Abstract: Sports clubs need data to operate their fan-centric shops that reflect actual fan behavior at the moment of purchase, not what they later report in questionnaires. The aim of this study is to design and evaluate a system that, by analyzing fans’ movement and visual attention inside the shop, predicts purchase probability, explains its behavioral rationale in manager-readable language, and converts the output into concrete operational recommendations for layout, sponsorship pricing, and capacity planning. The research was conducted in the design science paradigm and based on the six-step model of Peffers et al. (2007). The “Parsa‑Fan” system was built on the FanShop‑Bench corpus with 36,674 anonymized sessions in 9 shops over 150 days, evaluated temporally and against 9 competing models with equal input computational budget and identical hyperparameter tuning. The system achieved first rank on the ranking metric (area under the precision–recall curve 0.4157 versus 0.4118 for the strongest baseline), although this advantage is small and, after family‑wise error correction, is statistically significant only against two baselines. The main finding is that the system recovered three behavioral mechanisms as readable curves; among them, purchase probability increases up to about 4.49 minutes of presence and then decreases, i.e., long lingering is a sign of hesitation rather than interest. Valuation of sponsor boards based on measured attention showed that the common “opportunity to be seen” method overestimates value by a factor of 4.60, and demand‑distribution‑based capacity planning produced 28.23% cost savings compared to mean‑based planning. A systematic search following the PRISMA guidelines identified 208 records, screened 144, and included 35 studies in the comparative matrix. This search also found the closest prior work (Li et al., 2026). Therefore the claim of this paper is not “firstness” but a distinctive combination: the sports context, implementation of the stimulus–organism–response chain within the computational structure itself, a prescriptive layer that separates latent demand from censored sales, and an architecture that is structurally free of identity re‑recognition and devoid of affect inference.