eBay / Marktplaats

A Personal Experience for Over Two Million Daily Visitors

Customer challenge: Increase relevance of Marktplaats homepage by offering personalization for two million unique daily visitors, with over eleven million advertised items.

Provided solution: Optimized homepage showing more relevant offers based on search. Creating recommender systems for clothes and cars categories.

Outcome: 3,5% increase in time spend on platform and increase in clicks, to both free and paid listed items.

Offering a personalized shopping experience to over two million unique website visitors per day isn’t easy, especially when you advertise approximately eleven million items. For Marktplaats, GoDataDriven has developed smart algorithms that use machine learning technology to ensure that buyers see the right ads in any channel.

These “personal recommendations” have resulted in a 3.5% increase in user time spent on the platform. Because of these positive results, Marktplaats has assigned a critical role for data to support change and has made sure that all product teams include data scientists.

From Collecting to Predicting

Marktplaats had already been using data to optimize its platform for some time, but it wanted to improve its user experience even more, as well as the overall effectiveness of its ads. They asked GoDataDriven to help them create a data science competency within their organization.

Determining Applications

To develop its data applications, Marktplaats created a separate, multidisciplinary team consisting of a product manager, data scientist, developer, marketer, and tester. The team began collaborating with GoDataDriven, defining applications and developing initial proof of concepts.

“The ideas for data applications are mostly the result of different internal sessions organized around themes like advertising and user metrics,” explained Rutger Mooy, senior data pioneer at Marktplaats and eBay Classifieds Group. “The data team also collaborated with other teams involved in overall customer experience so we could better understand the way both buyers and sellers used the platform.”

Developing Data-Driven Applications

GoDataDriven chose to start with small experiments and proof of concepts and then develop them further as they proved successful. One of its first successful proof of concepts was showing personalized ads based on a search.

For this use case, it was a major advantage that Marktplaats doesn’t exclusively process search traffic, but also owns the ad platform. Since Marktplaats had direct access to the ad content, developing the recommenders autonomously was a lot easier to do.

Making Smart Use of Screen Space

Three-quarters of all traffic on Marktplaats comes from mobile devices. It’s important to make smart use of the limited screen space on mobile. Showing optimized content based on available user data has led to more relevant offers. Marktplaats’ optimized homepage (created by GoDataDriven), which shows both paid and free ads, has led to a 3.5% increase in time spent on the platform. The number of (sometimes paid) clicks also went up.

Marktplaats Homefeed

The Power of Combining Data

The Marktplaats web analytics showed that the platform had two types of users: those who used the search box on the screen (search) and those who navigated using the menu (browse). Marktplaats started combining the results for both types of users to show relevant ads, and the results for shown ads increased significantly.

“The search feature offers limited profile information. By combining typed searches with browsing behavior from the menu, we created a much richer profile. We used this enriched profile to personalize the platform, which led us to develop our first application that benefited the website user, the advertiser, and Marktplaats,” explained Mooy. “The website user is presented with relevant ads, the advertiser gets a higher click-through percentage on his ads, and Marktplaats generates more ad revenue.”

Recognizing User Behavior

Identifying patterns and making the data as complete as possible was necessary to improve the recommendation system. For example, Marktplaats can see a clear difference between the search behavior of male and female users. By comparing usernames to databases of typical boys’ and girls’ names, it correctly estimated the gender of 47% of its users.

Marktplaats Targeted Email

“The analysis of search behavior provided some interesting insights. For example, people who sold a particular size of children’s clothing often also purchased children’s clothing that was two sizes larger. With growing children, this seems obvious, but it’s good to see this trend reflected in the data. There also turns out to be a significant correlation between, for instance, coin collectors and metal detectors, “ Rutger Mooy explained.

correlation in data - Marktplaats

To help users make a selection from the over 180,000 cars advertised on the platform, Marktplaats introduced a collaborative filtering system that recognizes which cars strongly relate based on search behavior. Using this data, Marktplaats now emails relevant car ads to potential buyers. It has received excellent feedback on these emails from users.

Marktplaats - Recommenders

Marktplaats’Data-Driven Future

Marktplaats offers millions of users a unique experience by recognizing and providing relevant product recommendations. The entire organization has embraced the data-driven approach, and it is actively expanding its product teams with more data scientists. Other eBay labels have created and introduced smart data applications combined with technological developments based on Marktplaats’s positive experiences. “For example, image recognition, will make it even easier for sellers to place ads,” said Mooy.

Technology

For Marktplaats in particular, GoDataDriven used technologies such as Hadoop, Spark, Kafka, R, Python, Java, and Scala.

Marktplaats’ optimized homepage (created by GoDataDriven) which shows both paid and free ads, has led to a 3.5% increase in time spent on the platform. – Rutger Mooy, Senior Data Pioneer, eBay Classifieds

Industry

Retail

Project type

Data Science Workflow
Machine Learning

Technologies used

Spark
Kafka
R
Python
Java
Scala

“Image recognition will make it even easier for sellers to place ads.”

Rutger Mooy Senior Data Pioneer
Explore more cases