Ghost In the Machine: Marketing, Machine Learning and Artificial Intelligence

Jason Lark

The current dialogue around machine learning – the ability for a computer to accumulate and adapt to new information without human intervention – tends to focus on either the accessible, high-concept innovations like driverless cars and robot chefs, or jargon-heavy discussions about algorithms and Bayesian networks. However, it’s something that modern marketers must become familiar with, no matter how abstract or technical the technology might seem.

Machine learning can benefit your company: It’s simply a matter of understanding how to apply data-driven technology to enhance your marketing campaigns. This involves acquiring and developing marketing software – the automation tools capable of learning from and building on ‘experience’ – but it also means starting from a solid foundation of good, clean data. Without relevant, targeted information, your machines may well learn the wrong lessons.

Here’s what you need to focus on to get the right results.

Features, Goals and Results
The first priority for machine learning should be establishing the essential features of your campaigns. Machines ‘educate’ themselves by identifying and building on patterns, so it’s vital to make sure they’re looking for the right things. To start with, you can use a simple “who, what, when, where, why” format. This lets your program know which platform(s) your marketing collateral employs, a brief summary of its visual and text content, time of publication, its location on the website and any detail you think might be relevant and important.

With these boundaries established, it’s vital to define positive and negative results for your software. This will depend on your business priorities and your expectations: at the most basic level, where you’re trying to drive individual transactions (and therefore revenue and profits), a purchase would be a positive result and no purchase would be a negative. This is obvious enough, but machine learning allows you to be more sophisticated and specific in your approach.

If you’re trying to increase the time customers are spending on the site, for example – with the narrower but no less important goal of improving your web presence’s general appeal – you can define a positive as “Spent over 60 seconds on” and a negative as “Spent under 60 seconds on”. The same basic principle applies to things like customer spend (or lack thereof), click-through rate, or any other metric that’s important to your campaign.

Machine learning strategies only get the right results for marketers when they’re created with specific results in mind.

Machine learning has made personalization so sophisticated that some question whether there’s any real need for segmentation anymore. Given the practical awkwardness of treating every customer and data point as their own segment, it would likely be more accurate to say that machine learning can significantly improve marketing segmentation.  To get the most out of it, it’s vital to make sure your data set is already as relevant as possible.

Attribute-based segmentation methods have plenty of utility: demographic, behavioral, and attitude-based values are important to know and understand. Machine learning builds ‘clusters’ to identify patterns based on these attributes. The aim is still to optimize the experience for each kind of customer, from the hesitant prospect to the loyal, long-term buyer; the individual who’ll leave after one bad experience and the individual who’ll let you make it up to them. Machine learning can identify these types, refining its algorithm to attain the clearest, most in-depth view of your customer base – allowing you to personalize your service and gently nudge your target consumers down the sales and marketing funnel.

Channel and Context
You can also use machine learning to discern the effectiveness of each marketing channel, and the content you distribute through each channel.

If you know that customers with high lifetime value eschew social media for phone and email communication – and that the opposite applies to customers with lower lifetime values – you’re in a better position to tailor marketing material to them. Machine learning helps you determine more sophisticated patterns, such as if they like to be contacted on a certain channel at a certain time, or if not targeting certain segments will save you money if these segments are likely to take up too much time.

Marketing teams can’t reasonably be expected to understand the ins and outs of machine learning algorithms. But as their department begins to sit more horizontally across the business, they can be expected to determine what they do and don’t need from the technology. That means clarifying the team’s needs to IT, it means refining your core datasets, and it means opening your mind to consumer patterns and trends that might not have occurred to you – but may well have occurred to the machine.

Jason Lark is managing director at Celerity, which builds, implements and orchestrates Adobe Marketing Cloud technology to enable you to create more rewarding customer relationships.