bonprix

Sascha Netuschil

A Journey of Transformation and Innovation

After completing my degrees in automotive engineering and cultural anthropology, I began my career as a mechanical engineer and later transitioned to a role as a social media analyst. During this time, I discovered my passion for data analytics, both in terms of technology and methodology. This passion led me to join bonprix as a data scientist 11 years ago, at a time when the term "data scientist" was not widely recognized. I quickly realized that we were not fully utilizing the wealth of data we were collecting. While we engaged in reporting, data analysis, and A/B testing, much of our data remained untapped.

I viewed web tracking data as one of the most direct ways to understand customer behavior, preferences, and pain points. At that time, we were also building the technological infrastructure to process our data more efficiently, which allowed me to propose our first two in-house data science projects: dynamic marketing attribution and real-time session churn prediction. While these projects were not the easiest to tackle, and we faced significant challenges, they provided invaluable learning experiences that helped us refine our methods and processes.

Over the following years, we took on more projects, leading to the formation of a dedicated data science team and eventually a full department. Simultaneously, we continued to enhance our technological infrastructure and adopted a product organization model to develop data science products in an agile manner. Today, I lead a department of 10 data scientists, collaborating with various teams to manage four data science product teams. Together, we develop AI-driven products for our sales department, covering a wide range of applications, including recommendation and personalization services, marketing models (such as attribution and marketing mix modeling), customer models (like predicted customer lifetime value and purchase probabilities), and A/B testing.

Harnessing Generative AI and Causal ML: Transforming E-Commerce Experiences

It will come as no surprise that Generative AI is a significant technological trend for the future of e-commerce. Currently, the initial hype curve seems to be flattening out, revealing which use cases are realistically achievable with GenAI and which are more challenging. E-commerce lacks some of the psychological and personal advantages of brick-and-mortar retail, particularly in fashion retail. For instance, elements like try-on capabilities, style consulting akin to good salespeople and the easy resolution of service inquiries are often missing. GenAI presents an opportunity to enhance the user experience in these areas. For example, we can implement realistic try-on features for various body types, develop fashion advisor chatbots or enable natural language queries on our website.

“While tackling exciting projects can yield substantial benefits for the organization, it’s easier to succeed if you’ve already built trust in the technology within your company”

Moreover, GenAI has immense potential for reducing internal costs in retail, such as cutting expenses related to photo shoots. If we can map different clothing items onto the same model photo to create realistic results, we can save significantly on content production costs. While results from image-generating models were disappointing just six months ago, we are now seeing models that address these issues more effectively. Although image-generating AI has excelled at creating attractive images, many commercial applications require the generation of specific, deterministic images rather than just "some" beautiful pictures. I see the technology moving in this direction.

Apart from Generative AI, I see great potential in causal ML. By predicting not just what will happen but also why it will happen, we gain a deeper understanding of our options for action, ultimately enhancing personalized customer treatment.

Navigating Technology Choices: Overcoming Challenges in Retail Tech

As technology evolves, choosing the right solutions becomes increasingly confusing, especially in areas like generative AI and cloud technology. Companies face two main challenges: selecting technologies without lock-in effects and avoiding constant migrations that drain resources. One effective approach is to build technology-agnostic applications, allowing easy switching of underlying models. Start with common models and gauge satisfaction before exploring newer options. Additionally, companies, especially smaller ones, should weigh the make-or-buy decision carefully, as in-house development not only requires building resources but also ongoing maintenance, which can become costly and hinder overall progress.

At bonprix, we work on the Google Cloud Platform, where the key factors in our decision included the seamless integration of data science applications into our data workflows.

Strategic Success: Choosing the Right Use Cases for Impactful Innovation

When selecting your first use cases, it’s essential to choose wisely. While the common advice is to focus on those with the highest business impact, it's equally important to consider the organizational circumstances. Opt for use cases that have a strong likelihood of success, including technologies that are reasonably mature.

Additionally, be mindful of the environment; avoid starting with use cases that face significant resistance from your team—whether due to the changes they bring or skepticism about their value. You don’t necessarily need to choose projects that attract a lot of attention from C-level executives, as this can increase pressure on the team navigating a new area of expertise.

While tackling exciting projects can yield substantial benefits for the organization, it’s easier to succeed if you’ve already built trust in the technology within your company. Gaining management support through the delivery of smaller, yet impressive, AI products can pave the way for tackling larger initiatives later on.