Parameterizable algorithm for price optimization on brick and mortar retail products using multi-armed bandit algorithm (MAB) and Robust Quadratic Programming (RQP)

• Grant number: 22/02635-2
• Support Opportunities: Research Grants - Innovative Research in Small Business - PIPE
• Start date: November 01, 2023
• End date: October 31, 2025
• Field of knowledge: Physical Sciences and Mathematics - Probability and Statistics - Applied Probability and Statistics
• Principal Investigator: Marcus Vinicius Veiga Roggero
• Grantee: Marcus Vinicius Veiga Roggero
• Company: Infopreços S/A
• CNAE: Desenvolvimento e licenciamento de programas de computador customizáveis
• City: São Paulo
• Associated researchers: Caroline dos Remédios Barbosa ; Daniel Soares Lopes ; Diogo Feliciano dos Santos ; José Henrique Camargo Leopoldo e Silva
• Associated research grant: 19/22822-9 - Parameterizable algorithm for price optimization on brick and mortar retail products using Multi-Armed Bandit Algorithm (MAB) and Robust Quadratic Programming (RQP), AP.PIPE
• Associated scholarship(s): 24/19663-4 - Development of APIs and business rules for the dynamic pricing module, based on the intelligent models developed for intelligent pricing, BP.TT; 24/19609-0 - Design of an optimized User Interface (UI) for Price Optimization in Retail: Improving User Experience through Analysis and Implementation of UX Practices, BP.TT; 24/14819-6 - Analysis, monitoring and exploration of data to understand the results of the pricing models developed., BP.TT; 24/08003-3 - UI development, based on intelligent pricing models and customer needs, including the deployment of functionalities and analytical panels, BP.TT; 24/02033-8 - Parameterizable algorithm for price optimization on brick and mortar retail products using multi-armed bandit algorithm (MAB) and Robust Quadratic Programming (RQP), BP.TT

Abstract

The pricing of products in physical food retail is a daily challenge for retailers. The size of the assortment and the variables involved make optimal pricing difficult. InfoPrice is a retail tech pricing startup that has products for monitoring competitive prices and data analysis since 2013. The company has a Big Data infrastructure, processing to receive data from various sources - including crawlers/scrappers -, statistical models for the treatment and quality of information, and price panel. In addition, it has two patents: the NoBreak Gondola and the SmartPrice hardware used to reduce the time of collecting prices in physical food retail. The objective of the project is the development of dynamic pricing software. In phase 1 of the PIPE project, the feasibility of models for forecasting demand, price-demand elasticity, product sensitivity, price optimization by-product with tests with the MAB-type algorithm and by category with robust quadratic programming (RQP). The project presented in this work is phase 2 of PIPE, whose objective is to develop price optimization models to maximize the retailer's revenue or profit. The research plan aims to unite the MAB and the RQP so that the retailer has a consistent pricing policy for the product categories. Minimization of deployment time on clients will occur with backtests and simulations. The dashboards will be used to monitor the models with customers. Subsequently, the models will be transformed into API modules, which will be connected to the Automatic Pricing Platform (IPA). The IPA competes directly with two national companies, which have difficulties in developing solutions because they do not have technological bases and several foreign companies with very low penetration in the Brazilian market. The total value of the project is R$ 2.4 million, of which R$ 999,252.76 was requested for PIPE 2 and approximately R$ 1.4 million in compensation from InfoPrice. As it is software, the budget requested in PIPE 2 for the two years of the project is mostly allocated to grants (87%). The rest of the budget is allocated to specialized technical consulting (10%) and spending on cloud computing for processing the models (3%). The evolution of the project will be measured with the TRL technological maturity level. InfoPrice's counterpart during the two years of the PIPE phase 2 project will be for the payment of salaries of 20 employees fully or partially dedicated to the project, five of which are fully dedicated employees. In 2021, a Series A investment round was carried out, with part of these resources earmarked for the project. The IPA has great potential, given that the Brazilian retail market currently has a size of 2 billion reais for price intelligence. The revenue projection for this product line in 2024 is R$ 16 million. (AU)