If you have a smartphone loaded with apps, you’ve probably observed that those apps can have wildly contrasting pricing schemes. Some products are free but stuffed with ads, while others charge a flat fee. Still others offer in-app purchases and upgrades. How would an entrepreneur building out an app decide which of these approaches to pursue?
That’s the sort of question that Ron Tidhar, a graduate student in Stanford’s Department of Management Science & Engineering who is studying with STVP, is trying to answer in his research. “My interest is very much on how executives choose these business models and then develop them in a new market,” he says. In particular, he’s interested in revenue models—the money-making component of a company’s business model. He’s written papers seeking to understand when and why different ways of making money succeed or fail.
In comments edited and condensed below, Tidhar explains the pros and cons of different approaches to revenue. In the process, he touches on what Spotify got right, and compares two seemingly similar online clothing retailers that turn out to have radically different revenue models.
Why did you decide to focus on revenue models in your research?
RON TIDHAR: Revenue models are a new source of uncertainty for businesses. There’s greater flexibility nowadays, especially with digital products, in terms of which revenue model to choose. It’s so much cheaper to set up, sell and distribute software. And firms are able to delay that revenue decision while they’re subsidized by VC funding and say, “We’ll worry about how we make money down the line.” That ends up causing a lot of trouble for many firms.
We realized that there wasn’t a good answer in the literature to the question of which revenue model to choose when. That made it an interesting stream to go down.
You’ve done a couple of internships at Spotify. How did that shape your research?
RT: One of the things that people often talked about with Spotify was how their freemium revenue model was a source of their advantage and a reason why they could potentially be competitive with Pandora or Apple Music. I was intrigued by the why — why is that the case? Why did they choose that model when their rivals did not?
In our study, we actually identified two different kinds of freemium models, which is a new insight in the literature. We call them the bundled freemium model and the fragmented freemium model. Essentially, the difference is how many upsells the firm offers to its customers.
“We see the fragmented model quite often in gaming, things like Candy Crush.”
In a bundled model, there’s one or maybe two premium upsells. For example, Spotify has a free version and then a premium version, which includes offline music-playing and increased functionality. There’s just that one upsell. Imagine a Spotify where you had to pay separately for offline use, for playing specific songs, for receiving recommendations, and so on. Because those features are interrelated, it doesn’t make sense to break them up and sell them separately. It makes more sense to bundle them together.
We see the fragmented model quite often in gaming, things like Candy Crush. There’s a single free version with many different kinds of upsells: an extra life, extra points or other consumable purchases. A fragmented model works well when you have modular and consumable upsells.
One contribution of this research is to separate the freemium models and say, “There’s more nuance here in how these might be structured.”
What are some perils of the freemium approach?
RT: A classic issue with the freemium model is cannibalization. If you make the free version too good, then not enough of your users will be willing to pay in order to get the upgrade.
Then there’s the complementary problem of offering too little on the free product. People aren’t able to experience the value or don’t understand what the product is good for, so they’re not likely to pay to upgrade to the premium version. Figuring out that boundary is a really tricky problem that I think many executives often get wrong.
What types of companies can get away with a subscription model?
RT: Companies have to have really clear quality signals, so the customer knows that what they’re paying for is valuable, and valuable in a recurring way.
For example, Netflix is able to demonstrate continuing value because it’s always releasing new content, refreshing its catalog, and learning the preferences of its customers.
The customer is induced to pay the first time because of those quality signals, and is then willing to stay on because they know there’s going to be new content and better recommendations every month.
And what have you observed about the ad-based model?
RT: We realized that the advertising model is part of a broader revenue model which we call the third-party model. That’s one in which there is a third party that’s willing to pay. In many cases that is an advertiser, but not always.
The third-party model works well when the focal firm gets some kind of insight about its users that might be difficult for a third party to get otherwise.
Putting aside recent controversies, I think Google and Facebook demonstrate the vast monetary value that this model drives when built on the right product. The crux of this model is the combination of unique insights into their users (Internet search queries and social networks, respectively, which both reveal user preferences) in order to better target online advertising. I don’t think anyone quite anticipated just how profitable this business model would be.
“There’s been some interesting research about whether we should consider data as labor…”
Interesting counterexamples are Twitter and Snapchat. While they both command very large user bases, they don’t generate the same insights about their users as Facebook and Google. This is because Twitter is based on very short 280-character public messages often broadcast to loose connections, and Snapchat is based on disappearing photos and videos that — though funny — are often trivial or meaningless. (Both contrast with Facebook’s close friend networks and timelines of personal life events.) This can help explain why Twitter and Snapchat have struggled with their revenues when compared to Facebook and Google.
There’s an interesting change in consumer psyche recently around how we should be viewing this ad-based model. That’s because the consumer is actually paying for the service — they’re just not paying for it with money, they’re paying for it with access to their data. There’s been some interesting research about whether we should consider data as labor, suggesting consumers should be paid for work they’re doing generating data.
How might future research projects shed more light on this question of revenue models?
RT: I’m working on a follow-up study that’s in the preliminary stages at the moment. We realized that different firms, even in the same industry, can have very similar offerings to their customers, but because some part of their business model might be different, they end up looking like really different businesses altogether.
A great example is in the online retail space. Stitch Fix, an online retailer based in San Francisco, sells clothing by having users fill out a survey about their clothing preferences. Then they send that customer five items of clothing in the mail based on their sign-up survey answers. The customer will keep and buy whatever items they want and send back, at no cost to them, whatever items they don’t want.
Rent the Runway is also an online retailer, but they’re focused on renting clothes online. A user can scroll the website, pick a designer dress, rent it at a fraction of the retail price, wear it for a special occasion, and then send it back to Rent the Runway.
“The small, seemingly innocuous decision of how they want to make money from their products … has led to different businesses later on.”
Both of these businesses are essentially trying to deliver clothes to online customers. Neither of them focused on physical store locations. But because Rent the Runway has chosen a rental model, it means that 100 percent of the clothes that they send out will come back to them. Whereas Stitch Fix is trying to sell clothing; they’d like to have the majority of the five items that they send out actually be bought by the customer and not returned to them.
So the businesses they’ve developed turn out to be strikingly different. Rent the Runway has become the biggest dry cleaner in North America because they have to get really good at receiving clothes, cleaning them and then sending them back out again for the next renter. Stitch Fix has ended up being a master at retail data. They have gotten really good at predicting what clothes a customer will want to wear. Stitch Fix has hired from Netflix and elite data firms, whereas Rent the Runway has become a logistics business.
The small, seemingly innocuous decision of how they want to make money from their products — their revenue model — has led to different businesses later on. The question for this study, through a multiple-case approach, is: What does that early revenue model decision mean for what the business looks like later? And how can executives actually figure out what they have to get good at, given their early business model decision?
Is there an ideal time in the life of a startup to make a final decision about the business model?
RT: I think the key thing is to treat business model development as an experimental process. It’s very unlikely that an entrepreneur can precisely predict what the fully formed business will look like. Therefore, ensuring that entrepreneurs update their beliefs about what will work and what won’t as they go is a key part of the process.
Counterintuitively, some research from our group shows that entrepreneurs shouldn’t settle on a business model too quickly. The rationale is that optimizing (or over-optimizing) too soon leaves an early-stage business susceptible to market changes. It becomes more challenging to roll back decisions or adapt the business model to new conditions.