Shelf Report — Predicted Attention (CAS)

Published: 3 min read

#What is predicted attention in the Shelf Report?

The Shelf Testing Report can be enriched with AI-predicted visual attention for each product. The enrichment is generated from AI analysis of each product's packaging image and produces:

  • A predicted heatmap — a Composite Attention heatmap showing where attention is predicted to fall on the product.
  • Predicted CAS (Composite Attention Share) — a single predicted score per product.
  • Component shares — the predicted Speed, Reach, and Hold shares that make up the CAS.

Predicted attention is an estimate based on AI analysis of the pack design. It is not collected from real participants and is not a measured eye-tracking result.

#How Predicted CAS is calculated

For each product, the prediction produces three predicted component signals — Speed, Reach, and Hold. For the set of products that were analyzed successfully, each component's share is calculated relative to that set:

  • Speed share = 100 × the product's predicted Speed signal ÷ the sum of all analyzed products' predicted Speed signals
  • Reach share = 100 × the product's predicted Reach signal ÷ the sum of all analyzed products' predicted Reach signals
  • Hold share = 100 × the product's predicted Hold signal ÷ the sum of all analyzed products' predicted Hold signals

Predicted CAS = (Speed share + Reach share + Hold share) ÷ 3

Each component's shares total ~100% across the analyzed products, and the CAS values also total ~100% across that set.

#Predicted CAS depends on the comparison set

Predicted CAS is a relative score. It is calculated against the specific set of products that were successfully analyzed together:

  • Adding or removing a product changes every CAS value. The denominator is the set of analyzed products, so each share is recalculated when the set changes.
  • CAS is not a measured percentage and is not a probability. It only describes how the predicted attention of one product compares to the others in the same analyzed set.

Measured VAI (Visual Attention Index) is different: it is calculated from real participant eye-tracking data and does not change when products are added or removed.

#When Predicted CAS is unavailable

At least two products must be analyzed successfully for Predicted CAS to be calculated. If fewer than two products are valid:

  • The CAS and all component shares are shown as unavailable.
  • Products that could not be analyzed are shown as unranked and are excluded from the comparison set.

#Limitations

  • Predicted heatmaps and Predicted CAS are AI predictions, not measurements of real gaze behavior.
  • The prediction does not replace eye-tracking data from participants; use it to complement measured results.
  • Because CAS is relative to the comparison set, a high CAS does not mean "high attention" in absolute terms — it means the product is predicted to attract more attention than the others in the same analyzed set.