Quantifying the impact of marketing activities on company business goals is a complex task for Marketing Directors due to the interplay between internal initiatives and the influence of external factors, such as market variations or competitors' strategies. Marketing mix modeling (MMM) emerges as a regression-based method that quantitatively measures the contribution of different marketing channels to sales generation. Linear Regression is traditionally used within MMM to analyze sales or other Key Performance Indicators (KPIs) as outcomes, influenced by marketing activities and external elements as input variables. Yet, it often fails to capture the nonlinear and complex relationships inherent in marketing data. This thesis investigates the integration of tree-based ensemble algorithms, specifically Random Forest and Light Gradient Boosting Machines, into MMM development, focusing on determining its feasibility and potential benefits in terms of predictive accuracy and explainability. In the absence of direct estimates of the regressors' coefficients, typical of linear models, the study employs Shapley values, from cooperative game theory, to estimate the impact of each variable on the model's predictions. It presents a customized pipeline to simplify the understanding among analysts and stakeholders of the entire process of developing and evaluating MMMs. The effectiveness of the proposed pipeline is validated through an analysis using data from a national insurance company, grounding the work in a practical business context. The findings highlight that tree-based models surpass linear models in terms of predictive accuracy in the field of MMM, while maintaining a satisfactory level of interpretability. From a managerial perspective, this research introduces a new methodological decision framework that addresses the limitations of traditional parametric MMMs, improving the firm's understanding of marketing analytics and campaign effectiveness.
Quantificare l'impatto delle attività di marketing sugli obiettivi aziendali rappresenta una sfida complessa per i manager a causa dell'interazione tra le iniziative interne all'azienda e l'influenza di fattori esterni, come le variazioni di mercato o le mosse dei concorrenti. Il marketing mix modeling (MMM) emerge come un metodo regressivo capace di quantificare come i vari canali promozionali influenzino le vendite. Tradizionalmente, viene utilizzata la Regressione Lineare per studiare la relazione tra le vendite (o altri Indicatori di Prestazione) e le strategie di comunicazione o fattori esterni. Tuttavia, questo metodo spesso fallisce nel descrivere le relazioni complesse e non lineari presenti nei dati di marketing. Questa tesi esplora l'integrazione di algoritmi di machine learning basati su alberi decisionali, come Random Forest e Light Gradient Boosting Machines, per migliorare il Marketing Mix Modeling. L'obiettivo è valutarne l'efficacia e i benefici in termini di accuratezza predittiva ed intepretazione. In assenza di stime dirette dei coefficienti dei regressori, tipiche dei modelli lineari, lo studio impiega i valori di Shapley, tratti dalla teoria dei giochi, per stimare l'impatto di ciascuna variabile sulle previsioni del modello. In questo contesto è stata creata una pipeline per facilitare la comprensione del processo di sviluppo e valutazione del MMM tra analisti e stakeholder. L'efficacia di questa soluzione è stata confermata con un'analisi su dati reali provenienti da una compagnia assicurativa nazionale, dimostrando la sua utilità in un contesto aziendale concreto. I risultati evidenziano che i modelli ad albero superano i modelli lineari in termini di accuratezza predittiva nel campo del MMM, mantenendo un livello soddisfacente di interpretabilità. Da una prospettiva manageriale, questa ricerca introduce un nuovo approccio metodologico che, oltre a risolvere le limitazioni dei MMM parametrici, arricchisce la comprensione aziendale sia nell'ambito dell'analisi tecnica che nello sviluppo di strategie concrete di investimento.
Integrating tree-based ensemble algorithms for marketing mix modeling
CERRI, MARTA
2022/2023
Abstract
Quantifying the impact of marketing activities on company business goals is a complex task for Marketing Directors due to the interplay between internal initiatives and the influence of external factors, such as market variations or competitors' strategies. Marketing mix modeling (MMM) emerges as a regression-based method that quantitatively measures the contribution of different marketing channels to sales generation. Linear Regression is traditionally used within MMM to analyze sales or other Key Performance Indicators (KPIs) as outcomes, influenced by marketing activities and external elements as input variables. Yet, it often fails to capture the nonlinear and complex relationships inherent in marketing data. This thesis investigates the integration of tree-based ensemble algorithms, specifically Random Forest and Light Gradient Boosting Machines, into MMM development, focusing on determining its feasibility and potential benefits in terms of predictive accuracy and explainability. In the absence of direct estimates of the regressors' coefficients, typical of linear models, the study employs Shapley values, from cooperative game theory, to estimate the impact of each variable on the model's predictions. It presents a customized pipeline to simplify the understanding among analysts and stakeholders of the entire process of developing and evaluating MMMs. The effectiveness of the proposed pipeline is validated through an analysis using data from a national insurance company, grounding the work in a practical business context. The findings highlight that tree-based models surpass linear models in terms of predictive accuracy in the field of MMM, while maintaining a satisfactory level of interpretability. From a managerial perspective, this research introduces a new methodological decision framework that addresses the limitations of traditional parametric MMMs, improving the firm's understanding of marketing analytics and campaign effectiveness.File | Dimensione | Formato | |
---|---|---|---|
2024_04_Cerri_01.pdf
accessibile in internet solo dagli utenti autorizzati
Descrizione: Testo della tesi
Dimensione
10.73 MB
Formato
Adobe PDF
|
10.73 MB | Adobe PDF | Visualizza/Apri |
2024_04_Cerri_Executive Summary_02.pdf
accessibile in internet solo dagli utenti autorizzati
Descrizione: Executive Summary
Dimensione
578.29 kB
Formato
Adobe PDF
|
578.29 kB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/218374