Predictive Analytics of Digital Marketing and Sales Pipeline
Sandesh, Poudel (2019)
Sandesh, Poudel
2019
All rights reserved. This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-2019112622526
https://urn.fi/URN:NBN:fi:amk-2019112622526
Tiivistelmä
The purpose of the final year project was to integrate predictive analytical features to the Marketing Analytical tool of the case company. The primary objective of the project was to implement predictive models for classification of winning and losing sales cases in the pipeline and prediction of Key Marketing KPIs – Marketing Leads, MQL, and number of Visitors.
To execute the project, the data was collected from various social, advertisement and CRM channels of the case company. The data was collected using Python and processed with the R language. Machine Learning workflows were based on the functions and guidance provided by R packages - Caret and CaretEnsemble.
For both cases, predictive models were constructed and experimented with various machine learning algorithms and their combinations. The results were very accurate for classification problems and the prediction of numbers of website visitors. However, for two regression problems, the results were just adequate and further improvement was recommended. Overall, it can be concluded that all the defined objectives were achieved and the architecture has been set up to integrate additional recommended predictive analytical capabilities into the platform.
To execute the project, the data was collected from various social, advertisement and CRM channels of the case company. The data was collected using Python and processed with the R language. Machine Learning workflows were based on the functions and guidance provided by R packages - Caret and CaretEnsemble.
For both cases, predictive models were constructed and experimented with various machine learning algorithms and their combinations. The results were very accurate for classification problems and the prediction of numbers of website visitors. However, for two regression problems, the results were just adequate and further improvement was recommended. Overall, it can be concluded that all the defined objectives were achieved and the architecture has been set up to integrate additional recommended predictive analytical capabilities into the platform.