A carregar...
Publicação
Modelo de previsão do contacto decisor no contexto empresarial
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 2.69 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
No setor B2B de uma empresa de telecomunicações, cada empresa/cliente tem associados vários
contactos à sua carteira. Deste modo, a comunicação com um cliente específico traduz-se num grande
desafio, dado que se torna difícil para a operadora identificar qual o número mais eficaz para o contactar.
Note-se que o contacto com o cliente é um fator crucial para o sucesso de qualquer empresa, no que diz
respeito ao aumento a satisfação do mesmo, prolongamento dos períodos de fidelização e à identificação
de oportunidades comerciais.
Posto isto, o objetivo do presente projeto de trabalho consistiu em desenvolver um modelo
preditivo, que estimasse a probabilidade de um determinado contacto ser o contacto decisor de um
cliente. Por outras palavras, pretende-se que o modelo identifique o contacto da pessoa que decide a
recusa ou aceitação de uma proposta, em nome de uma empresa.
Para esse efeito, foi criada uma variável resposta binária, sendo que, foram testadas quatro
formulações distintas da mesma, a partir da informação dos resultados de negócio. Este projeto teve por
base metodologias de manipulação e modelação de dados, nas áreas de Estatística e Machine Learning,
sendo implementado na linguagem de programação Python. Neste sentido, para cada formulação da
variável resposta, efetuou-se uma comparação entre dois tipos de modelos, o Random Forest e o
Gradient Boosting, ambos com otimização Bayesiana de hiper-parâmetros.
Posteriormente, foi selecionado o modelo que apresentou melhores resultados e procedeu-se à
validação dos respetivos outputs, em campanhas de telemarketing. Desta forma, verificou-se que que o
modelo desenvolvido permitiu dar resposta ao desafio de identificar o contacto decisor, tendo-se
revelado ser uma mais-valia para a empresa de telecomunicações em questão. Por conseguinte, o
modelo garante ganhos de eficiência e taxa de decisão superior, em comparação com a metodologia
anteriormente utilizada para contacto de clientes nos processos de negócio.
In the B2B sector of a telecommunications company, each enterprise/client has several contacts associated with its portfolio. Consequently, it arises the challenge of identifying the decision-maker contact, since it becomes difficult for the operator to identify the best number to contact. Noteworthy the contact with the customer is an important factor to increase customer satisfaction, extending loyalty periods and identifying business opportunities. In this context, the objective of this project was to develop a predictive model that estimates the probability of a specific contact being the decision-maker for a customer. In other words, the model aims to identify the contact person, who decides the refusal or acceptance of a proposal, on behalf of a company. For this purpose, a binary response variable was created, and four different formulations of it were tested, using business outcome data. The project relied on data manipulation and modeling methodologies in the areas of Statistics and Machine Learning, implemented in the Python programming language. In this sense, for each formulation of the response variable, a comparison was made between two types of models, Random Forest and Gradient Boosting, both with Bayesian hyperparameter optimization. Subsequently, the model that yielded the best results was selected and its outputs were validated through telemarketing campaigns. It was found that the developed model successfully addressed the challenge of identifying the decision-making contact and proved to be an asset for the telecommunications company. Indeed, the model ensures efficiency gains and a higher decision rate compared to the previously used methodology for customer contact in business processes.
In the B2B sector of a telecommunications company, each enterprise/client has several contacts associated with its portfolio. Consequently, it arises the challenge of identifying the decision-maker contact, since it becomes difficult for the operator to identify the best number to contact. Noteworthy the contact with the customer is an important factor to increase customer satisfaction, extending loyalty periods and identifying business opportunities. In this context, the objective of this project was to develop a predictive model that estimates the probability of a specific contact being the decision-maker for a customer. In other words, the model aims to identify the contact person, who decides the refusal or acceptance of a proposal, on behalf of a company. For this purpose, a binary response variable was created, and four different formulations of it were tested, using business outcome data. The project relied on data manipulation and modeling methodologies in the areas of Statistics and Machine Learning, implemented in the Python programming language. In this sense, for each formulation of the response variable, a comparison was made between two types of models, Random Forest and Gradient Boosting, both with Bayesian hyperparameter optimization. Subsequently, the model that yielded the best results was selected and its outputs were validated through telemarketing campaigns. It was found that the developed model successfully addressed the challenge of identifying the decision-making contact and proved to be an asset for the telecommunications company. Indeed, the model ensures efficiency gains and a higher decision rate compared to the previously used methodology for customer contact in business processes.
Descrição
Trabalho de Projeto de Mestrado, Estatística e Investigação Operacional, 2024, Universidade de Lisboa, Faculdade de Ciências
Palavras-chave
Telecomunicações Machine learning Random Forest Gradient Boosting Otimização Bayesiana Teses de mestrado - 2024