Because of the ongoing changes in the competitive app publishing market, we decided in the early stages of the company to make a difference in some key points. The available professional SaaS products could not fulfill our expectations in different stages of our company history, so we decided to build a software stack to solve in-house issues ourselves. We have moved towards more advanced technologies in AppLike-history: What scales well stays in our tech stack.
Everybody within Applike supports the culture to not accept the given, to think out of the box and improve our business with incremental changes every day. This is reflected 100 percent in our software stack.
We started 4 years ago with our first in-house software – a completely unique and independent Mobile Attribution Tracking solution. Many competitors in the market were challenging this decision. Of course this step was motivated by saving cost, but in retrospect this piece of code was the foundation to put many things on top.
If you control the mobile app attribution and you have access to the underlying data, you can detect, investigate and prevent fraudulent marketing behaviors. If you know everything about your app users, the data science team can perform many cool analyses on the data, because we own it.
If you know what you pay for an attributed install effectively, your marketing team can do much better, efficient ROI predictions than others.
If you sell advertisements within your mobile apps, you should be really interested to show exactly those apps to users which are driving the highest revenue.
So – How could you be sure you’re doing all this and doing it well, if you haven’t implemented the logic behind these features, from attribution to monetization yourself?
Today, we feel we gained a significant competitive advantage by owning, building and adapting the Marketing Technology behind our business ourselves.
Our different inhouse developed software stacks are helping our company to succeed in various fields.
Let’s take a look:
All companies which are doing performance marketing for their own apps are using a mobile attribution technology provider, e.g. adjust, appsflyer and many others.
Those platforms help to distinguish the different marketing traffic sources for app downloads. For example they report if a new app download was generated by facebook advertisements or by any other website. It also supports some basic ROI overview of ad spends, i.e. how much you earned back for every dollar spent on user acquisition. All of them have fraud prevention on their feature list and they are integrated to ad-networks. All of them claim to be the best in various niches. We recognize they each have different strengths and weaknesses.
Our big concern was that we wanted to understand what is really going on behind the scenes. We wanted to have full transparency and full access to all data points.
If you want to build high towers, you have to spend good time on their foundation:
We first built a state of the art mobile attribution system for Android and iOs. Later we added SDKs for the different platforms (React Native, Cordova, Unity and so on).
As a next step, we needed deep integrations with ad-networks, e.g. we integrated with APIs to pull KPIs from the partners, and we set up tracking and creatives where the boring steps are automated.
With all data in place we started to implement a user friendly dashboard for our business teams. Setting up campaigns and monitoring their performance over time is mandatory for our marketing successes.
It’s a big challenge to maintain such a complex system, and there are new features requested regularly.
Our vision is to build a software platform which automates all marketing efforts. Why should a marketing manager focus on struggling through different data sources and many dashboards, if this could be solved by a machine and they could focus on big decisions and creative outcome?
Our marketing management suite is deeply connected to the major user acquisition networks. Within one dashboard many different campaigns can be controlled and monitored. This software is connected to our own attribution technology.
Our data science team has real time access to all data which relevant to our company. We have already collected more than 10TB of data. We implemented different technologies, which make possible to work with this data in real time and to have a acceptable access times. We host these solutions on AWS and with Kubernetes. We used Hadoop and S3 as good approaches to store data. Performing analyses, building descriptive and predictive models is really great in Spark on the selected data. On the other hand we transfer our MySQL-data into our Redshift-cluster, which offers easy, understandable and fast SQL-query access to our data, enabling BI to iterate on ad hoc analyses quickly. Kinesis and Kafka distribute all data redundantly and of course in real time. All of these technologies enable us to answer different questions from different departments coming up every day.
Everybody in data science is working for one big goal: There is no business question that can remain unanswered!
Our inhouse sales team sells and manages over 50.000 different campaigns for different advertising partners. This includes an interface to upload creatives for our partner campaigns, and interfaces to manage and to monitor them. This is an easy usable dashboard, built nicely in React.It is important that every manager can access all relevant campaign KPIs in a second. With many different data-points to monitor it is helpful to show them in a variety of pretty understandable graphs. Our account managers do not need to export any csv to gather information anymore, instead they get their answers in their custom-built web applications!
Built a data science based API with response times smaller than 20ms? We show advertisements to monetize the users in some of our apps. Personalizing the apps to the targeted user improves their experience and our monetization. The service includes different parameters into the decision which ad should be distributed. We analyse personal fit as well as the price impact. We developed a fast and precise recommendation engine ourselves, and mixed its decisions with a price prediction model that includes all KPIs relevant to the campaign.
The challenge for the development team is to have all relevant data in place and have the optimization provide interesting apps for the user fast enough.
Mobile Ad Fraud is a big topic in the industry and some are saying at least 20% of all mobile ad traffic is fraud. Many companies are struggling with fraud. From our perspective, publishing and advertising ourselves, it is an even bigger challenge. We need a sophisticated anti-fraud system to stay clean. In AppLike we have a dedicated team responsible for building the strongest solution on market. We combine different approaches (which is really unique in the market), like typical machine learning models, behavior analytics and detecting smartphone manipulation. This makes us much more efficient than other competitors on the market. This service is deeply connected to the attribution services.
Do you like to get in touch with Spark and Hadoop? Are you interested in finding the next security breach? Do you have deep experience with android, especially with the NDK and the underlying linux OS? Then help us to make our system even stronger!
If you own a website, you want to improve the experience for your visitors. One typical tool to try such improvements are A/B tests. You might try for example, if a different button color makes your site more appealing. When you sample randomly from your users in web traffic, you can use the IP as an identifier, and distribute different versions of your website to different IPs. This is transparent to you as the testing developer, enabling easy analytics according to the IP-address, it is hardly noticeable for users, and you can easily revert to the old version or the new version for all users.
With the Google Play Store A/B tests identifying which users got which app version in your own data can be tough, rolling back to the classic button color or distributing the new flashy button color to all users will have to wait for the next release cycle – you expect to reach only about 50% of your existing users within 24 hours, and you shouldn’t expect to ever reach 100% of users.
Have you ever considered to do A/B tests for mobile apps – and doing it live, maybe within a day? Imagine the possibilities, if product managers or marketers could each test different versions of our apps within the day and follow those incremental improvements every day. We have built a service, which does exactly this for native apps as well as hybrid apps.
We run different mobile apps with different unique selling propositions. Those are mostly developed in Java, Swift, React Native or Unity. Our frontend developers are always interested in using the newest technology, for example currently Kotlin, React Native with Typescript or Flutter.
It is mostly known how to do proper testing in Backend CI pipelines and proven best practices exist. One of our big challenges was to integrate a full circle test of our mobile apps into the CI server. To do this we need to run them on emulators and test them under more realistic circumstances, we mock their backend calls and gather the results. Are you interested in these results and do you want to help us improve with them?
We are startup, in particular we #move_fast, so we like to test new product ideas quickly with minimum viable products – MVPs – and we iterate on the product after we have tested the relevant KPIs, usually how many users, who see our Play Store page, also install, how many of them stay, how many of them get how we want them to use our productDo you want to test an MVP on a Friday for a product you just excited us about on Monday? Our in-house design team, the motivated Frontend-developers and their no less excited Backend-counterparts will ship any exciting app from concept quickly to the Play Store, and proudly iterate on the next great app idea with you.
AppLike GmbH © 2018