Forecasting across hierarchical levels — for example, national, retailer, and category — is often inconsistent and inaccurate when done independently. We show, via a CPG case study, how applying hierarchical reconciliation methods with Tickr’s Forecasting Platform improves forecasting accuracy by an average of 14.9% enabling better business decisions.
Shopper Journey Data provides insights into why shoppers choose to purchase certain products at specific Retailers over others. Analyzing timestamps and basket-level purchases from the Ohio region, allows us to observe whether shoppers make subsequent trips for specific products on the same day. Our findings illustrate that walk-out rates vary by retailer, food category, and brand. Convenience and dollar stores have higher walk-out rates compared to grocery stores, likely due to their smaller size and limited assortments. Certain food categories have greater variation in walk-out rates across retailers, indicating that certain retailers are better at capturing the appropriate product assortment or provide more appealing pricing and promotions.
This post addresses the challenge of identifying substitutable goods in consumer behavior analysis by leveraging large-scale shopping cart data. We introduce a sequential probabilistic model called “SHOPPER” developed by Ruiz, Athey, and Blei, to analyze hundreds of thousands of shopping trips encompassing millions of transactions. Our approach captures the complexity of consumer purchase behavior, considering various factors such as shopping purpose, seasonal variations, in-store promotions, and personal preferences.
In a previous post, substitutable goods were identified by leveraging large-scale shopping cart data to estimate a sequential probabilistic model called “SHOPPER,” developed by Ruiz, Athey, and Blei. In this post, the analysis is expanded by analyzing shopping cart data from several retailers across various geographical regions. Similar products are queried to investigate how the ranked lists of substitutable products vary across different retailers and geographies. These results provide insights into what products are substituted at various retailers and for the same retailers in different geographies.
At Tickr, we enhance the data science life cycle using large language models (LLMs). We develop a solution, Generative Predictor Search (GPS), which integrates LLMs to improve the accuracy and efficiency of time series forecasts and provides intuitive interpretations of influential factors on forecasted variables. This approach not only reduces forecasting errors by 15.6% compared to naive univariate autoregression models but decreases run time by 13 times, establishing GPS as a leading solution in explainable forecasting.
In today’s data-driven world, businesses are increasingly turning to advanced technologies to gain a competitive edge. Large language models (“LLMs”) have emerged as a game-changer, enabling businesses to have intelligent conversations with data and extract valuable insights. In this study we explore the effects of LLM temperature, a concept borrowed from statistical physics and thermodynamics, on the impact of LLM business applications.