The most innovative marketers routinely pick up new things, try them out, and succeed or fail. Why are marketers struggling so much to adapt to artificial intelligence and machine learning? In this series, we’ll explore machine learning and artificial intelligence to build a foundation for understanding the field – and how it applies to marketing.
AI Applications in Marketing: AI-Powered Content Distribution
Distributing our marketing is both easier and harder than ever. Easier, because there are more opportunities than ever, more channels for us to find relevant audiences. Harder, because there are more distractions than ever, more channels for our audience ignore or avoid us on. In our smartphones, millions of more entertaining choices exist than our content, if our content is dull. Machine learning and artificial intelligence help us to overcome some of these obstacles in three different ways:
- Programmatic advertising
- Influencer identification
- Identifying highly complex audience data patterns
Programmatic advertising is machine learning-powered advertising. Gone are the days of setting up manual bidding and placements in ads; today, programmatic advertising does much of the heavy lifting in terms of setting up tests to determine which ads should go to which audiences.
With programmatic advertising, we provide our ad creatives and parameters for a campaign, and let the system do the rest. A programmatic advertising system will run hundreds or thousands of different tests to determine the best possible performance for our ads, find and bid on thousands of different ad slots and inventory opportunities, and run our ads for us.
Who uses programmatic advertising? Everyone from AdWords/Google Display Network users to the most sophisticated demand-side platforms uses programmatic advertising. If you’ve ever run a campaign in AdWords/Google Display Network in which you selected “Smart bidding” as the bidding choice, you’ve used programmatic advertising.
Identification of Influencers
Influencer marketing is all the rage in digital marketing, largely because many companies haven’t figured out how drastically the ad landscape has changed. Today, influencers account for 7.7% of marketing resource allocation, according to IDC. Yet an astonishingly large number of companies still do influencer identification using decade-old methods, finding influencers with nothing more than Google, qualifying them by follower counts.
Modern influencer marketing uses machine learning techniques such as graphing databases and sophisticated mathematics to find influencers. Algorithms such as Betweenness Centrality, Eigenvector Centrality, and Degree Closeness help us to map out social networks much more intelligently, then determine what kind of influencer we’re after.
Here’s an example using Social Media Marketing World’s 2017 conversations:
We might consider, for example, quality of connection when identifying influencers. We may not want the person who is connected to everyone; we may want a person connected to our specific audience profile. I’d rather work with an influencer who’s influential with CMOs, for example, than consumers.
Much of the best influencer identification software, like AI software in general, is available for free, such as Apache Tinkerpop, Apache Giraph, and other open-source tools. With effective, modern influencer identification, we improve our content distribution efforts.
Patterns in Complex Audience Data
Finally, content distribution often fails because we don’t know our audiences well enough. We make broad assumptions using older tools like personas, or worse, we just assume everyone wants the same thing. Many patterns in data are invisible to the naked eye, to the average human mind, especially when we consider many different data dimensions.
Combining advanced statistical tests like random forests with simulations like multi-armed bandits gives us the ability to forecast and test many different assumptions about our audiences to prove or disprove what will resonate with them. For example, suppose in our random forest exploration we discover that LinkedIn is our best channel for generating desirable outcomes.
We might then take the known statistics and analytics from LinkedIn and feed that data as a starting point to a multi-armed bandit.
We’d test a series of probabilities – such as type of post (image, text, video) – to the software and make a simulation of likely outcomes. Based on our testing, we’d move forward with the outcome that’s most likely to generate the results we want.
To wrap up this series, we’ll discuss some of the things you can do to prepare in your personal career for an AI and machine learning future. Stay tuned!
Want to read more like this from Christopher Penn? Get updates here:
Get your copy of Marketing Blue Belt!