Prisjakt is a price- and product comparison site with the goal to guide consumers into making smarter purchasing decisions. The company started as a hobby effort in the southern Swedish town of Ängelholm back 2002. Today Prisjakt has grown into a 230-people strong company dedicated to provide consumers with as good pre-purchase information as possible.
The user experience on Prisjakt is built around the idea of having all relevant information in one place. This means that if a consumer is looking to buy the latest iPhone, all the retailers selling that iPhone should be displayed in the same interface, “the product page”.
Creating relevant product pages requires intense indexing work where identical products from all reachable e-commerce sites have to be connected. Usually, this connection is done through industry product codes such as European article number (EAN). If a product lacked this type of code the Prisjakt team would previously sort them manually.
This is where Machine Learning (ML) comes into the story.
Scaling reach through ML
Through ML, the sorting of codeless products now happens automatically. This might seem like a small task, but let’s revisit the fundamental product offering of Prisjakt. The goal of the service is to display all the prices of a given product. Can you imagine the amount of data such an endeavour in the case of an iPhone might generate? Every day? Every hour? You get the idea.
A common challenge for machine learning solutions is to ensure sufficient amounts of data for models to train on. In the case of codeless products, there is usually a lot of data to be found on popular products (such as an iPhone). For less common products, let’s say a Game of thrones collector’s item, there are not that many past examples (i.e. not a lot of data) for the model to learn from.
In order to cater to eventual shortfallings of the machine learning model, which often related to there not being enough past examples for the model to learn from, the staff of Prisjakt still “own” the indexing and could manually change any connection made by the machine, or shut the feature off altogether. Keeping humans in the loop is essential to the success of the model and an important strategic decision by the Prisjakt team – both in terms of delivering product quality and protecting the agency of employees.
Empowering the team
Grzegorz Drozdowski works in the indexing team at Prisjakt’s Krakow office. He thinks the new ML solution enables his team to spend their days working in a more efficient way:
“The best part is that we got rid of some monotonous work, which ML does. We can now focus on things like the correctness of the data we present or having all available offers under the most popular products,” says Drozdowski.
While the idea of getting rid of monotonous tasks might indicate that working at Prisjakt used to be less innovative, Drozdowski is quick to highlight that coming up with new ideas and finding new ways to provide value is part of the Prisjakt culture.
“There has always been time for being innovative in our company”, he says.
Finding new ways to add transparency to the purchasing experience is built into the DNA of Prisjakt. As noted by Grzegorz Drozdowski, the Prisjakt team is excited to explore new ways of utilizing AI technologies for providing an even better customer experience in the future.