Freemium is an increasingly popular online business model, where a user gets a core product or service for free but pays for advanced or premium features. Revenues for freemium sites are largely dependent on premium subscriptions, which are not easily won, making monetisation a major challenge. How do sites convert free users to subscribers and what is the value of premium features to users? Professor Jui Ramaprasad and her colleagues seek to unravel the mystery behind premium subscription adoption and also understand how premium features impact users in the context of online music and dating sites, with surprising results.
The freemium business model merges free models, which we see on commonly used sites such as Twitter and Facebook and are monetised by advertising, and subscription models, which we see on the Financial Times and many other news websites.1 On freemium sites, users can access many of the sites’ features for free. Often, these features are enough to give consumers a nearly complete user experience. However, certain “premium” features are only accessible if users pay a monthly subscription. For example, on Last.fm, an online music streaming site, users who access the site for free can stream songs from the entire catalogue available on the site, but the listening experience is interrupted by commercials after every third song. Those who pay the premium subscription of US$3.00 per month, however, are able to engage in an uninterrupted, advertisement free music listening experience.
Freemium is a prevalent online business model used by sites ranging from those that offer media such as music streaming sites to those that offer the chance at love – online dating sites. However, sites that employ this model often rely on premium subscribers for much of their revenues and have found that converting users to premium subscribers can be quite a challenge. As Bapna and Umyarov (2013) show, premium subscribers on Last.fm are 32 times more profitable than free users, but only make up 1% of the site’s users.
So a key question for managers of these sites is: what drives people to pay? And for users, the question is: are users better off with the premium features? That is, should they want to pay for a premium subscription to access these features? The research my colleagues and I are conducting looks at these two questions in two contexts: music and mating. In the context of music, we draw on prior work and look at the interrelationship between social engagement and premium subscription adoption. We know from Oestreicher-Singer and Zalmanson (2013) that climbing the ladder of social engagement, which includes making friends and engaging with them on the site, drives premium adoption, and we know from Bapna and Umyarov (2013) that premium users influence their non-premium friends to adopt subscriptions. In fact, a user has a 50% higher likelihood of purchasing a premium subscription if one of her friends does. In our latest research, Ravi Bapna, Akhmed Umyarov and I explore this relationship further, reversing the question to ask whether paying for premium features drives social engagement. If it does, we would see what we call a “virtuous cycle of social engagement”: social engagement drives subscription to premium; and then if paying for premium drives engagement, including making more friends who are both free and paying, we would see an increase in the impact of peer influence where premium subscribers influence their non-paying friends to convert. For managers of sites with freemium models, understanding this cycle demonstrates the importance of incentivising user engagement in order to convert users to subscribers
A second part of studying freemium models is understanding the value of the premium features. In particular, we look at one specific information technology (IT)-enabled feature that is part of many premium bundles on online dating sites − the feature that allows users to browse anonymously. While the default setting on many of these sites is non-anonymous browsing, where users leave a trail when they visit other users’ profiles, users who purchase a premium subscription are able to browse anonymously. That is, the premium user can view other users’ profiles without leaving a footprint on the profile he has visited. The question my co-authors and I ask is how this feature impacts behaviour and outcomes.
The Virtuous Cycle: Social Engagement and Premium Subscription
How many push-ups could you do if you bought an energy drink for US$2.99? Similarly, how many crossword puzzles could you solve? What if the price of the drink was discounted to US$1.49? Research conducted by Baba Shiv, Ziv Carmon, and Dan Ariely (2005) found that although the drink is the same in both cases, discounting its price sets a lower expectation for the person buying the drink and results in lower performance in both physical, cardiovascular activities and mental tasks such as solving puzzles. In the same vein, we ask: does paying (for a premium subscription) lead to higher expectation and higher “performance”? Theories of cognitive dissonance starting with Festinger (1957) suggest that payment may change behaviour because individuals want to balance inputs – their payment – with what they get out of it. Taken together, these studies suggest that payment for premium may impact behaviour, and in particular, increase engagement with the site.
In our study, we look at whether payment influences engagement in the context of the Last. fm online music listening site. Users of Last.fm can socially engage in many ways, for example, through making friends, “loving” tracks, and “shouting” on other users’ walls (similar to posting on a friend’s Facebook wall). The goal of our work is to see if those users who decide to pay for premium features increase their social engagement.
While the ideal research design to determine this causal relationship would be to run a field experiment, this is challenging to implement. Instead, we employ a quasi-experimental approach using propensity score matching (PSM) techniques. In this method, we create a treatment group and a control group as though we are conducting an experiment. We match a treatment group of 4,000 random users who have subscribed to premium in the five-month study period to a control group of 4,000 random users who have not adopted a subscription in the study period, but do adopt after the five-month period is over. We match these two groups on observable characteristics (such as gender and age) and observable behaviour (such as making friends, shouting and listening) to ensure that we are comparing two groups of users who are similar. Further, by matching users who subscribe to premium in the study period (the treatment group) with those who adopt in a future period (the control group), we account for time-invariant unobservable characteristics that drive premium subscription. Accounting for such unobservable characteristics has been a challenge of the PSM technique in the past.
When we compare the treatment and control groups, i.e. subscribers and non-subscribers (but future subscribers), we fi nd that payment does indeed drive social engagement: premium users make 59% more friends, do 21% percent more listening, love 80% more songs, and create 5.5 times the number of playlists compared to non-premium users. This data, combined with the previous research described above, shows the existence of a virtuous cycle between social engagement and premium subscription. That is, increasing social engagement is crucial to freemium sites as it kicks off a cycle that drives premium subscription, which leads to more social engagement and importantly, making more friends, which through peer infl uence converts more users to premium, and so on as the cycle continues
Digging Deeper into Premium Features
We know that a small percentage of users pay for premium subscriptions, and that converting users from free to fee is a key driver of revenues for sites. But what do we really know about the premium features themselves? Do they create value for those who pay for them? This is the question that motivated Ravi Bapna, Akhmed Umyarov, Galit Shmueli and I to look more deeply at premium features, starting by focussing on one premium feature – anonymous browsing – in the premium bundle of a large North American online dating site, which we shall call monCherie.com.
Why online dating? There is an abundance of dating sites online, and many of them, ranging from SeniorPeopleMeet.com to the well-known Match. com, employ the freemium model. Online dating is a growing industry. A 2011 Financial Times article reported that approximately 46% of the single population in the United States uses online dating to find a partner. In India, six million people used online dating sites as of 2012, and this number is expected to grow to 115 million people by 2015.
The premium features that these sites offer include enhanced search, browsing without advertisements, message read receipts and anonymous browsing. We focus on anonymous browsing to start our analysis.
Premium users make 59% more friends, do 21% percent more listening, love 80% more songs, and create 5.5 times the number of playlists compared to non-premium users. This data, shows the existence of a virtuous cycle between social engagement and premium subscription. That is, increasing social engagement is crucial to freemium sites as it kicks off a cycle that drives premium subscription, which leads to more social engagement and importantly, making more friends, which through peer influence converts more users to premium, and so on as the cycle continues.
Anonymous Browsing
Think about your daily morning routine of logging into your Facebook account, browsing your newsfeed, following links from one friend’s post to another acquaintance’s pictures, and so on. Now imagine if Facebook changed its settings such that your browsing behaviour was visible to those whose profiles or photos you were looking at. How would your behaviour change? With that in mind, we look at how the anonymity feature on online dating sites, specifically the ability to browse other users’ profiles without leaving a trail, impacts behaviour. This is a distinctively IT-enabled feature; consider for a moment that you are at a pub, a party or any other lace where you might meet a potential partner and are “browsing” the crowd: are you able to turn on a switch that allows you to be anonymous?
For our study, we partner with a large North American online dating site, which we call monCherie. com, and conduct a large-scale randomised field experiment. We take a random sample of 100,000 new users in one geographic area and endow 50,000 of them with the ability to browse anonymously (note that the default setting on the site is non-anonymous browsing). This is our treatment group, with the remaining 50,000 users forming the control group. We collect data on these users’ behaviour on the site for three months − the pre-treatment month, the treatment month and the post-treatment month. We observe the behaviour of each user in relation to incoming views (other users’ views of the focal user), outgoing views (the focal user’s views of other profiles), incoming messages, outgoing messages, and a measure of a match that we construct using the messaging behaviour. Based on discussions with senior management of the online dating site and Weick’s (1979) concept of a “double interact” (the idea that the exchange of three messages is part of the sense-making process that people use when they organise in a variety of contexts), we use an exchange of three messages between two users as an accurate signal of a match. The online dating site itself uses this as a signal of the arrangement of an offline date and given that they know the content of the messages and train their recommendation engines using this measure, we consider it to be a robust measure;however, we also consider longer series of message exchanges (our results do not change). We also have a variety of demographic characteristics of these users, including age, gender, attractiveness based on other users’ ratings, sexual orientation and subscription. We focus on the heterosexual population in this study, and, given gender asymmetries and established social norms in matching markets, we split our analysis by gender. These gender differences are apparent in our data, which show that women receive five times the number of views as compared to men and 10 times the number of messages.
We find that anonymity does breed disinhibition. In particular, we find that women in our treatment group – those users browsing anonymously – visit 12% more profiles and men in the treatment group view 10% more profiles than their counterparts in the control group. Turning to our key measure of success, a match, we see that this disinhibited viewing does not necessarily translate into more matches, which was initially surprising to us. Specifically, we see a 7% reduction in matches for men and a 14% reduction in the number of matches for women using the anonymity feature.
Comparing the treatment and control groups in the treatment month, we find that anonymity does breed disinhibition. In particular, we find that women in our treatment group – those users browsing anonymously – visit 12% more profiles and men in the treatment group view 10% more profiles than their counterparts in the control group. Turning to our key measure of success, a match, we see that this disinhibited viewing does not necessarily translate into more matches, which was initially surprising to us. Specifically, we see a 7% reduction in matches for men and a 14% reduction in the number of matches for women using the anonymity feature.
Why does this happen? To understand this, we break down the matching process. It turns out that matches for women are mostly what we call “received matches,” which means that they are initiated by an incoming message to the woman from the man. When we look at both received views and received messages, we see that both of these decrease significantly for both men and women under the anonymity setting. That is, all in-bound communication is being inhibited by anonymity – and this translates to a reduction of matches initiated by incoming messages. This impacts women harder because more than 75% of total matches for women are these “incoming matches,” consistent with traditional social norms where men typically initiate contact with women. With the anonymity feature, we take away the trigger, or what we call the weak signal that women leave for men since they do not initiate message exchanges. These weak signals are what often trigger the incoming message and therefore the incoming or received match. Without these signals, women lose a key mechanism that increases their matches.
While many of us cannot imagine that lack of anonymity would be beneficial in the context of other social networks such as Facebook or Last.fm, we see that in online dating, this feature – which many people desire – actually reduces the number of matches that a user gets, particularly in the case of women.
Conclusion
The freemium model is prevalent in the landscape of online business models today, but its premium subscription component is still somewhat of a mystery. The research I have described here is ongoing, but it is an initial step towards demystifying the process behind driving premium subscriptions and understanding the value of premium features.
Motivating social engagement is important in these communities, as it unlocks the door to the virtuous cycle between payment and engagement. And features that are perceived to be minor and desirable, such as anonymity, may actually have a significant impact on individuals and the value they get from the site.
