Facebook Pages

On the Facebook Pages Science team I lead a group of 12 machine learning PhDs to develop a set of core reconnection surfaces across the entire Facebook platform.

There were many challenges that had emerged at Facebook when I arrived in MPK in 2015. The platform had overgrown. It had become almost impossible to find and understand what was relevant to you based on your personal experience. I came in to design some of the first recommendation surfaces for the platform. These become some of the most important areas of the entire platform. We experienced a tremendous growth on all of the surfaces that the algorithms were implemented on. This was the first time that we began to correlate user intent across their journey through facebook. We looked at how a person might come from Newsfeed, move to Search, go to an event then navigate to the Page that created the Event. This yielded an incredible insight across a large population. The algorithms took weeks to generate the data and train on. They were exquisite, they had the perfect balance of consistent long term engagement without an artificial novelty fall off curve.

By the end of this period I had become the lead designer across Pages, Places, Local, and Events. I was working with 5 other designers and over 100 engineers. The stress was super intense, and as you can imagine the political climate at the time had become like a cauldron of malcontent. I completely melted down at some point, as my manager in that position had been quite the troll and largely unsupportive, which in my opinion is the largest contributing factor to burn out. Luckily I had a relatively good relationship with the director of design in that department, and she helped me navigate a team switch. That brought me to the Connectivity Lab.

By the end of my tenure on that team we had nearly doubled the number of active businesses on the Facebook platform to about 70 million active pages. The consortium of designers who were working on the recommendation algorithms was one of the most influence group of people on the platform at the time, and many of them designed the core products that were launched around that time including live and marketplace. My work primarily focused on interest based advancement of small and medium sized businesses on the platform. Because of that audience’s disproportionate influence that lead me to organizing content across many teams. Businesses accounted for less 10% of the total audience but 50% of the content created on groups and events once those platforms were opened to businesses. So I got the opportunity to oversee how those design surfaces interacted visually and functionally.

At the same time I was juggling all kinds of fun and interesting minor pieces of functionality. There was a point when Trump was amping up his campaign that people started to complain that they wanted to follow him on Facebook without liking him. This was the first time that people had not liked a figure they needed to follow for informative purposes on the platform. I had to design a solution to divide like and follow so that people could follow Trump without supporting him publicly on the platform. In retrospect this was a really interesting trend that started way before his election.

I was really fortunate to learn for the first time about deep user research in both a field and lab setting. I got to design interactive prototypes in Framer that demonstrated a lot of the functionality I was designing for the recommendation surfaces we were working on and do field research. The first place that I got to test designs in market was in India. We had a large group of people, maybe 12 people, who all went to Delhi to learn more about the market there. We also went to the city of Meerut, which immediately become a somewhat dramatic affair as Google maps had taken us down a dirt road which was largely unpassable. Eventually we met a group of men dressed in military fatigues and in an unmarked truck. For a solid twenty minutes they made our driver follow them and we were all largely certain we were being kidnapped as we live tweeted our demise, we eventually reached the road and realized it was just a group of police who were out in the fields drinking. One of the primary lessons from that trip were that maps, addresses, and social structures were largely independent and not very technically encoded. We reformed the entire facebook address structure to accommodate better directions.