Monetizing Mobile Machine Learning

The current state of play when it comes to monetization options and opportunities for mobile ML experiences

If you at all follow the economics of mobile platforms, then you’ve almost inevitably heard about (and likely saw) Epic’s recent viral campaign in response to Apple’s decision to ban the incredibly popular, free-to-play game Fortnite from the App Store.

You can read much more about what’s happening between Epic and Apple, but it boils down to this: The App Store has banned Fortnite for violations of its terms of service, specifically because Epic has been using its own in-game payment system. This makes some sense, given Fortnite’s status as an elite freemium gaming experience.

From where we stand, we don’t have a whole lot to offer on this specific debate. But it did get us thinking about the bigger picture of monetization. Specifically, monetization as it relates to capitalizing on mobile machine learning experiences.

While the mobile ML industry might not be mature enough to offer robust revenue projections or historical engagement metrics, it’s important to start thinking about the platforms for—and the opportunities and challenges of—monetizing mobile ML experiences. That’s what I’ll attempt to do in this blog post.

First, a review of three central platforms for deploying mobile ML experiences. iOS, Android, and (perhaps surprisingly) Snap.

Apple and iOS

Of course, the two most logical outlets for mobile machine learning experiences are on the two primary mobile platforms: iOS and Android.

We’ll start with Apple, since they’re at the center of the monetization debate right now. Like many other Apple systems, publishing and monetizing apps in the App Store is a bit of a closed loop (this closed loop, primarily, is the source of the dispute with Epic). Here are they key details:

  • App publishing fee: App publishing fees are baked into a user’s Apple Developer Program account. Currently, the fee for this program is $99 USD per year. More on what you get with that Developer Program here.
  • In-app purchase revenue sharing: For iOS apps that include chances for users to upgrade a subscription or access new features via additional payment, there are a couple things to keep in mind. The first surrounds revenue sharing. Apple takes 30% of that revenue. The second involves the payment mechanism. Developers must use Apple’s official In-App Purchase system.
  • Paid apps: Same rate applies here for apps that charge an upfront fee for an initial download. 70/30 split.
  • Subscription-based apps: For apps that are free-to-download but require a subscription (i.e. Pandora, Hulu, etc.) to access certain features, the App Store again requires the use of Apple’s payment system if subscriptions are purchased in-app. The App Store takes 30% of subscription fees in the first year, and then 15% for each subscriber that continues their subscription into year 2 and beyond.
  • Selling physical goods: For app’s that sell their goods directly to consumers (meaning the purchases are of external products and not in-app features), Apple does not share revenue or require the use of their payment system.

Android and the Google Play Store

The primary difference between iOS and Android seems to be that Android is a much more open ecosystem. Play Store and App Store policies are similar in many ways, including revenue share rates and the requirement to use an official in-app payment system. In fact, Fortnite also got kicked off the Play Store, though this news received much less fanfare and didn’t seem to factor into Epic’s strategy in picking this fight.

More specifically, Epic didn’t run an in-game ad or file a lawsuit against Google because on Android, Epic isn’t beholden to Apple as the lone distribution channel. Players can side-download and play Fortnite on Android’s more open ecosystem.

That said, not many app development teams have the product, brand recognition, or money in bank the to run their own independent app store like Epic does. As such, most developers and teams will have to rely on the Play Store for distribution and monetization.

  • App publishing fee: Unlike the the $100 yearly price tag for Apple, Google Play requires only a one-time payment of $25 USD to create a Google developer account.
  • In-app purchase revenue sharing: Google takes 30% of in-app purchase revenue. And similar to Apple, developers must use Google’s official in-app purchase system. It’s assumed that the Play Store is more lax about their policies in general, but draws a hard line on games.
  • Paid apps: Same rate applies here for apps that charge an upfront fee for an initial download. 70/30 split.
  • Subscription-based apps: Once again, same as Apple’s App Store. Google Play Store takes 30% of subscription fees in the first year, and then 15% for each subscriber that continues their subscription into year 2 and beyond.
  • Selling physical goods: For app’s that sell their goods directly to consumers (meaning the purchases are of external products and not in-app features), Google does not share revenue or require the use of their payment system.

Snapchat and Lens Studio

Mobile machine learning experiences are no longer limited to full, natively-developed apps. Snapchat, long a leader in advanced on-device computer vision, has added new functionality to its Lens development platform Lens Studio that allows the use and monetization of custom neural networks.

In 2015, the incredibly popular social content platform added Lenses to their mobile app — if you’ve ever played with Snapchat, you know these well. They’re essentially augmented reality (AR) filters that give you big strange teeth, turn your face into an alpaca, or trigger digital brand-based experiences.

Developers and creators have long been able to access Snap’s internal ML models when building with Lens Studio, but now, with their new framework SnapML, creators can implement custom ML models to unlock truly limitless possibilities for more immersive augmented reality systems.

Keep in mind that Snap Lenses aren’t full apps — they’re AR filters built with Lens Studio. While Lens Studio is a robust and technically-complex platform, developing, publishing, and monetizing Lenses is quite a different process.

Put simply, Snap Lenses are a much easier access point for mobile machine learning experiences. Full apps are harder to create, and building an audience for your app from scratch is difficult, to say the least.

  • Lens publishing fee: There is none. Lens Studio is a free-to-use development platform that’s integrated with Snapchat.
  • Lens revenue sharing: None. Again, quite a different experience and set of economic incentives. Snapchat serves as a creative ad platform, so they generate revenue from downstream ad buys. Lens creators (sometimes facilitated through Snap, sometimes independently) work directly with brands that want to advertise on Snapchat, so the cost doesn’t fall to creators.
  • Growing AR advertising market: In 2018, AR advertising revenue hit $453 million, and that number is only expected to grow. This is especially true with more people working, shopping, and staying at home. And combining AR with advanced machine learning models promises to make mobile mixed reality experiences (like those on Snapchat) more immersive, interactive, and unique.

Monetizing Mobile Machine Learning: The Possibilities

The truth is, in most cases, machine learning experiences perform the best and most consistently on iOS. The device landscape is less fractured (i.e. clear differences between iPhone generations), and developers have a clearer understanding of what kinds of AI-accelerated hardware can be accessed with models (still a bit murky in some cases, but it’s getting better). Building on Android is improving in this area, but it will always be difficult to build ML experiences that are optimized for such a large number of devices and chipsets.

Depending on how core-to-the-experience a machine learning model is, monetization will likely look quite different. Here are a few hypothetical possibilities for iOS and Android:

  • Create an experience that uses machine learning in the service of selling an external good or service — whether through visual search, virtual try-ons, or other new ways of showcasing products to consumers. While avoiding revenue sharing is a bonus, revenue sharing might just come in a different form (i.e. revenue sharing with partner brands)
  • Apps that provide ML-powered features as part of in-app upgrades can be successful, but the earning potential is certainly diminished by the 30% both platforms take off the top. This especially makes sense for creativity apps that include otherwise rich editing tools that can entice users to upgrade to access virtual green screens, portrait modes, artistic filters, and more.
  • In-app upgrade features: This is perhaps the most intuitive approach for many mobile ML experiences, and particularly for photo, video, and creativity tools. Gate new, ML-powered image and video editing capabilities behind a higher tier that requires an upgrade.
  • Link to external sale of goods/services: This is probably the most feature-specific method of monetizing mobile ML models. Consider visual search applications like Shnap that allow users to search a large catalogue of brands and products by providing an image of a product or pattern they like. By creating an independent, external marketplace, experiences like these can provide their app to users for free and monetize through providing cutting-edge on-device experiences to partner brands. There are certainly more markets like this to tap into, but visual search provides us a helpful conceptualization of this approach.
  • Paid apps for full ML-powered feature set: For those most confident in their mobile ML experiences, upfront, pay-to-download apps can signal a premium feature set. For particularly robust photo and video editors, this approach can make sense. Just make sure the experience backs up what you’re promising.
  • Free app with ML to build brand engagement: Of course, app development teams can also opt to keep their apps entirely free. ML can also help build engaging brand experiences and give users ways to share content organically on social channels. Things like user-generated content moderation allow apps built around social sharing to automatically screen incoming visual and text-based content. Other brands might allow users to access promotions through things like brand-based scavenger hunts. Or, as we also explored above, brands can now use platforms like Snapchat to create and distribute custom ML-powered experiences.

What We Still Need to Learn

The truth of the matter is, it’s tough right now to assign ROI, engagement, or direct revenue to machine learning-powered features. Much like in the early days of many other technological advancements, the paths to monetizing and building business value aren’t always entirely clear. Many of the most-used mobile ML experiences are baked into first-party apps and operating systems.

But as mobile developers and creators are equipped with increasingly-accessible platforms and tools for building and deploying machine learning models, we should see its use become more pervasive, and with that use, a lifting of the current fog around monetizing mobile ML experiences.

Avatar photo


Our team has been at the forefront of Artificial Intelligence and Machine Learning research for more than 15 years and we're using our collective intelligence to help others learn, understand and grow using these new technologies in ethical and sustainable ways.

Comments 0 Responses

Leave a Reply

Your email address will not be published. Required fields are marked *