AI-Driven Personalization in B2B E-Commerce Product Recommendations: A Case Study of Etra Oy’s B2B Webshop
Pysyvä osoite
Kuvaus
Opinnäytetyö kokotekstinä PDF-muodossa.
Driven by the growth of digital commerce and the increasing complexity of industrial product catalogues, business-to-business buyers are facing information overload when trying to identify relevant products among tens of thousands of items. How recommender systems are designed is developing, yet empirical research on recommender systems in industrial B2B e-commerce remains limited, particularly from a holistic organisational perspective. This thesis aims to examine the limitations of the current recommender system used in the B2B webshop of Etra Oy, and to answer the research question "How can personalized AI-based product recommendations enhance cross-selling performance in Etra Oy's B2B e-commerce store?".
This study’s theoretical background and literature review cover artificial intelligence, different recommender system types, cross-selling in B2B e-commerce, and the DeLone and McLean Information Systems Success Model. This master's thesis is a qualitative single-case study with a pragmatic philosophy and an abductive approach, in which the empirical data were collected through six semi-structured interviews with employees from the case company. In addition, descriptive quantitative data on product sales and webshop user behaviour were used to support and contextualize the qualitative findings.
The findings suggest that the effectiveness of the recommender system is shaped mainly by organizational and data-related factors, rather than by the sophistication of the algorithm itself. The manual, rule-based logic that the company's recommender system is using currently appears to limit scalability, while inconsistent product data and limited resources con-strain recommendation coverage. Recommendations, when available, are perceived as accurate and trusted, but approximately sixty percent of products currently lack them. The findings of this thesis highlight that improving recommender system performance in B2B e-commerce, in this case, the company, requires prioritising data standardisation and organisational capacity before implementing advanced AI-driven recommender systems. Rather than transitioning directly toward AI-driven models, a staged approach, which first addresses data inconsistencies and decreases reliance on manual product relationship management seems necessary in order to introduce a scalable and reliable recommendation logic, while taking the B2B needs into consideration.