Better shopping experience thanks to artificial intelligence

Finally, the first own apartment! And even a small budget for the furnishings. But how do you find exactly the chair that suits your personal taste in the huge selection in the online store? This is exactly the situation that inspired Luca Indermühle and Ramon Herzig, two prospective diploma students at the iCompetence study program at the University of Applied Sciences FHNWat the end of 2017 for the topic of her bachelor's thesis. The ambitious goal: to develop a machine learning algorithm that sorts a web store according to one's own tastes - and to do so without collecting historical data about the users.

"It was important to us not to simply invest the time and energy we put into our final thesis in an external project that might then disappear in a drawer," says Indermühle, explaining the motivation behind their self-initiated project, which ultimately aims to ensure a better shopping experience. The two budding computer scientists first developed a survey tool to obtain as much data as possible. This was crucial to the success of the project.

From thesis...

In just six months and countless hours of work, the first prototype was finally created: an algorithm that independently learns which optical features are relevant for an object. Their supervisor Martin Melchior, Professor of Data Science at the FHNW, was impressed by their work ethic: "The two of them delivered an excellent bachelor's thesis, showed above-average motivation and heart and soul - in a field that was still new to them."

The positive feedback encouraged Indermühle and Herzig to continue pursuing the project after graduation - initially mainly in their spare time, alongside their 80% jobs. "At first, we mainly developed the solution ourselves without looking for customers - that wasn't ideal," says Indermühle. "Actually, we learned exactly the opposite in our studies: not to work on a finished product in the basement, but to go out to customers early on to validate the solution," Herzig adds.

...to your own company

But at the beginning of 2021, the duo took the big step: they found their first customer, set up a limited liability company and finally quit their jobs to focus entirely on InnoFind to concentrate. Today, they are successful on the road. "Our studies at the FHNW University of Applied Sciences have given us a big head start here," says Indermühle. "The iCompetence degree program combines computer science with design and business topics. Setting up our own company was not completely new territory for us as a result." They received support in this step from their former lecturer for Internet and Management at the FHNW, Louis-Paul Wicki. As a startup mentor, he brings a lot of valuable experience to building a young company and is happy to share his knowledge and experience with his former students. "It always pleases me to see how our graduates successfully take off in practice," says Wicki.

"The algorithm learns by itself"

Today, InnoFind already has six customers. Meanwhile, their algorithm can assess much more than just chairs. At one of its customers, the home accessories specialist Trenddeko.ch for example, it helps users find exactly the right poster for their own wall out of 25,000. It's simple for users: If they like a poster, they get a good rating; if they don't like one at all, they get a negative rating. The more ratings someone distributes, the easier it is for the algorithm to make suitable suggestions. But no one has to laboriously assign keywords to the posters, such as "owl," "hand drawing" or "blue.

"The algorithm learns on its own which visual features are relevant for a categorization," explains Indermühle. Even if, for example, a new poster trend emerges - flying elephants, for example - the artificial intelligence recognizes and trains itself to recognize and group the appropriate features. This not only makes the shopping experience more pleasant and faster for customers, but also pays off for the store operators. This significantly increases the number of successful sales.

For a better shopping experience: tests also on the street

The next goal is to further develop the software in such a way that users do not have to actively evaluate products - but rather that the product selection continuously adapts without them noticing anything. Instead, their behavior will be evaluated live. What do users click on? What do they look at for longer? In this way, the selection in the online store will almost magically adapt to their own tastes - and thus ensure an even better shopping experience.

But this further development no longer happens in the quiet chamber. The two young entrepreneurs from Thun are networked in their city's startup scene and work closely with their customers. But they also rely on personal contact to perfect their product. You can regularly meet Ramon Herzig and Luca Indermühle at the Thun train station, where they show random passersby the current status of their software with a tablet and ask for honest feedback. They also learned this process at the FHNW. "It's fun - and people cooperate well. In the process, we learn a lot about how users behave, and we can optimize our solution so that it's well received by both customers and their users." And that is the best prerequisite for success.

Source: University of Applied Sciences Northwestern Switzerland - School of Engineering