- Retail
This project presents an innovative approach to estimating On-Shelf Availability (OSA) within nanostores, key components of retail channels in emerging markets like India. Utilizing sales data and field study findings in Mumbai, we developed and validated two distinct models: a probabilistic model and a Machine Learning model. The probabilistic model was constructed based on sponsors’ available sales data and by the formulation of five key assumptions, which were afterwards assessed through both qualitative and quantitative components of a field study. This approach revealed that certain assumptions were not fully validated, thereby weakening the model. Moreover, the approach revealed new insights into the purchasing behavior of store owners which had not been previously considered, such as their tendency to buy from wholesalers instead of directly from the sponsor company. In contrast, the classifier Machine Learning model, notably Random Forest, yielded superior accuracy. This classifier model was trained on actual inventory data gathered during the field study and is capable of relying solely on features derived from the sponsor company’s data without the need for qualitative assumptions. Our findings underscore the significance of robust modeling techniques based on relevant data to enhance OSA estimation for Consumer-Packaged Goods (CPG) companies operating in nanostore contexts. Our recommendations include the adoption of the Machine Learning model, emphasizing its scalability and robustness as well as the importance of data collection to extend OSA visibility beyond Mumbai or India to broader regions. This study offers valuable insights and actionable recommendations for tackling the lack of empirical data for OSA estimation within nanostores.