Does Device ID Coverage Matter for Deep Links?

At Branch, we designed our linking infrastructure to be the most sophisticated solution in the industry. Thanks to our 30,000+ integrated partner apps, the Branch Link Graph is the largest collection of cross-platform (web and app) user identities in the world — every month, we see a link graph for almost every smartphone on the planet. We are obsessed with chasing down new edge cases to solve, which means your users always get the best possible experience.

Legacy attribution providers often make statements that sound something like this: “We cover over 98% of device IDs in [some country], which makes our deep links better.” As the team behind the most robust linking network in the world, we would like to help you unpack this claim.

Here’s the answer: Device ID coverage alone means absolutely nothing for deep links.

Statements like these are more than just irrelevant — they’re irresponsible. At best, they indicate the companies making them have a fundamental misunderstanding of how deep links work. At worst, these companies are using large numbers to mislead you (and could even be trying to cover up major functionality gaps in the products they’re offering). Either way, you deserve to understand all the details before you end up stuck with an inferior system that damages your user experience.

In this blog post, we explain why device ID coverage is useless for deep linking. It’s almost like trying to clap with only one hand. We will also give you some recommendations on what to look for instead.

Deferred Deep Linking: A Quick Review

A normal deep link takes your app users directly to a specific place inside your mobile app. A deferred deep link can do this even when the user has not yet installed your app — context is preserved through the app installation process. Deferred deep linking allows you to build all sorts of powerful functionality, like automatic referral programs and customized onboardingbut only when it works reliably.

A reliable deep linking system requires two things:

    1. It must remember users who click links in a web browser (or in an app that opens links in a web browser, which is basically every app). Think of this as writing a name tag.
  1. It must recognize those users again when the app opens. This is like reading the name tag from earlier.

In other words, deep links build a bridge into your app from the outside world, so that your users can travel across. The technical term for this is “matching accuracy”, and being good at it is very important: any deep link system that sends Jack across a bridge that you built just for Sarah is worse than no deep linking at all.

Here’s the thing: device IDs (IDFA on iOS and GAID on Android) are only available on the app side of that bridge. Without a Link Graph to help you reference those device IDs on the web, you could have a database containing every single device ID in the world and it still wouldn’t help you model a single deep linked user.

When legacy attribution providers claim to cover 98% of device IDs, that number is utterly meaningless for deep linking because they all have 0% coverage on the web. They have no Link Graph. Matching accuracy requires data at both ends of the bridge.

Matching via Basic Probabilistic Modeling

Since these networks can’t use their device ID collections for deep linking, how do they match users? The answer is basic probabilistic modeling. Here’s how it works:

    1. When the user clicks a link, the network generates a web basic probabilistic model based on ephemeral device data including IP address, OS version, screen resolution, click time, and more.
    1. When the app opens for the first time after install, the network generates a corresponding app basic probabilistic model using the same data.
  1. If two basic probabilistic models match, the user is deep linked.

Here’s the problem: these basic probabilistic models are often so similar that 30-40% of them get paired to a completely different user (if you’re on shared wifi or a busy cell tower, the error rate can be over 50%). Basic probabilistic models also start decomposing as soon as they are taken, which means they become pretty much useless within an hour or two. Attribution networks typically claim to offer some sort of advanced basic probabilistic model filtering algorithm, but you can’t make gold out of dirt — the entire process is inherently inaccurate.

Before you know it, basic probabilistic model matching errors mean Jack is walking across that bridge you built for Sarah.

The Persona Alternative

There is one way to use device IDs for deep linking: tie them to persistent, anonymous identifiers on the web. In other words, build a cross-platform Link Graph of cookie/device ID pairs. This allows you to recognize users at both ends of the bridge, while still respecting their privacy, without relying solely on basic probabilistic models.

Branch has spent the last three years building exactly such a worldwide link graph. This is the same technology behind our cutting-edge Attribution Solution product, and makes us the only deep linking platform able to offer strong matches with 100% guaranteed accuracy, even when users haven’t installed your app yet. Since the Branch Link Graph is pooled between every app in the Branch network, it is constantly growing.

And yes, even the Branch network sometimes has to fall back on basic probabilistic model as a last resort. When this happens, we clearly indicate it so that you can elect not to perform deep linking if you choose. This means the worst case scenario for a Branch deep link still beats the best anyone else can offer.

The Numbers that Actually Matter

Next time someone tries to dazzle you with an outlandishly high “device ID coverage” number, claiming it means better deferred deep link accuracy, ask this question: “how big is your strong match pool? Or in plain English: “for first-time installs, how often can you guarantee my deep link model accuracy, and how often are you just guessing?”

Without strong matching, your user experience is left to luck — Jack could get his content… or he could get the content you intended to provide for Sarah. There’s no way to be sure, and device ID coverage tells you nothing. The Branch Link Graph allows us to power millions of strong matches all around the world, which no other deep link platform can offer.

Happy deep linking!