Machine Learning in Digital Advertising: the good, the bad and the ugly.

by | Mar 28, 2019 | Uncategorized

A few days ago Google announced that it’s putting machine learning into the hands of every advertiser.

 

So, what does this mean for advertisers?

 

First, let’s define machine learning in relationship to digital ads. It works as an advertising platform (such as Google) defines mathematical models to describe the relationships between certain aspects of user’s online behavior and demographics and the probability of them to perform certain actions online. After a period of time, the algorithms analyze the data and adjusts the model to maximize the desired online action. For example, it sets up initial user clusters based on gender, age, geolocation, visits to certain websites and frequency of interactions with certain apps and assigns the likelihood of these users to download an app or buy something on the app. After some data is collected and analyzed, the probabilities are reassigned and the process iterated – the machine learning in action.
 
Real world example? There’s a Google advertising product employing powerful machine learning techniques called the Universal App Campaign (UAC) which has been available for a while.
 
How does it work? Google explains:
Unlike most AdWords campaigns, you don’t design individual ads for Universal App campaigns. Instead, we’ll use your ad text ideas and assets from your app’s store listing to design a variety of ads across several formats and networks. All you need to do is provide some text, a starting bid and budget, and let us know the languages and locations for your ads. Our systems will test different combinations and show ads that are performing the best more often, with no extra work needed from you.
 
So good!
Basically, add your app, some text, some images. Sit back and enjoy. Here’s a cup of chocolate for you!
You don’t need to hire an online marketing person anymore! Set up takes in 15 minutes – well, a bit longer as there are always tracking issues. Perfect if your goal is not to have less work!
 

However, as usual, there’s a catch!

 

UAC allows you to optimize for desired in-app action but what if paying for every desired in-app action is not good for your business?
 
Say, if you are e-commerce selling low priced products at low margins it makes sense for you to pay only to acquire new customers and then you use CRM or other free (“free”) marketing to get them to repurchase. And what if these users are already buying from your website or store – there are cheaper ways to get them to download your app or tell their friends to download it. Although algorithmically these are the most likely users to download your app through the ad and perform in-app action.
 
Or say, you are a food ordering app and you are willing to promote an offer during non-peak hours only, otherwise you have customers ordering at regular hours anyway. Machine learning ad technologies don’t know that and would keep serving your ads to your repeat clients, or during your peak hours, as there are higher probabilities of these users to perform the desired action.
 
Takeaways for your business? Track, evaluate, see what works for the bottom line and what does not. Forget about vanity metrics and the number of people who saw the ad.  The rule is if it doesn’t deliver the returns the channel is off.
 

And.., lastly, dust off your old math books – understand algorithms and your business better than they do!