Five things to know about machine learning

B2B | Data Science | Machine Learning | Performance Media | SFBJ

Reprinted from the South Florida Business Journal.

The definition of machine learning comes under the umbrella of artificial intelligence, but is less about human-like decision-making and more about writing code, called algorithms, that will automatically change outcomes as new data becomes available. In other words: These algorithms evolve and improve as data changes, without developers having to write new code.

‘Suggested products’ started it all

The first thing for marketers to realize is that machine learning crept into their worlds in early e-commerce days in the form of “suggested products.” Amazon.com has made an art of suggested products. A further adaptation has been an upgrade in “retargeting” web visitors. As online shoppers know, when you click on a product photo, it follows you around the web – no matter what site you visit. This is basically pixel tracking.

But through machine learning, retargeting is getting more intelligent. A retargeting campaign can now present ads based on predicted user intent, or those that perform better for the persona visiting the site. These methods of following consumers and suggesting products around the web continue to be more intuitive and sophisticated.

Programmatic advertising

A second thing to know is that, while machine learning is, at times, intertwined with programmatic advertising, they are somewhat different. Both use big data, but programmatic ad buying is a process of specifically using software to buy digital ads based on audience segmentation and characteristics. Media buyers bid for impressions in real-time bidding protocol.

With this and other ad-tech software, demand has increased for using machine learning algorithms that predict future consumer behaviors. These algorithms optimize ad buys that maximize conversion rates for purchase, signup, bookings and the like. Their success is reflected in minimized costs per acquisition and improved overall brand performance.

Multiple sources of data needed

Third, data modeling to identify target customer characteristics by combining databases from disparate sources is still important to the overall success of these programs. Several years ago, Facebook released customer “look alike” targeting by allowing users to input customer email lists and matching them up with their extensive user information. This made finding Facebook users who looked like your customers – because they had similar interests – a snap. But the results could only be used for Facebook marketing, so you still needed to combine and model databases for your targeting activities.

Predicting performance

A fourth realization is that, by using machine learning for advertising, you will have the potential to predict ad performance before the ads are run. This is only possible with significant ad history of the brand, similar products and campaigns. Still, this growing intelligence makes marketers, agencies and even financial officers more excited about how machine learning can take the mystery out of advertising results.

Sales and marketing work together

A fifth important point is that, even with the significant targeting prowess of machine learning algorithms, marketing and sales need to closely align their activities. A target prospect who has already spoken to a salesperson may not welcome heavy online targeting. This makes helpful, relevant ad content more important than hard-sell messaging that can be a turnoff for a “hot prospect.” The same is true for ads with offers that may differ from offers the sales staff is using.

Overall, advances in machine learning for marketing and advertising continue to make ads more relevant, reaching your prime customers and prospects with the right message, in the right place at the right time – a definite competitive advantage.