Harnessing the power of machine learning to drive smarter decisions
Founded in 2014, ChimpWorks has built their reputation on the back of a series of highly successful mobile games.
They produce simple and satisfying short session games like Baseball Boy, Rush and Rider. As a company, they’re continually adapting to the changing mobile gaming landscape and improving every aspect of what they do.
To achieve his goal, ChimpWorks knew that automation was critical.
Creating automated reports that acquisition and marketing managers could access on a daily basis was where they needed to be. That meant establishing dashboards that allowed the team at ChimpWorks to plan, predict and make data led decisions.
ChimpWorks relied on a combination of BigQuery and Amplitude to manage their data and analytics. Together, these tools helped them understand their player level data. However, they lacked the ability to drill down and go deeper. Critically they were missing the ability to accurately attribute which of their marketing channels was driving conversions.
And, of course, working across two analytics tools made it difficult to establish a genuine single source of truth.
“We were missing a lot of key aspects of the data. We could see our player level data, but we didn’t get accurate attribution data. It was impossible to say whether our players came from our Facebook Ads, Google Ads or if they were organic.”
As well as wanting accurate attribution data, ChimpWorks also wanted to drive even more value from the data. His aim was to take the richer data from AppsFlyer and use it as the foundation for two machine learning projects; the first focussed on optimizing UA creative, the second a forecasting tool.
There’s a simple truth about all forms of advertising. Some adverts work. Some don’t. Much of that success is down to the creative execution.
The marketing team at ChimpWorks wanted to understand which combination of creative elements produced the most successful adverts. Understanding the relationship between colors, characters, text and the flow of the advert would help produce more effective creatives.
“To do that manually would be quite a bit of work. We pulled the data from AppsFlyer and trained a model to just pick up the objects in our ads. We can look at the individual elements and compare which has the highest retention level. And that really helps us direct our creative team. We stop them wasting time on producing creatives for useless A/B tests that we know won’t work.”
Predicting the future
Machine learning has also been applied to help ChimpWorks marketing managers make more accurate forecasts about campaigns.
They have the ability to forecast the performance of campaigns over 3, 7, 14, 30, 60 and 90 days.
Now, at any point in the lifecycle of a campaign, marketing managers can answer a really simple question – “Is this game performing as we’d predicted?” and, of course, if the answer is no, corrections can be made in real-time.
“Our marketing managers can instantly see if a creative is not performing well and more importantly they can drill down and see why it’s not performing well. We can save them around two to three days of work. In one example we managed to increase our ROI by 37%. We always see an improvement based on campaigns that use this model.”
Collaboration has been key to ChimpWorks’s success.
“Support tickets are picked up instantly and transferred to whoever is best placed to help. Most of my answers actually come from Jonatan, my AppsFlyer Customer Success Manager. He’ll talk to the support team and come back to me personally, which is very nice.”
The data team at ChimpWorks are constantly exploring new ways to support their marketing colleagues.
That means harnessing AppsFlyers native capabilities as well as its ability to provide the trusted data for their machine learning projects. What that ultimately means is more success for ChimpWorks and a better experience for their team and their players.
“The best thing is that AppsFlyer gives us cleaned up data. And, we can go fairly deep with that data as they allow us to mix and match whatever combination we need from their API. Most data partners just give us a table and tell us ‘Hey, this is what you get.’. Now we just load up what we need from the API, which makes my job much easier.”