Intern Series: How Machine Learning in Digital Marketing Affects You

by Ovative Group
August 12, 2016

This summer, we asked our talented group of college interns to contribute to the Ovative blog. We asked them to choose a topic that interests them, do a deep dive into the subject, then develop a post to share what they learned. Enjoy this post below, then check out the rest of the Intern Series content!

 


 

If you work in tech, you’ve probably heard of machine learning. If you work in data, you’ve definitely heard of machine learning. Research budgets are exploding in private sector and academia alike, all driven by the advent of data-storage efficiency and the ever-increasing digital footprint of the contemporary Joe. Computer scientists and data scientists come together to hold hands and sprint to market in what is one of the largest development-oriented rat races in recent tech-sector memory. It’s no surprise then, when the new player in town sets its eyes on one of the more easily quantifiable areas of the economy: digital marketing.

Hold on, this is all going so fast. What even is machine learning?” you’re probably asking, if you come from a traditional marketing background.

The long and short of it is that machine learning (ML) is a subsect of artificial intelligence that we use to write algorithms that allow computers to recognize patterns and analyze data in a way that people can’t. If humans are modeling around a couple variables, they can typically predict more accurately than computers can. But designing models to incorporate hundreds of non-linear (or possibly even non-apparent) variables? Computers take the cake every time.

Okay, you’ve got my attention. But how integrated is machine learning into digital marketing currently?” you ask again, because you’re smart and realize that this is a big deal.

mother-board machine learning
Here’s a picture of a motherboard. We’re not really sure why it belongs here, but in doing research for this blog post we found that a lot of similar articles had a picture of a motherboard, so we’re going to have one too.

For purposes of this blog post, we’ve categorized the many applications of ML in digital marketing into four categories:

1. Display & targeting

Applying convolutional neural networks, ML can autonomously recognize discrete characteristics of content that you post. Third party software companies such as GumGum offer image-recognition type marketing that determines when you (the Twitter, Tumblr, or Instagram socialite) have posted an image with a logo or brand-name in it, even if your image caption makes no mention of it.

Alternatively, image recognition can be used to target consumers who take photos with products like yours. Pinterest just rolled out their own program to automate this process in-platform. For example – setting up a campaign that uses image recognition to target people who take pictures holding beer bottles for ads of your or your client’s beer. Now that’s efficient marketing.

Happy youngsters, enjoying efficiently marketed beverages
Why are they smiling so much? Because you’re going to sell them some delicious beer. Mmmm, efficiently marketed beer.

It doesn’t end with images, though. Companies like Indico have created comprehensive AI that crawl through user posts and text for population segmentation to predict any individual’s emotion, political alignment, personality type, personal information (age, gender, location) – all things perhaps not explicitly stated in your profile or text information. Spooky.

2. Real-time-bids for demand-side-platforms

Employing ML to optimize real-time bidding for media buys, companies like Datacratic are computing unique valuations for individual impressions in a demand-side platform environment. The idea is this: your machine determines, in a fraction of a second, the probability of conversion for an impression based on a two-sided funnel.

On one end of the funnel, it evaluates the previous performance metrics of the ad content (CTR, CR, Bounce Rate, Engagement, etc.) and on the other end analyzes an individual’s demographic background using either 1st or 3rd party data. It then projects customer value by predicting future performance, such as determining that individual’s future life-time value based on where they are in the consumer cycle and their likelihood of conversion.

Optimization doesn’t only occur in the area of time efficiency, but also in creating individualized weighting of attributed value for which targeting parameters forecast higher ad performance. Think about it this way: for some individuals, you might learn more about their consumer patterns using their age to predict future behavior. For other individuals, geographic location might be more important for predicting future behavior. Machine learning algorithms can help us understand the importance of any particular characteristic by looking at that person’s other characteristics.

3. Preventing customer churn

Utilizing supervised learning methods, ML algorithms can take the information of former customers who are already “churned” and use that to predict not only the likelihood of future churn, but also the when of future churn. This allows digital marketers to specifically target customers at risk for a churn prevention strategy, such as offering an exclusive “sale” or in-store credit.

This is particularly valuable because it prevents collateral cart shrinkage – if you were to make the offer available to everyone, people already intending to buy at full price would produce less revenue.

This is a butter churn. It's the only kind of churn anyone wants.
This is a butter churn. It’s the only kind of churn anyone wants.

4. Optimizing for incrementality in multi-touch attribution modeling

Attributing value to different channels in the consumer path to conversion is a difficult task. There are as many methods as there are marketing measurement teams, and some are a lot better than others. For those who can afford it – there exists the sweet nirvana of algorithmic attribution modeling. It is data driven, ML generated, and completely customized to the client. According to a study by the Interactive Advertising Bureau in 2012, only 11% of marketing service providers and measurement teams reported using algorithmic attribution modeling, although the number is undoubtedly higher now.

But even the best of the best models cannot measure incrementality. Not that this is an insurmountable roadblock for the good measurement teams – but it generally requires using the tool in tandem with a robust testing strategy, which can require a significant labor investment.

The future looks promising, however, and we data nerds believe someday soon there will be completely scalable, completely incremental modeling.

Actual picture of O/G AI tech.
Actual picture of O/G AI tech.

And most importantly: How do these things affect YOU?

That depends on who you are. As a consumer, you can expect that as time goes on the ads you are served will be more relevant to your buying interests. For better or worse, they may be served to you based on information you didn’t believe you had disclosed (in one case Target exposed a pregnancy).

As a marketer, this means adding more tools to your tool-belt in creating more effective campaigns and potentially lower CPCs. Setting up these ML algorithms is no mean feat, however. Data scientists are always coming up with new ways and improving upon old ways to solve these statistical problems. If you’ve got the time and the inclination, we suggest you head over to the wiki page and dive down the rabbit hole.

Latest Posts

/now
Marketing Measurement in an Increasingly Privacy-Oriented World
/now
How Healthcare Marketers Win: Digital Opportunities in Response to COVID-19
/now
Navigating Media Pauses: How to Reboot with Success
/now
Intern Spotlight Series: Summer 2020

Subscribe

Enter your email below to receive notifications of new posts:

  • This field is for validation purposes and should be left unchanged.

Popular Tags