I didn't make it on Okcupid Why

Connection and Compatibility: This is how OkCupid uses analysis to help people find a partner

Today, about a third of Americans have used a dating app or site, and 12% have either been in a committed relationship or married someone they met through online dating, according to a recent report by Pew Research. Meeting the right person may seem like magic, but when you use a dating app or website, meeting the right person is a calculated process. Online dating has always been a data-driven, scientific, and effective way of connecting people who have common goals and interests.

There are a variety of online dating apps that have sprung up over the years that cover pretty much all interests, social groups, and affiliations. OkCupid has been around since the beginning, and today the use of business intelligence (BI) and product analysis tools is what makes the platform so successful.

Informed by the data, driven by the heart

Data is at the core of the mission here at OkCupid. Our obsession with data is why OkCupid makes over 4 million connections a week, over 200 million a year, 5 million contact referrals a day, and has more mentions in the wedding section of the New York Times than any other dating app.

I've been with OkCupid for three years and lead our data science team that takes care of platform analysis. Watching meaningful human connections develop is exciting, but it's rare to open a dating app and find love right away. The users have to stay for a while so that the app can learn their likes, dislikes, deal breakers and other information in order to find a suitable "match".

One of the main differentiators of OkCupid is the use of questions to create a "match score" that determines the compatibility of one person with another person. The more questions we ask, the more information we get and the better we can connect users with others. To do this, however, we need to understand the huge amounts of data that we receive.

Creating the perfect data infrastructure

The focus of the data analytics team is on understanding how the OkCupid platform works and what we can do to improve it. Our work ranges from traditional business intelligence (BI) reporting to algorithm development and optimization with a macro focus on user experience (UX) and product optimization.

The data infrastructure at OkCupid consists of mParticle, Looker and the Product Intelligence Platform (PI) Amplitude. mParticle collects and stores our customer data, which we send to Looker for general business reports and to Amplitude for deeper analysis of user behavior and our customer experience.

When my team first started using Amplitude, we got the idea that it was primarily intended for event tracking and segmentation. Over time, we learned that we can use it to measure engagement, identify user cohorts, analyze various user journeys and find leading indicators for conversion and retention. Amplitude is explicitly designed for this type of analysis. That meant we could access meaningful insights much faster.

BI and Amplitude: Better Together

Creating the most engaging and fun product possible requires a lot of A / B testing and data analysis. This allows us to determine which aspects of our product customers like and find ways to increase engagement. It doesn't matter if you're a high-intent user looking for a long-term relationship or a casual user looking for something casual. We need to understand who these different users are, how they engage on the platform and what behaviors and motivations lead them to stay loyal to the platform or to leave it over time.

Traditional BI tools like Looker, Tableau or Power BI can perform this analysis, but they require a lot of time to build data models to answer our product questions. They also have their limits when it comes to the depth of insights we can get from the data we have.

With Amplitude, we can make meaningful use of unstructured data and begin to understand our different users and their journeys in our product. On this basis, we can build more structured reporting, identify product experiences that customers find most valuable - and incorporate more of them into OkCupid.

With Amplitude, for example, we can identify and understand the various behaviors that suggest users are staying in the app for a long time. And for users who log in and then quickly exit the app, Amplitude provides us with user paths. We can analyze these to determine what happens most often before a user ends their session. As a result, we can figure out which aspects of OkCupid we should change - or remove altogether.

A traditional BI tool like Looker can access all information in our data warehouse and perform traditional aggregations and pivots very easily. But Amplitude shines when dealing with time series analysis and anything that isn't well structured.

A concrete example: The following question can easily be answered with a BI tool: “How many likes has a user received over time?” Amplitude offers additional value here, because it is important to understand what the user is talking about That led to these likes. Did a user get the Like button through a notification, navigation or other functions of the app? Which path did users go from there and what was their typical engagement pattern with the various functions? So instead of just knowing that a user liked 20 people today, we can start building a story about that user's experiences and preferences. Maybe you liked 20 people today and spent a lot of time texting each of them. This is different from someone who liked 20 people today but did so in quick succession.

The nuances in our clients' experiences are hard to see when dealing with aggregations. Looker builds on established data storage systems. So to answer a question like the one above, you'd have to create a custom report, merge multiple records, or even write SQL. When using Amplitude, the differences are easy to see when we have the user journey ahead of us.

Better teamwork and faster launches

Amplitude is mainly used by my data science team and our product teams. Both groups ask questions about user journeys and engagement, but they also need answers to different types of questions. For example, we have a team that takes care of our onboarding flow. This team takes care of exit points for new users. Another team is more focused on long-term retention. So they put much more emphasis on "sticky behaviors", those that ensure that people keep coming back to the site and thus increase the chance of success in love.

With Amplitude we can create, save, and distribute all of our various charts and dashboards across the organization. That means we don't have to duplicate the work; the teams regularly share results and make decisions based on the same data set. Even though we have a self-service approach to our data, it's a really collaborative process that saves us time and leads to more informed decisions.

With Amplitude, we can look at the structured data without spending additional development time creating new views. Whenever we start a new feature, we simply set up an event for it in mParticle and send it to Amplitude with the corresponding user and event properties. Previously, an analyst had to write manual queries in Python or SQL for precise data tracking within the platform. With Amplitude, we don't need the help of an analyst. We can watch the new events load in real time and evaluate them instantly in amplitude diagrams.

This powerful combination of teamwork and functionality ultimately means that we immediately understand whether a product bet is working or not and can iterate much faster than before.

A concrete direction in times of change

The widespread lockdowns have shifted the dating norms this year. Instead of lamenting the loss of traditional dating protocols, we had a new question to answer: How are people changing their usage behavior on our platform in order to adapt to a COVID-19 dating world?

First off, we found that users spend a lot more time delving into conversations. Personal encounters are no longer that easy, so you have to spend more time getting to know someone better within the app. Thanks to our powerful combination of BI and PI, we can quickly access specific data on these new patterns. We're creating even more opportunities for users to forge deeper virtual connections, filling some of the void that physical distancing has created for many people.

Finding love through data analysis might not seem romantic, but we know it works. Our mission at OkCupid has always been to bring love into the world. And with the right data - applied in the right way - we are helping people do just that.

Nick Aldershof

Nick Aldershof heads OkCupid's analysis team. Over the course of his career, he has focused on optimizing products with strong network effects to build healthy ecosystems and markets. He is passionate about using data science and analytics to optimize business processes and drive growth through new insights and business strategy improvements.

More from Nick