Ranking Workshop
This was a cross-functional workshop I had the opportunity to facilitate. The objective was to create a clear vision and shared understanding of Catawiki's ranking system, including defining essential ingredients and the problem's significance for users and the business.
Ranking is not a problem to be solved
It is the means to an end. So, what's the problem we're trying to solve with Ranking?
As Catawiki keeps growing, increasing objects in auction, and attracting more passionate enthusiasts; there's an opportunity to improve our effortless discovery, while maintaining the treasure hunt experience for our collectors.
On the other side, there's the business need to protect and maintain the brand perception as the home of special objects throughout a vast range of categories.
What elements can be ranked, and on what surfaces?
Anything we can have multiple at a single page can be ranked.
We have to consider all the use cases/Surfaces where these elements will appear to the users.
🧑🏻🍳 Ranking Ingredients (Features)
What can we take into account when ranking elements on any Catawiki Page?
User segments:
How can ranking serve each user's needs?
Predictive Machine Learning
& Personalisation
To be able to match users with objects (products), it was very important to deeply understand the role of predictive machine learning in how we order elements in any given page for any given user.
Then, we came up with a framework to help us put it into action
There are infinite ways to apply this framework.
And impossible to know what will be the sweet spot for every use case, every user and every surface. But hey! we have to start somewhere. We started formulating hypotheses and examples.
🚀 Boosters per user segment
After the workshop, we got our hands dirty and launched our first AB Test focusing on our base model + 1 booster
Target: All existing users
If we rank objects based on our base recommendation model and we add a booster on the factor "Time remaining", more users will become bidders because we are showing them the most relevant objects that are ending soon.
What was our hypotheses?
Results
It turned out to be the most successful experiment of this kind ever run in Catawiki, so of course, we continued exploring with more and more hypotheses.