Ask anyone in ad tech what’s the “hottest topic” right now, and you’ll likely hear the same answer: machine learning. And who could blame them? Until now, ad tech has always been quite binary. If you land the results you wanted, then great; if not, then it’s time to make some tweaks. Machine learning turns this all on its head as it can analyse a data set and determine a successful solution in mere milliseconds – a process that might have taken a human hours, days, or even weeks.
But the real question you should be asking is not who’s adopted it, but rather who’s leveraging machine learning to deliver the biggest impact. You see, while the reasoning behind the uptake of AI is the same for most ad tech companies, how it’s used in practice can be wildly different, meaning you’re not always guaranteed the same level of success.
Building the entire mobile ad tech ecosystem
Appodeal is a perfect example of these differences. Our machine learning models are, at their core, deep neural networks. In layman’s terms, they’re essentially trying to replicate the human mind. For those of you who’ve watched many movies, don’t worry, we’re not trying to replace anyone with robots. All this means is that, much like our brains, our models consist of multiple layers, enabling them to deliver successful outcomes for all our clients rapidly.
A/B testing also remains a core part of our work at Appodeal. The difference now is that it’s used to test the efficiency of our machine-learning models. This is important because models can degrade in certain scenarios, such as if there is a significant change to the dataset they’re based on. So you do still require human input, although for the most part, these models can be left to run independently.
The other thing that technologically sets Appodeal’s machine-learning offering apart from the rest of the market is that we cross the entire mobile ad tech ecosystem. We have access to data not just from our demand-side platform (DSP), but also collated from the programmatic ad exchange, mediation and even publishers. We also have Chardonnay, our in-house gaming publisher, which lets us speed up these feedback loops and better understand how our models impact our customers.
That’s important because more data typically equals better results from the model, especially when filtered through an advanced neural network like that which we offer. These constantly learning algorithms can spot more patterns and better adapt to user behaviour.
What does all this mean for your results? It empowers us to deliver higher-quality ads that attract not only more users to your apps, but also the right users – those who will be truly engaged and genuinely contribute to growth. We primarily concentrate on achieving strong Return on Ad Spend (ROAS), as this is what most publishers today see as the key performance indicator. We also place a lot of value on metrics such as retention rate, average revenue per daily active user (ARPDAU) and customer lifetime value (LTV).
Ensuring your machine learning is ready for tomorrow
However, it’s not just about what’s on offer in the here and now; you must also consider where things are headed. For that reason, we’ve invested a lot of time and effort into research and development to ensure we (and our customers) remain ahead of the curve.
Vector is a perfect case in point. Around one year ago, we established a dedicated team under this name to work toward bringing machine-learning products to the entire Appodeal ecosystem, which is currently focused on bolstering the potential of our demand-side platform. Other teams within Appodeal are also developing ML solutions, such as our Exchange team, which has dedicated experts building models specifically for that area, including one that determines the optimal bid floor.
Additionally, it’s vital to consider how fast the mobile marketing landscape has been known to change. Take Apple’s App Tracking Transparency, for example, which allows users to opt out of sharing their data with the Identifier for Advertising (IDFA). Four years after its launch, it continues to be a challenge for many marketers, with around half of all users choosing to hold onto their privacy, resulting in a severe reduction in targeting data.
While those targeting Android fare somewhat better, Google did try to follow Apple’s footsteps with the Privacy Sandbox, but this was dropped in April after being stuck in development limbo for many years. There’s nothing to say that Google won’t decide to reignite its plans in the future, and we end up living in a world where we don’t have access to identifiers for our users across the entire mobile ecosystem.
Well-executed machine learning can help us prepare for such a scenario. At Appodeal, we are developing a solution that instead uses contextual information to target users, looking at things like the actions users are taking, the reasons why they’re taking those actions, and the behaviour being displayed in different apps. By connecting all those data pieces through the power of AI, we can help overcome this identification challenge.
