Towards an Efficient Method of Modeling 'Next Best Action' for Digital Buyer’s Journey in B2B
| Autor: | Nishant Niranjan, Kiran Rama, Stig Berg, Parag Chitalia, Anit Bhandari, Nandini Seth |
|---|---|
| Rok vydání: | 2017 |
| Předmět: |
Ensemble forecasting
Digital marketing Computer science business.industry 020208 electrical & electronic engineering 05 social sciences Rank (computer programming) Context (language use) 02 engineering and technology Machine learning computer.software_genre Email address Order (business) 0502 economics and business Digital asset 0202 electrical engineering electronic engineering information engineering 050211 marketing 020201 artificial intelligence & image processing Asset (economics) Artificial intelligence Literature survey business computer Simulation |
| Zdroj: | Machine Learning and Data Mining in Pattern Recognition ISBN: 9783319624150 MLDM Advances in Data Mining. Applications and Theoretical Aspects ISBN: 9783319627007 ICDM |
| DOI: | 10.1007/978-3-319-62416-7_8 |
| Popis: | The rise of Digital B2B Marketing has presented us with new opportunities and challenges as compared to traditional e-commerce. B2B setup is different from B2C setup in many ways. Along with the contrasting buying entity (company vs. individual), there are dissimilarities in order size (few dollars in e-commerce vs. up to several thousands of dollars in B2B), buying cycle (few days in B2C vs. 6-18 months in B2B) and most importantly a presence of multiple decision makers (individual or family vs. an entire company). Due to easy availability of the data and bargained complexities, most of the existing literature has been set in the B2C framework and there are not many examples in the B2B context. We present a unique approach to model next likely action of B2B customers by observing a sequence of digital actions. In this paper, we propose a unique two-step approach to model next likely action using a novel ensemble method that aims to predict the best digital asset to target customers as a next action. The paper provides a unique approach to translate the propensity model at an email address level into a segment that can target a group of email addresses. In the first step, we identify the high propensity customers for a given asset using traditional and advanced multinomial classification techniques and use non-negative least squares to stack rank different assets based on the output for ensemble model. In the second step, we perform a penalized regression to reduce the number of coefficients and obtain the satisfactory segment variables. Using real world digital marketing campaign data, we further show that the proposed method outperforms the traditional classification methods. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|