An Algorithmic Data Pipeline Architecture for the Production of Personalized Telecom Product Offers

Telecommunications sales management require a better understanding of the client's profile and customer needs before selecting and proposing the “right” product at the “best” price, choosing from an increasingly complex collection of offers and tariff packages. To this end, various methods are aiming to understand and estimate the user's behavior, predict traffic and willingness to pay. Based on such information, sales channels select and propose a product at an acceptable price, which will both be attractive to the customer and provide satisfactory and sustainable revenue to the company. Such methods require complex machine learning exploration and exploitation algorithms, applied to high volumes data of variant complexity, with the additional requirement to execute the whole process instantly so that sales team can access the proposed offers in front of the customer. Furthermore, interaction outcome is fed back to the system to enable continuous updating of data, events and finally offers. The AutoSPRice project goal of Neurocom is a fully automated and autonomous revenue and customer lifecycle value management system, which produces the most appropriate offers for each sales channel both in batch and real-time conditions. The system can predict, using the appropriate machine learning algorithms, how future telecom subscriber needs will evolve at individual and group level (company, family). It also assesses the user’s financial capacity and willingness to buy from different offerings. Hence, it enables sales channels to identify what is the right product that best meets a customer’s needs by using reinforcement learning. Combined with systems that can perform big data management and processing, it immediately provides responses out of large datasets. Using exploration-exploitation algorithms, it identifies the price to achieve the best chance to be accepted by the customer and the provider. The system has been developed and tested extensively under realistic usage scenarios.