The following is a guest contributed post by Sanjay Castelino, vice president of Marketing at Spiceworks.
Every business cycle has its trend du jour. Today, that’s machine learning—the idea that we can employ armies of intelligent systems to analyze data for previously undetectable patterns. Common wisdom espoused by advocates of the technology says the secret to meaningful customer engagement is in the data. Get enough machines to surface the right patterns and they’ll learn, on the fly, to create experiences that engage.
Sound promising? It should, because pie-in-the-sky sales pitches always do. Unfortunately, today’s machine learning technology hasn’t been tuned to measure the elements of engagement, such as the quality and depth of customer experiences. They’re instead designed to measure website clicks and other minutiae.
We’ll need years and perhaps even decades of meaningful innovation to get to the point where machines can design rich experiences from scratch using nothing more than raw data signals. In the meantime, we can use the minutiae—the clicks, browsing habits, and more—to better understand what our customers care about most when it comes to how we sell and service our products.
The expectations chasm
Understanding what our customers care about is more important than you may think. In his book, Hug Your Haters, author Jay Baer points out that 80 percent of surveyed businesses believe they deliver superior customer service yet only 8 percent of their customers agreed. The resulting expectations chasm is troubling and can make it difficult to design engaging experiences.
Think of how internet advertising works today. How many Google Now cards do you swipe away each day? A lot, right? How often does a Facebook ad for something you have no real interest in follow you around to every webpage you visit, stalking you like unwanted paparazzi begging for a shred of your attention? All the time, I’d bet, and yet Facebook and Google are the twin titans of internet advertising. They’ve built global networks of data centers filled with advanced systems that run custom algorithms meant to serve users with relevant, timely ads. The fact that even they get it wrong should be telling.
They are, after all, the early adopters of machine learning for marketing purposes. As a result, I should be able to look at the paid ads they present me with the same level of interest I do with all the unpaid content that comes my way online. The reality is, I pay little attention to both, which is why Facebook, Google, and others continue to invest heavily in this unfortunately-named technology that isn’t learning as fast as we’d like.
When machines lack creativity
The term “machine learning” also sounds predictive. Analysts certainly treat it that way. Virtually everyone who tracks the technology rates it as a subset of the multi-billion dollar market for artificial intelligence.
In practice, there is prediction involved with machine learning today, but let’s face it, it’s in the adolescent stages and not where we need it to be. Software and servers conspire to collect and sift through data to find patterns and take action on those patterns with little sense of nuance or creativity. This, in a nutshell, is why Facebook ads you’ve ignored for months continue to follow you around the web.
Frankly, it’s frustrating to watch as machines infer the wrong conclusions again and again in fumbling attempts to engage with their target audience. But perhaps machine learning systems can accelerate our marketing efforts if we simply streamline our asks of the systems we’re putting in place.
Responding to the right signals
What’s that look like? First, we need to admit that it’s asking too much for current machine learning systems to translate patterns into actions that produce instant results. Experiences take time to create, test, and refine. So instead of relying on machine learning technology to understand the nuances of human behavior and accurately predict the most relevant ads or content, we can simply deploy the technology to run a series of tests on our campaigns and compare the results of every new test to the established baseline.
That way, we can truly understand how customers respond to existing experiences and adjust them accordingly to increase engagement. Also, it takes the difficult and often dreary work of number-crunching out of the hands of marketers paid not for their spreadsheet prowess but for their creativity.
Ultimately, there is opportunity for marketers to take advantage of machine learning technology, and compared to where we were five years ago, you could call today’s technology a giant leap forward. But if your perspective is that machine learning should be like talking to the best friend who knows you intimately, we’re not there yet.
