Mobile Attribution for IOS has profoundly changed since Apple’s ATT was announced last spring. It was just the beginning of nothing less than a disruptive change for marketers, advertisers and a whole bunch of players in the digital ecosystem. The biggest tech firms in the world are moving towards what what Google calls a “more privacy first web”, as the most used search engine in the world has just announced that in the next two years it will remove support for third party cookies.
Safari now gives the user the option to choose not to be tracked while navigating and the same goes on Firefox. As tracking becomes less and less available, data-driven decisions will inevitably become harder to take and marketing efforts — with efforts we mean “money and time spent” — could potentially be harder to measure.
This is especially true in the mobile marketing business, where the adoption rate of IOS 14.5 will force the industry to change completely and reinvent itself. Quick.
We put together some of the things that people in the mobile industry need to know and take care of in this day and age, what machines can do to help your business and why they will be so important in the upcoming months.
WHY ARE ATTRIBUTION MODELS SO IMPORTANT?
An Attribution Model is the strategy a marketer/analyst uses to understand which action is responsible for a conversion. For example, if you own an app and start promoting it via paid search, and you see that downloads skyrocket, you want to know what caused such an increase.
Understand what is working and what is not is the first step of a successful media plan. Digital Marketing actions are often a “trial and error” process, where different things are tested for some time before the decision maker can asses what is working and what is not.
Mobile Attribution in a nutshell
Imagine you have just developed an e-commerce app and you want to promote it so that you’ll increase the number of downloads.
Ok, your app is ready and you want users to download it. As soon as you start advertising on different channels you clearly see an increase in the number of downloads
A Mobile Attribution Model is what data analysts use to understand what is/are the reason/s behind the higher number of downloads that your app is receiving. The purpose of the model is to give credit to the different marketing efforts in order to highlight what is working and what is not; this way, you can optimise your investment.
The problem with IOS 14.5 is that attribution has become harder and Apple’s SKAdnetwork — we will cover this topic extensively in the next paragraphs — comes with different limitations.
Let’s first dive deep into the difference between the two main attribution models: Probabilistic and Deterministic.
We like to cover these topics also in our newsletter: you can subscribe here
Deterministic attribution: the winner takes it all
The deterministic attribution model is supposedly the most accurate way of tracking and measuring your marketing action, as it uses different identifiers to keep track of the user behaviour. These identifiers can be GAID (Google Advertising ID), Apple’s IDFA (Identifier for advertiser) or other custom identifiers that keep track of the user behaviour.
In statistics, a deterministic model basically means that you know all the data beforehand. There is no “randomness” involved. You know — for example — that a user has downloaded your app and you can exactly track her journey and pinpoint the reason that led to that action. Hence, the solution is binary: 1 or 0, 100% o 0%, full or empty.
The user clicks on an ad and downloads the app: the ad takes all the credit for the download. This model is considered the “gold standard” for analytics, because it implies that you know the ID of the device that has clicked on the ad and whether or not that user has installed the app.
This model is as good as it sounds, but taking “randomness” out of the picture is impossible in the IOS 14.5 era.
ATT (Apple Tracking Transparency) allows the user to choose not to be tracked, therefore invalidating the IDFA that we mentioned before in the paragraph. This means that when a user opts-out of tracking, is not possible to monitor her actions outside of the app.
Not knowing the full user journey simply means that there will be “randomness” involved. But randomness is not the right term: probabilistic attribution, as you’ll see, works on data.
Probabilistic Attribution Model: how does it work?
While the deterministic doesn’t allow for “randomness”, as it is aware of the entire user journey, the probabilistic is actually looking for clues and follows a set of probabilities that the install or any other action was driven by one or more campaign.
This means that the credit for a given action will be distributed across different touchpoint, given that the analyst doesn’t have the device ID and therefore is not in possession of enough data to use the deterministic.
The Probabilistic Model is completely anonymous and respectful of the user’ privacy. This kind of attribution is based on probabilities and not id matching. For more on this, you can check out this article.
What we want to highlight is that the industry needs to move towards a probabilistic model in order to preserve the user’s privacy, which is something that every marketer should strive for.
If you want to know more about how we respect privacy, you can check out this article.
In fact, after ATT, the only deterministic model that will be allowed and that respects the user privacy is Apple’s SKAdNetwork, as complex and limited as it is.
SKADNetwork is more important that it used to
SKADNetwork is Apple’s privacy-friendly way to attribute impressions and clicks to app installs on iOS apps. It shares conversion data with advertisers without revealing any user-level or device-level data. This means that advertisers are getting a smaller, filtered fraction of the data.
Everything runs on apple proprieties: the attribution happens within the app store and then is verified on apple’s servers, where data are “cleaned” and then sent to the advertiser. The goal for apple here is to preserve sensitive data, such as the person identity or the IDFA, if the user has decided not to share it. The data are processed in order to guarantee the user’s privacy,
SKAdNetwork attributes conversions in a complex way — you can read this article to know more — and the entire process is intricate and comes with some limitations, as apple wants it to be.
Those are some of the major ones:
- In-app events are harder to measure: it is harder for analysts to understand how the users behave within the app, making it difficult to calculate a crucial metric such as LTV
- Not all apps and websites support SKAdNetwork: some apps need to make changes in the source code to support SKAdNetwork.
- Campaign Results come with a 24–48 hours delay
The last two points are especially painful for advertisers. As we said before, marketing is a “Trial and error” process, which means that when you’re making an “effort” — money and time — you want to know if it’s working or not.
You may wonder what does it mean that results come with a major delay.
Let’s make an example.
Imagine that a user is playing a game, sees an ad on the game, clicks on it, goes to the store and downloads the app. After the app is installed, a 24 hour timer begins. This timer can restart if a defined event occurs — i.e. the user registers or make a purchase — or expires if no events occur. Then, data are sent to the advertisers within a timeframe that can last up to another 24 hours.
So, SKAdNetwork gives advertisers limited data — remember, apple “cleans” the data beforehand and not all the apps support SKAd — with a delay that ranges from 24 to 48 hours.
This looks as it sounds: less data, less room for optimisation. And Apple has made clear that there are no ways around, no “cheat code” is available.
But there is a solution for attribution in the mobile business; is called “predictiveness” and it involves machines.
Attribution after IOS 14.5: what can marketers do?
IOS updates typically have a high adoption rate, as described in this post. When it comes to ATT — allow tracking or not allowing — the picture is still unclear. According to a report by adexchanger, the “opt-in” rate on a daily basis is between 4%-13%, which means that the strong majority of the users is disallowing tracking. In addition to this, Apple’s communication strategy is focusing a lot towards privacy and “a more privacy first web” — by the way, they have just published a great ad about this topic, you can check it out here — .
This means one thing: is likely that the vast majority of the users won’t allow tracking.
Therefore, IDFA is not expected to be a reliable identifier anymore and app owners need to figure out new ways to interact and get to know their audience.
On one hand, SKAdNetwork has definitely become more important, and will increasingly be more necessary to owners and analysts to understand the way users interact with their products. But Apple’s very own model, as described in the previous paragraphs, comes with a set of limitations — again, because apple wants it to be limited — and we believe that machine learning will be of great help for marketers.
If you want, you can follow Mapendo on linkedin!
In the near future, machine learning will become more important, as it is a mean to predict the success of a marketing effort.
Algorithms are created to do things a human being couldn’t do even with an infinite amount of time available. Artificial intelligence helps us understand trends, user behaviours and journeys in a way that is respectful to the user.
In the iOS 14 reality, with limited data available, machine learning can give analysts and marketers a complete view of how their efforts are performing.
.png)
