This Small Business Innovation Research (SBIR) Phase I project will establish the feasibility of optimizing the effectiveness of commercial displays using a predictive model of human attention deployment. Traditional solutions for optimizing displays require human data. For example, during A/B testing, the target audience is exposed to alternate versions of a display (comparables) and responses such as clicks are used to determine effectiveness. A key shortcoming of traditional optimization solutions is that the number of comparables that can be tested is severely limited by the available audience. These shortcomings will be eliminated if display optimization could be performed based on computer analyses rather than human data. The Phase I research will establish the extent to which computer-generated attention scores can predict catalog selections as measured by mouse clicks. It is anticipated that catalog items with high attention scores will be selected more frequently than catalog items with low and average attention scores.
Billions of dollars are spent annually on optimizing commercial displays using techniques that require human data, such as A/B and multivariate testing, contextual and behavioral targeting, consumer research, etc. If successful, this SBIR project will result in a software-as-a-service solution for optimizing commercial displays based on attention predictions. This proposed innovation will enable large scale display optimizations that are practically impossible to perform based on human data. Another key advantage of automated optimization is that it could be performed before exposing audiences to ineffective displays. As a result, the proposed innovation will increase revenue gains from commercial displays. This innovative solution could be applied to a variety of online and offline displays, including catalogs, shelf plans, and graphic ads. Beyond display optimization, attention models could lead to important scientific and technological advances, commercial applications, and health benefits.
Our NSF SBIR Phase I project established the feasibility of optimizing online product displays based on attention predictions. The first aim was to quantify the relationship between EyePredict scores and consumer selections. EyePredict scores quantify the likelihood that products will get noticed. They are generated by a computer model of visual attention that was developed and validated based on cutting-edge neuroscience research. In this project, consumer selections were measured by what people clicked on when asked to select items of interest from online catalogs. We performed a total of 12 validation (A/B) tests that yielded 326,258 clicks by 7,051 adults from all age groups and both genders. A series of correlational and causal analyses established the extent to which EyePredict scores account for and predict consumer selections. To prove a causal relationship between EyePredict scores and consumer selections, we used an automated optimization engine to generate ten pairs of catalogs in which a target item appeared with either an average EyePredict score ("before optimization") or a high EyePredict score ("after optimization"). The target items were 10 branded virtual gifts, including a Cal helmet, a Coors can, a Garnett jersey, a Jackson glove, a Lakers jersey, three Prius badges (green, love, and white), a Terminator skull, and a Toms loafer. Each catalog pair included the exact same collection of items but in different positions and the target item always appeared in the top-left quadrant. Optimization led to increased click-through rate (CTR) for 9 out of 10 targets. A one-tailed t-test demonstrated that this result is highly significant [p<0.001]. The target with the median increase in clicks after optimization – the Lakers Jersey – was associated with a 175% CTR increase. The project outcomes demonstrate the feasibility of optimizing online displays using a predictive model of visual attention. The broader implications is that the EyePredict technology can make commercial displays more effective by helping to bring relevant items to the attention of consumers.
