A JuryTap session gives you a range and a median from a panel of real people who heard your case and answered alone, on their own phones, before anyone talked. This page says what that number is, what it is not, and the research behind both. We would rather you hear the limits from us than from the other side.
What it is
Individual answers, not a deliberated verdict. Each juror answers privately. You see every answer, who gave it, and when.
A range first, a median second. Dollar awards are the noisiest judgment jurors make. Twenty-four honest answers spread out; the spread is the information.
Your own calibration. Use it to decide what you will say, and what you will walk away from. Whether you show it to the mediator is your call, and work product is yours to keep.
What the research says
Deliberation pushes awards up. In experiments with more than 500 mock juries, the group's award ran systematically above the median of what its jurors wanted before they talked. A private poll is a floor more often than a forecast. Schkade, Sunstein & Kahneman, 100 Columbia Law Review 1139 (2000).
The first number jurors hear pulls their answer toward it. The amount a plaintiff asks for moves what mock jurors award. Decide what anchor, if any, your presentation shows, and read the range in that light. Hastie, Schkade & Payne, Law & Human Behavior 23(4) (1999).
Lawyers are overconfident about outcomes, and experience does not fix it. In a study of 481 litigators, lawyers who put their own odds of meeting their trial goal at 64% met it 57% of the time. Years of practice did not improve the accuracy. Goodman-Delahunty, Granhag, Hartwig & Loftus, Psychology, Public Policy & Law 16(2) (2010).
Plaintiffs misjudge offers more often than defendants do. In 834 California personal injury cases that went to trial after settlement talks failed, the plaintiff did no better than the rejected offer 53% of the time. Defendants erred less often but at far greater cost. Kiser, Asher & McShane, Journal of Empirical Legal Studies 5(3) (2008), Table 7.
Laypeople are a fair proxy. A meta-analysis of 53 studies and 17,716 mock jurors found damage awards did not differ between student and community samples. Bornstein et al., Law & Human Behavior 41 (2017).
AI jurors are not people. When GPT-4o, Claude and o1 were run as jurors on a case with a human comparison group, the models returned guilty verdicts at less than half the human rate (21% against 49%) and carried their own biases. Sun, Zappalà et al., Law & Human Behavior 49(5) (2025).
How to use it
Read the range before the median. Ask what the top quarter of the panel heard that the bottom quarter did not; the free-text answers usually say.
Treat the number as a floor on a deliberating jury, not a ceiling, and as evidence about this case rather than a prediction of a verdict.
Note the anchor. If your presentation named a figure, the panel heard it.
Run it before you commit to a number out loud. A second panel on the same case tells you more than a bigger first one.