“There are plenty of fish in the sea…” To a modern dater, this old adage about finding love seems almost eerie in its prescience of the emergence of online dating. With the rapid rise of Match.com, Tinder, Bumble, and more, it is unsurprising that recent estimates suggest that the proportion of the U.S. adult population using dating apps or websites has grown from 3% in 2008 to over 15% today .
One such app, Hinge, launched in 2012. Its basic premise is to show a user some number of profiles for other suitable singles. If a Hinge user spots someone of interest while browsing, he or she can reply to a particular element of that person’s profile to start a conversation  – much in the same way a user on Facebook can “like” and comment on another user’s newsfeed posts.
This model is not a massive departure from the formulas used by older competitors like OkCupid and Tinder. However, Hinge differentiates itself with the pitch that it is the best of all the platforms in creating online matches that translate to quality relationships offline. “3 out of 4 first dates from Hinge lead to seconds dates,” touts their website .
One way that Hinge purports to offer better matches is by deploying AI and machine learning techniques to continuously optimize its algorithms that show users the highest-potential profiles.
Hinge first tests the AI waters
Hinge’s first public foray into machine learning was its “Most Compatible” feature, launched 2017.
The Hinge CEO shared that this feature was inspired by the classic Gale-Shapley matching algorithm, also known as the stable marriage algorithm . Gale-Shapley is most famously used for matching medical residents to hospitals by assessing which set of pairings would lead to ‘stability’ – i.e., which configuration would lead to no resident/hospital pair willingly switching from the optimal partners they are each assigned .
At Hinge, the ‘Most Compatible’ model looks at a user’s past behavior on the platform to guess with which profiles he or she would be most likely to interact. Using this revealed preference data, the algorithm then determines in an iterative fashion which pairings of users would lead to the highest-quality ‘stable’ matches. In this way, machine learning is helping Hinge solve the complex problem of which profile to display most prominently when a user opens the app.
Hinge creates valuable teaching data using ‘We Met’
In 2018, Hinge launched another feature called ‘We Met,’ in which matched users are prompted to answer a brief private survey on whether the pair actually met up offline, and what the quality of the offline connection was.
This was a simple, but powerfully important, step for Hinge. In addition to allowing Hinge to better track its matchmaking success, it can also use this data as feedback to teach its matching algorithms what truly predicts successful matches offline over time. “‘We Met’ is actually focused on quantifying real world dating successes in Hinge, not in-app engagement,” writes an analyst from TechCrunch . “Longer term, [this feature] could help to establish Hinge as place that’s for people who want relationships, not just serial dates or hookups.”
Recommendations and actions
In the context of increasing competitive intensity in the market, Hinge must continue to do three things to continue its successful momentum with AI:
- Increase ‘depth’ of its dataset: Invest in advertising to continue to add users to the platform. More users means more options for singles, but also better data for the machine to learn from over time.
- Increase ‘width’ of its dataset: Capture more information about each user’s preferences and behaviors on a micro level, to improve specificity and reliability of matching.
- Increase its iteration cycles and feedback loops (e.g., through ‘We Met’): Ensure algorithms are truly delivering the objective: quality offline relationships for users.
Outstanding questions as Hinge looks ahead
In the near term, is machine learning truly a sustainable competitive advantage for Hinge? It is not yet clear whether Hinge is the best-positioned dating app to win with AI-enhanced algorithms. In fact, other dating apps like Tinder boast much larger user bases, and therefore much more data for an algorithm to absorb.
In the long term, should Hinge be worried that it may stunt its own growth by improving its matching protocols and tools? In other words, if the implementation of machine learning increases the number of stable matches created and leads to happy couples leaving the platform, will Hinge lose the user growth that makes it so compelling to its investors?
(Word Count: 782)
 Aaron Smith and Monica Anderson, “5 facts about online dating,” Pew Research Center [website], February 2016, http://www.pewresearch.org/fact-tank/2016/02/29/5-facts-about-online-dating/, accessed November 2018.
 Hinge’s FAQ section, https://hingeapp.zendesk.com/hc/en-us, accessed November 2018.
 Dale Markowitz, “Hinge’s CEO says a good dating app relies on vulnerability, not algorithms,” The Washington Post, September 2017, https://www.washingtonpost.com/news/soloish/wp/2017/09/29/hinges-ceo-says-vulnerability-not-an-algorithm-is-the-key-to-a-good-dating-app/?noredirect=on&utm_term=.98040cd2701d, accessed November 2018.
 Diamond Siu, “Hinge dating app will begin using machine learning to make better matches,” Mashable [blog], July 13, 2018, https://mashable.com/article/hinge-facebook-dating-machine-learning/#QFvMlV4cGOqy, accessed November 2018.
 Muriel Niederle. “Matching”, February 11, 2007, The New Palgrave Dictionary of Economics, 2nd edition, Palgrave Macmillan, https://web.stanford.edu/~niederle/Palgrave%20Matching.Approved.pdf, accessed November 2018.
 Sarah Perez, “Hinge is first dating app to actually measure real world success,” TechCrunch [blog], October 2018, https://tcrn.ch/2yD03Ji, accessed November 2018.