After a period dominated by branding and CSR, the current objective of retailers is focused on maximizing the ROI of marketing actions to release the stock accumulated during the months of inactivity, as well as to communicate the new launches to a consumer who has adopted new purchasing habits. Since the online environment is now strategically important, data science applications in e-commerce within the retail sector already have a long history of trying to improve online store efficiency, automate processes and refine the user experience to maximize sales.
Data science applications in retail
Although surveys such as the Big Data and AI Executive Survey 2019 conducted by NewVantage Partners reported that only 30% of companies in the United States were considered data-driven, the crisis triggered by the SARS-CoV-2 pandemic has accelerated the digital transformation. Retailers operating in online environments and already applying data science techniques have a competitive advantage during these uncertain times. Below, we review three of the most prominent applications:
1. Price management with a data-based approach
Managing pricing policy with the support of real insights rather than intuition is a valuable tool for boosting sales at times of low demand while taking advantage of its increase during seasonal peaks. This is possible as long as the retailer can unify their data under common characteristics and without excluding any sales channel.
With this initial classification, where data is stored in data lakes, processed and used to feed machine learning models, it would be possible to:
- Customize the discount or pricing policy based on the user: the data analysis is responsible for detecting common patterns and identifying customer clusters based on their previous online behavior as well as historical interests and purchases. This way, it is possible to design a strategy of promotions tailored to the types of users found and, therefore, with a greater probability of conversion.
- Define prices by segments: in this case, the products and services’ catalogue as well as the price are set taking into account wider audience segments. For example, one option would be to have standardized prices for the majority of users, but also to integrate another strategy with more aggressive discounts aimed at those customers whose purchase engine is based on the price. Lastly, they would include premium offers with services or special features for those seeking extra security.
2. Customized e-commerce upselling and cross-selling recommendations
One of the great references in these data science techniques is Amazon and its advanced algorithm to show recommendations to marketplace users. In this sense, the most established machine learning models are designed to:
- Offer an improved version of the chosen product (upselling).
- Recommend other products related to the purchased item such as a fitness ball with its air pump.
- Suggest items that similar users have purchased at the same time. This is something very common in fashion e-commerce, where the aim is to boost the sale of complete outfits.
The advantage of using machine learning to manage these recommendations is gaining efficiency by not needing to run hundreds of A/B tests to make decisions: it is the algorithm itself that determines which products to show each user in a personalized way. This is achieved by submitting the model to prior training, in other words, it is necessary to classify the data beforehand and set traits that relate the items to each other. After the implementation, these algorithms refine the selection of recommendations and optimize it constantly. They become very precise, as they are fed by an increasingly extensive data history that allows them to thoroughly analyze user reactions to suggestions.
3. Advanced attribution models with machine learning
The complexity of users’ shopping routes in retail has grown due to the implementation of an omnichannel strategy in which the boundaries between online and offline actions are blurred; the increase in the number of sales platforms; and the emergence of new advertising environments such as TikTok. In marketing, the analysis of the buyer’s journey has always been a subject of study, but until now it was impossible to trace with such a high level of precision. Those responsible for this are the advanced attribution models with machine learning.
The attribution models that work with machine learning record the user’s online behavior at all times. To do this, they integrate data collected from external platforms (for example, from advertising campaigns on different channels) and information from the site. Based on this realistic scenario, it is possible to control the profitability of each channel, redistribute budgets or evaluate the effectiveness of the affiliate network. Another advantage of this type of attribution model is its ability to detect fraudulent clicks.
Data science: a source of certainty for retailers
Making the right decisions has become crucial to the survival of many retailers, who are trying to make their way in a consumer environment deeply marked by uncertainty. For this reason, those retailers that apply these data science techniques in their e-commerce will be one step ahead of their competitors by being able to enrich the online experience of their users and have all the information to react to changes with flexibility and speed.
