Did the Keynote Land? Read the Real Room, Not the Survey

A keynote either landed or it did not, and the room knew within the first ninety seconds. Yet the standard way to find out is a form emailed hours later, answered by a self-selected minority, after the talk has been flattened into one remembered impression.

This is recall bias, one of the most replicated findings in survey methodology. People store a peak and an ending, then reconstruct the rest. The slow build that earned the room, the joke that fell flat, the data slide where attention drained, all of it collapses into a number that tracks how the session closed.

Response rates make it worse. The people most likely to answer are the delighted and the furious. The large middle, the ones who drifted at minute eighteen and came back at minute thirty, rarely answer at all.

Almost every keynote is already recorded. A camera on the stage, sometimes a second on the audience, captures exactly the evidence a survey throws away: the room reacting in real time, second by second. Proof of Impact reads that footage and turns it into a measured account of how the talk performed, without asking anyone to remember anything.

Reading a room and reading one person are two different measurements, and it is worth being exact about which is which. A single-subject read works from the 468-landmark facial geometry described by Kartynnik and colleagues (2019). A room read does not. It detects the faces it can pick out across the frame, counts them, and estimates from a few keypoints per face how much of the room is oriented toward the stage. That is deliberate rather than a shortcut: an audience face lands at roughly 25 to 35 pixels in a venue frame, and a fine-grained facial read at that size measures noise. So the room read is aggregate by construction. It reads the room as one body, and scores no individual.

Proof of Impact measures the audience signals it can read from the room. Attention tracks where the room is oriented, heads turned toward the stage rather than phones and exits. Synchrony measures how together the room moves, because a room that laughs, nods, and stills in unison is a room being carried. Engagement tracks how that holds across the session.

Two more close the loop. Congruence reads how consistently the visible reaction holds together, drawing on the Duchenne distinction (AU6 plus AU12) from Ekman and Friesen's Facial Action Coding System. It describes the expression, not whether anyone meant it. Momentum, our Momentum Signature, tracks whether the room builds or fades across the arc of the talk.

The engine reports the strongest genuinely-measured one as an index out of 100, under its true name, not as an invented composite. When the footage is too dark, too distant, or too brief to read honestly, the engine abstains rather than returning a number.

A single number is a verdict, not a lesson. The more useful output is the arc. Proof of Impact reports how the room's engagement moved across the opening, middle and closing thirds of the session, so a programme team can see where it built and where it sagged without a survey. The report does not pinpoint a clock time and it does not read any individual.

This is where measurement becomes craft. A programme team can see which sessions carried a room and which did not, and can compare the same slot across years. None of it requires interrupting the audience, handing out clickers, or trusting anyone's memory.

Conferences run on sponsorship, and sponsors want proof rather than vibes. A satisfaction average from a thin response pool is a weak artefact to attach to a major activation. One room-level index out of 100, drawn from the room on the day and reported with the phase of the session where the audience leaned in, is a defensible measure of the attention the session delivered.

For organizers, that turns the post-event report from an apology for low response rates into an asset. You can rank sessions by how they performed, brief next year's speakers with evidence, and price stages by measured attention rather than seat count. For a bureau, a consistent Impact Index across engagements is a credential a testimonial cannot match.