15:35 | 16:15
Keywords defining the session:
Takeaway points of the session:
- Recommendation tools can be used to offer products to our customers that better fit their needs, but also, at the same time, helps the company to better define the stock and portfolio, reducing the operational costs and then optimizing the returns for the company. As a result, a win-win solution can be achieved for both, customers, that have a better decision process, and the company, with improved profits.
- From a business point of view, in order to define the company portfolio, the available information in the Big Data about what customer purchase should be complemented with information about what customers think and declare to want. Conjoint Analysis is a powerful market research techniques that provide customers preferences about the different aspects that define the products, and then, can be used to complement the preferences obtained from the Big Data.
In a commercial offer, knowledge of user preferences is a beneficial factor for both the customer and company. From the point of view of business, personalization of products allows to generate value to the offer and therefore to the company’s income.
We know commercialization of Smartphones is a high changing market where the user has certain preferences, often unknown but not unpredictable thanks to the large amount of information available from previous purchases.
Characterization of clients based on their preferences and device use habits is necessary to optimize the efficiency of the commercialization of offers. Therefore, it is proposed to design and implement a platform that allows to determine the different profiles of customers and give a recommendation of terminal for each one of them.
From the technical view, we have been working in a system able to work with huge, real time data and multi-source information, with online self-learning in digital channels, and with the possibility of not having historical information to learn about, i.e., with cold start. Algorithm is based on the reinforcement learning model Contextual Bandits, like Netflix frame’s content selection, which is updated with real campaigns results. Pilot results was successful improving local campaigns.
However, in this work we wanted to add more value to the recommendation system adding the customer’s opinion in order to enrich the output of the tool.
To do this, we carried out a market research asking customers about Smartphone preferences. The research was done online, using the Conjoint Analysis technique.
Conjoint analysis allows the researcher to obtain the “utility” that customers give to different aspects in the purchase process, in this case, the Smartphone selection. Each smartphone can be characterized by different attributes (brand, price, screen size, memory, main camera, selfie camera, processor…), and each attribute has a different range of levels (Brand can be Apple, Samsung, Huawei or other).
The utility allows to find, for example, to what extend customers give more value to 128 Gb of Memory instead of 64 Gb, or to what extend customers are willing to pay 500 € instead of 250 €, and the most important, what a Smartphone should offer to increase the price from 250 € to 500 € (a better brand, more memory, bigger screen, better processor?).
Conjoint Analysis also allows to explore the complete universe of possible Smartphones, not only the current Smartphones available in the company stock, or, in the best case, in the market.
The utilities are then used to complement the scores obtained with the Reinforcement Learning algorithm using previous purchases. To do this, due to the huge different information available from Big Data (millions of sales) both information sources, Big Data and Conjoint Analysis, are clustered in order to do this match.
The result is a tool that allows Marketing Units to define the best portfolio to offer our customers taking into account previous sales (Big Data) and the customer declared insights (Conjoint Analysis), using all available Smartphones in the market and in the company stock, optimizing the profitability conditions.