Improving Asset Allocation in a Fast Moving Consumer Goods B2B Company: An Interpretable Machine Learning Framework for Commercial Cooler Assignment Based on Multi-Tier Growth Targets

Renato Castro, Rodrigo Paredes, Douglas Kahn · 2025

Read Improving Asset Allocation in a Fast Moving Consumer Goods B2B Company: An Interpretable Machine Learning Framework for Commercial Cooler Assignment Based on Multi-Tier Growth Targets by Renato Castro, Rodrigo Paredes, Douglas Kahn (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-06642.txt.

Introduction

Renato Castro, Rodrigo Paredes, and Douglas Kahn predict which shops will actually grow after receiving a free commercial beverage cooler. In fast-moving consumer goods, placing coolers is a costly bet on future sales, and machine learning rarely guides that placement. Churn and demand models are common, but asset allocation decisions stay mostly manual.

The study follows more than three thousand traditional trade clients of a Central American brewer, with a year of sales before and after installation. Gradient boosting models, explained through SHAP values, forecast which clients clear ten, thirty, or fifty percent growth targets. The best model scored strongly at every threshold, and simulations suggest better returns plus savings from skipping clients unlikely to grow.

Catalog record

CREATORRenato Castro, Rodrigo Paredes, Douglas Kahn
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE26 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-06642

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Improving Asset Allocation in a Fast Moving Consumer Goods B2B Company: An Interpretable Machine Learning Framework for Commercial Cooler Assignment Based on Multi-Tier Growth Targets Renato Castro, Rodrigo Paredes, Douglas Kahn Abstract In the fast-moving consumer goods (FMCG) industry, deciding where to place physical assets, such as commercial beverage coolers, can directly impact revenue growth and execution efficiency. Although churn prediction and demand forecasting have been widely studied in B2B contexts, the use of machine learning to guide asset allocation remains relatively unexplored. This paper presents a framework focused on predicting which beverage clients are most likely to deliver strong returns in volume after receiving a cooler. Using a private dataset from a well-known Central American brewing and beverage company of 3,119 B2B traditional trade channel clients that received a cooler from 2022-01 to 2024-07, and tracking 12 months of sales transactions before and after cooler installation, three growth thresholds were defined: 10%, 30% and 50% growth in sales volume year over year. The analysis compares results of machine learning models such as XGBoost, LightGBM, and CatBoost combined with SHAP for interpretable feature analysis in order to have insights into improving business operations related to cooler allocation; the results show that the best model has AUC scores of 0.857, 0.877, and 0.898 across the thresholds on the validation set. Simulations suggest that this approach can improve ROI because it better selects potential clients to grow at the expected level and increases cost savings by not assigning clients that will not grow, compared to traditional volume-based approaches with substantial business management recommendations. I Introduction The fast-moving consumer goods (FMCG) industry faces growing pressure to improve how physical assets are allocated to their customers. In the brewing and beverage industry, commercial coolers are one of the most visible and capital intensive assets, directly in trade marketing execution and sales results. When strategically placed in B2B retail environments, these assets can significantly increase sales volume and strengthen customer relationships, making them a key lever for commercial growth. Despite their strategic importance, cooler deployment decisions are still largely guided by experience, legacy practices, client loyalty, or simple volume-based metrics. These rules often overlook critical signals in customer behavior, historical trends, or the local market context, limiting their effectiveness. As data availability improves and Return on Investment (ROI) expectations increase, there is a need for smarter and more predictive approaches that link asset investments with measurable commercial outcomes. In this context, the ROI expectation is defined by the ratio of sum of the incremental margin and the cost savings of not investing the assets in not potential clients divided by the total coolers investment: purchase, transportation, installation and maintenance. Although research in B2B customer modeling has focused primarily on churn prevention and retention strategies [ 4 , 6 ] , few studies have addressed the challenge of asset allocation under growth uncertainty. Unlike churn models, which aim to flag customers at risk of leaving, cooler allocation requires identifying clients with high potential to grow when supported by physical investments. This paper introduces a machine learning framework to predict which B2B beverage clients are most likely to deliver strong sales growth after receiving a commercial cooler. We frame the task as a multithreshold binary classification problem using sales uplift targets of 10%, 30%, and 50%. The models are trained on 3,119 B2B clients using 12 months of pretreatment and posttreatment data and leverage gradient boosting algorithms (XGBoost, LightGBM, CatBoost) with SHapley Additive exPlanations (SHAP) for interpretability. Unlike traditional Customer Life Value estimation, which targets long-term value, our approach focuses on short-term intervention-specific results, making it highly relevant for tactical allocation decisions in B2B operations. Our findings offer a scalable, data-driven method to guide asset deployment, helping decision makers target the right clients, at the right time, with higher confidence and ROI. The main contributions of this work are: • A comprehensive framework for predicting potential customers in B2B FMCG asset allocation scenarios targeting different growth tiers (10%, 30%, 50%) to support flexible business strategies • Integration of SHAP-based interpretability to understand feature importance and propose new strategies in the operation • Empirical validation on real-world data from 3,119 B2B clients with demonstrated business impact II Related Work II-A B2B Customer Churn Prediction In the FMCG research space, Mirković et al. [ 10 ] built a churn model using only invoice-level data and still achieved strong performance, showing that robust predictions are possible even with minimal inputs. Their use of a multi-slicing strategy to boost data efficiency, and their finding that ensemble methods consistently outperform single models, are closely aligned with the modeling approach used in this work. Other studies on demand prediction [ 12 ] and FMCG market analysis [ 7 ] have shown the promise of ML in the sector, but the majority focus on forecasting or segmentation. The application of machine learning to asset allocation decisions, especially in B2B contexts, is still a gap in the literature that this work aims to address. II-B Interpretable Machine Learning in Business Applications The adoption of ML in business-critical applications requires model interpretability to build stakeholder trust and enable actionable insights. Daz et al. [ 9 ] applied explainable AI techniques to the prediction of churn of B2B customers for enterprise software, using SHAP values to reveal the most influential factors behind the risk of churn. Recent advances in interpretable ML have shown particular promise in customer prediction tasks. Maan et al. [ 8 ] developed transparent churn prediction models using XGBoost with SHAP values, while telecom research has achieved high AUC (0.79) with XAI-Churn TriBoost while maintaining interpretability [ 1 ] . The integration of SHAP with gradient boost models has proven effective in identifying key drivers of customer behavior [ 11 ] , enabling targeted strategies based on the importance of the features [ 2 ] . II-C Resource Allocation and Asset Optimization Although there is extensive research on customer prediction and churn management, the application of machine learning to physical asset allocation in B2B contexts remains limited. Wang et al. [ 13 ] proposed a general framework for resource allocation using historical scenario data, demonstrating the superiority of machine learning over conventional optimization methods. Colias et al. [ 3 ] addressed the optimization of B2B products using mixed logit models and non-linear programming, integrating machine learning with economic theory to optimize features and pricing. However, the specific challenge of predicting ROI for physical asset deployment in FMCG B2B contexts represents a gap in the literature. Our work addresses this gap by combining profit-driven modeling approaches with interpretable ML techniques to optimize asset allocation decisions. III Problem Formulation III-A Business Context The allocation decision involves selecting which B2B clients, such as retailers, restaurants, and so on, should receive these assets to maximize overall business impact. Given limited resources and high asset costs, a commercial cooler with all related logistics costs is roughly $974. Optimal allocation is crucial to direct these assets to clients with attributes that support sales growth, while avoiding placement with clients unlikely to show significant growth. III-B Dataset Description Fig. 1: Monthly number of unique B2B clients receiving coolers from January 2022 to March 2025. Our dataset comprises transactional data from 3,119

Read the complete text