There is a widespread misconception about fans who check their phones while watching a live game. The assumption is that they are disengaged: distracted by social media, doing something unrelated, tuning out. The data tells a different story.
Fans who use a second screen during live sports are often the most engaged fans in the audience. They are not escaping the broadcast; they are extending it. They want more information about what they are watching. They want context, analysis, and something specific to argue about with the people they are watching with or the community they are part of online.
This distinction matters enormously for how sports media and fan platform products think about second screen content. If you assume second screen users are distracted, you design content that competes with the broadcast for attention. If you understand that they are looking for analytical depth, you design content that layers on top of the broadcast and makes the viewing experience richer.
What Fans Are Actually Looking For
We built a set of intuitions about second screen behavior by watching how fans interacted with prediction data during live games in our early pilots. The pattern was consistent: the content that drove the most second screen engagement was not live score updates (fans could see those on the broadcast), and it was not retrospective highlights (those belong post-game). It was forward-looking analytical data.
Specifically: what is the win probability right now, and why did it just shift? What is the upset-risk score for the remaining portion of the game given the current state? Which player performance outcomes are still in play that matter to how this game ends?
These questions are not answered by the broadcast. A broadcast commentator will describe what just happened and provide emotional context. They are not giving fans a quantified model of how likely different game-ending scenarios are. That quantified layer is the second screen's natural territory.
Pre-Game Second Screen: The Often-Missed Window
Most second screen product thinking focuses on in-game engagement. This makes intuitive sense: the game is happening, the broadcast is live, the emotional intensity is high. But in our experience, the pre-game window is actually the richer opportunity for prediction data.
The 30-60 minutes before a game starts is when fan communities are most actively discussing the matchup. Predictions have been made. Lineups have been announced. Whatever last-minute news has emerged has been processed. Fans are in a state of pre-game anticipation and they want to engage analytically with what is about to happen.
A pre-game prediction card on a second screen app that shows win probability, an upset-risk flag, and two or three specific prediction signals gives fans exactly what they are looking for at this moment. It gives community platforms specific content around which discussion can organize. It gives individual fans stakes in the game beyond their emotional rooting interest: they have a probabilistic prediction to track against the actual outcome.
Why Confidence Scores Are the Right Format
There is a specific reason why confidence scores work for second screen contexts better than declarative predictions or analysis paragraphs.
Second screen consumption is a divided-attention experience. Fans are watching the broadcast and checking their phone. They are not reading a 500-word analytical piece. They are scanning for something at a glance that either confirms their intuitions or gives them something to push back on.
A confidence score plus one key driver delivers all of this in a format readable in two seconds: "67% home side, key driver: defensive efficiency over last eight games." A fan who knows that defensive efficiency has been inconsistent will push back immediately. A fan who agrees will feel like the analysis confirmed their reading of the situation. Either way, they are engaged.
Upset-risk flags perform particularly well because they introduce surprise into a format that is inherently surprising. Telling a fan that a game the broadcast is treating as a clear-cut outcome has above-average model variance creates a specific kind of anticipation: they are now watching for the upset scenario to materialize or not. That is a different quality of engagement than simply watching to see which team scores more points.
Building Second Screen Retention, Not Just Activation
The second screen product challenge is not getting fans to open the app. It is keeping them coming back. This is where prediction data, used well, creates retention rather than just activation.
The mechanism is outcome tracking. A fan who saw a 67% win prediction before the game and watched the favorite win will come back next week partly because they want to check the model's track record. A fan who saw an upset-risk flag and then watched the upset happen will have a strong association between the platform and analytical credibility. Both experiences reinforce the value proposition of the second screen product beyond the game itself.
For fan community platforms, the retention loop is slightly different: fans come back because the discussion that prediction data sparked did not end with the game. Post-game analysis of whether the model was right, why the upset risk materialized or did not, and what the next matchup looks like draws the same analytically-minded fans back into the community thread. The prediction data becomes a thread that connects pre-game, in-game, and post-game engagement into a coherent experience.
A Note on In-Game Prediction Updates
We build pre-game prediction data, but a natural question for second screen product teams is whether in-game probability updates are worth the engineering investment.
In-game probability updates are valuable when they can respond meaningfully to game events in near-real-time: a turnover that shifts win probability, a key player going to the bench, a scoring run that crosses a threshold where the upset scenario becomes more or less likely. If your in-game data pipeline is fast enough to produce updated probabilities within seconds to minutes of game events, this is a strong second screen feature.
If your data pipeline is slower, and this is a realistic constraint for teams without a dedicated real-time data infrastructure, in-game updates can actually undermine credibility. A probability that updates 15 minutes after the event it is responding to will appear wrong relative to what the broadcast has already moved past. In that scenario, a stable pre-game prediction with a clear model update cadence (updated once at game time, not after each possession) is more reliable than an in-game update that is systematically late.
We are explicit about our own update cadence in our API documentation, and we advise clients to surface that cadence to their users. Second screen fans are analytically minded; they can handle knowing that a model updates twice per day rather than per possession, as long as you tell them that upfront.