Currently, around six million apps in the Apple and Google stores vie for the attention of users. According to AppAnnie, more than $ 86 billion was transacted globally in 2017 – in Apple and Google app stores, as well as third-party Android stores. That’s just over 74 billion euros.
In Germany, users currently have an average of 90 apps installed on their smartphones. But everyone knows from their own experience: just a third of it is used regularly. For app operators, therefore, the visibility of their apps in the app stores is the big goal, which is why millions are put into advertising. To grab users’ attention, something else is worth a lot more: a regular data analysis.
Look good and give a little more gas at the beginning
Winning users for apps requires a smart combination of marketing, app design, and data analysis. In the app stores the star rating, user reviews as well as selected texts and screenshots play an important role. This directly affects the conversion rate, that is the number of downloads per user that sees the app page. Furthermore, the ranking of the app is important, because a good ranking ensures that more users find the app by itself, so organic. Special Boost or Burst campaigns can increase the number of visitors and downloads in the short term, and thus also for a better ranking and more organic users in the long term.
Find users based on data
Both in boost campaigns and in “normal” performance marketing campaigns, mobile users are shown promoting apps, either while browsing the web or within other apps. For example, on the publisher side, ad impressions cost one euro for one thousand impressions (CPM) depending on placement. App manufacturers, on the other hand, prefer to pay in a performance-based manner, ie only when the download has been completed, for example at EUR 10 per download. If the download is not guaranteed, there is a risk of having to pay for commercials, but may not get any downloads and therefore no compensation.
Therefore, you have to think carefully, which user you want to display which advertising. “Historical data and machine learning are used here,” explains Dr. med. Steffen Wachenfeld, Chief Product Officer at WeQ, a start-up specializing in user acquisition for apps. “One can let computers learn, for example, on which operating system versions certain games convert well or at which times a shopping app is downloaded in contrast to a taxi app.”
Test display optics
When it’s clear which users want to recommend a specific app when, it’s important to make the ad itself, the creative, as appealing as possible. Creative Optimization looks at what a creative – whether banner, video, or playable ad – should look like for the click-through rate (CTR) and conversion rate (CR) to be particularly high. Even with a simple banner, there are many variables that can be tested against each other: the colors, the position of text and picture elements, the key visuals, the call-to-action etc.
Which combinations work differs from target group to target group. In addition to classical sociodemography, holidays and events such as the World Cup or the affinity of the user have an influence on the advertising success.
Coping with the flood of data
Once the ad is online, a typical performance marketing company in the mobile app world sees thousands of data points every second. This data must first be structured and preprocessed, says Wachenfeld. For this they are sorted, subjected to a statistical analysis and enriched with additional data. Basic is, for example, the order of IP addresses and based on an assignment of location information and an evaluation of frequencies of specific geographical areas. If you want to continue doing it, you can assign additional information to this geodata depending on the time, such as weather data. Then, for example, a taxi app can be advertised exactly where it is raining and there is an increased demand.
Differentiate important from unimportant
Time, holiday, banner color, district, browser version or display size – the characteristics of users and the ability to combine different-looking advertising at different times are endless. Artificial intelligence can also learn from historical data and help find the right properties, says Wachenfeld, WeQ’s AI expert. Special algorithms make relevant features more meaningful over time, making them more and more important in advertising decisions. Non-relevant features allow the algorithm to die slowly, making it easier and faster to make decisions based on fewer properties, saving space, time, and money.
What is normal?
Attention: If the previously observed patterns suddenly change completely – for example, if suddenly twice or only half as many downloads of a weather app occur – should be analyzed immediately: What is going on? Campaign managers look up and may find that a horse race is taking place in London and everybody wants to know if they need the umbrella. Or – if the download rate drops – may just be a technical problem.
To automate this task with pattern recognition, it is important to clarify what is normal. To do that, look at the past and observe variances. If you notice too much deviation from the usual variances, the campaign managers are automatically notified by the system. For example, is it realistic for a user to download a larger app in less than a second? and to open? Here you have to investigate, if this was not a real user, explains Dr. med. Guard field. Even if after 1,000 started downloads and some elapsed time not a single user has opened the app, this is not normal. Then it is important to find out, for example, if the download link was broken or if the post-backs, ie the notifications about the installs, work correctly on the way from the app via the app manufacturer to the advertiser.
Monitoring the server
To monitor server landscapes, the company Dynatrace, which was founded in Austria, recently showed an interesting solution at its in-house exhibition in Barcelona. The company is engaged in application performance monitoring, ie the monitoring of IT systems. In order to demonstrate the complex interaction of different servers and systems, such as app servers and payment servers, Dynatrace visualized its IT landscape in an AR application. This is comparable with a 3D map, in which the nodes of each server are like small continents in a virtual landscape. For the technician, the visual representation makes it easier to see where a mistake has crept in and is paralyzing the network. In other words, where payment is currently not working, where ads can no longer be played or where downloads are currently not possible – all in real time and worldwide. If you only realize that in the US is currently doing little in terms of app downloads but not always a network error must be guilty – it can just be just night.