Alexey Pisarevsky has more than 6 years experience in the mobile industry. A marketer with a mathematical education, Alexy graduated from the faculty of applied mathematics in MIPT, Russia. From the beginning of 2013 he became the CEO of Mobio, a full-service mobile app marketing platform. In this post he covers how to efficiently buy mobile traffic, how to analyse its quality and more. You can download the full ebook here.
This is the first part of the article about mobile traffic buying and optimization based on the eBook of the same name. In this part we talk about how to set up tracking system properly and work with relevant indicators.
You need to assess the quality of traffic when you get it to your app. We hope you already trace everything around the app with AppsFlyer, MobileAppTracking, Adjust, ADX or other system. But simply embedding the tracking codes is not enough. You must set it up properly and register changes to the important indicators.
- Create the events funnel
Numerous advertisers assess the quality of the traffic only by conversions to the ultimate event (purchase etc). This is wrong. In fact, the majority of purchases occur quite late into the campaign and user’s lifetime, it takes time for a user to purchase, so you need to set up a conversion events funnel, register each conversion and crunch the numbers.
For example, in case of a mobile game the events can be:
1) getting through the Tutorial
2) buffing up to level 10 and then
3) making the first payment.
For a flight booking app the chain of events may include the first search, then the tap on a search result and then the payment.
- Pick a day each event in the funnel should be registered on
Let’s consider «buffing up to level 10» in a mobile game. It is important to understand that the later we analyze the traffic the higher the conversion rate will be. We can get a picture like the one below:
Here, we take a cohort of users who came on day zero and look at what happened to them further on. On the first day, only 15% of users have reached level 10, but throughout the following days more has followed. It is clear that analyzing this event/indicator on the second day makes little sense, it is too early, while on day 7 we have sufficient amount of data to draw conclusions from. Let’s take another event, getting through the Tutorial. The picture may look like this:
Tracking systems typically count all events, the period of time does not matter. Therefore, you need to refine the data to learn, say, the number of registrations on day 7. For example, you can export all events and installs data into a .csv and work it out leaving only the events that have less than 7 days between install time and event time.You can analyze the data gathered from this event on the second day, there is enough data. If we consider the “purchase” event in a flight booking app, it can occur months down the line, so checking it after a week into the campaign is too early. That is why it is very important to have “early” events in the funnel, those you can analyze as soon as possible, preferably on the second day. This approach lets you quickly cut off low quality traffic and leave only those channels that show the potential.
- Simultaneously compare conversions delivered by different channels
Let’s consider this example:
At first, it looks like source 1 is much better than source 2: it brings twice as much conversions into registrations. But if you consider only the registrations that occurred latest on day 7 after the install, the picture will be different:
It turns out that source 2 is better than source 1, it’s just that there wasn’t enough time for source 2 to show its efficiency. And this is why you need to take one and the same period of time when comparing channels.
- Count unique converting users and not the total amount of conversion events
A simple example:
It looks like source 1 is much better than source 2, but if you count unique users who have made deposits rather than the total amount of deposits, the picture may change:
That means that source 1 brought a user that paid a lot, he is a high-roller, but source 2 delivered great conversion rate and brought a number of paying users that will pay much more in the long run. That is why you want to count unique users and not the amounts.
- You must have enough data to assess a source
How many installs do you need to adequately assess conversion potential of a source? And what is the definition of “adequately” here? Mathematical statistics offers us the notion of “confidence interval”. This interval determines the limits of the measured value (conversion) with an acceptable probability level (e.g., 95%).
However, you can apply common sense without diving deep into the theory. The sense suggests that the higher the conversion rate the less installs you need to assess the traffic’s quality. For example, with 80% conversion into “Getting through the Tutorial”, 100 installs will let you know whether the traffic acquired is good or not.
If the conversion rate is 10%, you need at least 300-500 installs generated by the traffic source to assess its quality. And if the CR is 1-2% it takes some 1,000 installs, otherwise the random factor is too great. That is why the first conversion event in the funnel should have high CR potential! It lets you see the quality of traffic from the source with small amount of installs. This is what a funnel made with the recommendations given above in mind can look like:
- 2nd day retention
- 7th day deposit
- 30th day deposit
- Make sure you tally up all events correctly
Many advertisers and publishers have the events measured incorrectly by the tracking platform. Examples:
- “Purchase” event also tracks older purchase restorations (from other devices).
- “Completed rides” event also tracks rides canceled by the user (taxi app).
- “Registration” event also tracks authorizations of users already registered in the database.
Thats is why you should always get into finest details with the developer, set the tasks clearly and test the correctness of measurement. Otherwise you’ll get a distorted picture and your traffic quality assessments will be wrong.
- Take into account the type of traffic when assessing the source quality
Types of traffic differ, and so does its quality. For example, organic traffic usually converts better than paid, and Facebook campaigns can yield the top-quality users due to the targeting capabilities of the network. Keep this in mind when assessing quality of traffic from a source, otherwise you risk screening out decent sources while comparing, say, paid and organic traffic.
- Do not forget about LTV when calculating ROI
Remember that users “live a long life”. High-quality hardcore mobile games or, for example, travel apps can keep them interested for over a year. In such a case, 10% ROI for the first month is a very good result.
Thanks to Alexey for the amazing insights into mobile traffic buying and optimization. You can download the full ebook here.