity windows. If it doesn't, you've seen the raw data and methodology — no cost.