Recommender Systems - Part 1: Motivation & Basics

Content

This blog post answers two areas of questions:

  1. For whom is a Recommendation System relevant? Why?
  2. What basic variants are there? How complex is a corresponding implementation?

I focus on a solid overview in this post. Details are described by my colleague Josef Bauer in the subsequent parts two and three.

What does a Recommendation System do?

A recommendation system specifically addresses the individual interests of individual consumers. Conversely, it prevents the interruption of a customer journey due to overwhelm or frustration. Therefore, recommendation systems are central, especially with large product and service offerings. It is important to consider the user behavior of customers. While some pages explicitly invite browsing, others focus on quickly finding the optimal product.

Does my company benefit from the implementation of a recommendation system?

Digital B2C business models benefit the most from recommendation systems, as they directly influence the most important key performance indicators (KPIs). An introduction significantly enhances the user experience. Additionally, consumers increasingly expect good recommendations as a standard feature. The optimization of an existing recommendation system also serves to unlock untapped potential.

In addition to this main effect, recommendation systems have a number of side effects:

  • They lead to a higher understanding of data among the involved teams and promote a conscious engagement with actual user behavior and preferences.
  • Companies can use the insights gained for optimizing both the offering and the user experience.
  • Implemented recommendation systems can be used as an additional testing tool, as parts of the traffic can be controlled through them.

Many details determine how well this instrument can be used. Recommendation systems can still be roughly divided into two flavors:

konsumenten-vs.-produkt

Option 1: Product-based Recommendations (Item-based Approach)

The first option is closely oriented to the available products. The consumer sees products that are similar to those already considered. "Similarity" can mean many different things in this context. Examples include the product category, the price, or even the manufacturer. Of course, a combination of different dimensions is also conceivable.

Option 2: Consumer-based Recommendations (Consumer-based Approach)

In the second option, the focus is not on searching for similar products, but on similar customers. The resulting recommendations may strongly resemble or not resemble the first option at all. Such an approach attempts to model the preferences and decision-making logics of customers underlying their purchasing decisions, and less the "objective" similarities between products.

Which option fits my use case?

In times of the GDPR, the selection is significantly driven by data availability and legally permissible usability. However, a product-based approach is suitable for the introduction. From a practical perspective, this option can be well maintained as a table that is recalculated at fixed intervals. This greatly facilitates integration into mobile websites and apps. The consumer-based approach offers greater potential but depends on a reliable streaming infrastructure.

What are the first steps?

geschaeftsmodell-infrastruktur-datascience

To implement recommendation systems, inventories are required at three different levels:

  • How directly is the business model related to a Recommendation System? Is it a central component or more of a bonus? How significant is the actual added value for the customers?
  • How easily can a Recommendation System be integrated into the existing infrastructure? Are the necessary data available and can they be used for such purposes?
  • Is the necessary data science know-how available in the company? Is there already a team that can take on this task, or do new structures need to be created?

If these three questions are sufficiently answered, the actual work can begin. The details are described by Josef Bauer in his next post

Link
https://www.btelligent.com/blog/recommendation-systems-teil-1-motivation-grundlage/
Summary
This blog post addresses two main areas: the relevance of recommendation systems and their implementation variants. Recommendation systems cater to individual consumer interests, preventing customer journey disruptions due to overwhelm or frustration, especially in large product and service offerings. Digital B2C business models benefit significantly, as these systems enhance user experience and meet consumer expectations for quality recommendations. They also foster better data understanding among teams and can optimize both offerings and user experience. Recommendation systems can be categorized into two approaches: item-based, which suggests products similar to those already viewed, and consumer-based, which focuses on recommending products based on similar customer preferences. The choice of approach is influenced by data availability and legal considerations, with item-based being easier to implement. To implement a recommendation system, businesses should assess the system's relevance to their model, integration ease with existing infrastructure, and the availability of data science expertise. Once these factors are evaluated, the implementation process can begin, as detailed in subsequent posts by Josef Bauer.