• SUBSCRIBE

    THINK DIFFERENTLY ABOUT DIGITAL MARKETING FOR RETAIL

    Join innovative marketers like you and stay connected with the latest data-driven research and commentary, direct to your inbox
    • Perspectives grounded in proprietary research, data, and experience
    • Expertise from digital marketers and analysts who understand your role, business, and vertical
    • Answers to critical digital retail marketing questions, before you’ve even asked them



The What, Why, and How of Machine Learning

Ryan Williams

“What is ‘machine learning’?” is high on the list of most-commonly-asked questions of the Sidecar team.

Machine learning underpins Sidecar’s optimization technology for product listing ads, and we’re not shy about making that fact known. But we understand that the term might sound a little bit sci-fi to some.

However, since it’s so central to what we do and who we are, we thought a simple explanation of machine learning was in order — no computer science degree required.

Rock or Not?

At its simplest, machine learning means giving computers the power to teach themselves to make decisions using historical examples, rather than explicitly programming them to perform a task.

To illustrate this concept, I’m going to borrow an example from The Data Skeptic, a popular — and highly recommended — podcast by Kyle Polich.

Say you want a computer to identify whether a given song was performed by Pearl Jam (Polich uses jazz, but I’m a die-hard PJ fan), there are basically two ways to go about it.

You could feed the computer every existing recording of Pearl Jam — we’ll call these inputs — and label each one as “Pearl Jam.” Your program could then correctly identify a tune on the radio as “Pearl Jam” or “Not Pearl Jam,” by comparing what it’s hearing with the contents of its database and looking for an exact match.

Machine learning means giving computers the power to teach themselves to make decisions using historical examples, rather than explicitly programming them to perform a task.

But there’s one major catch: when Pearl Jam writes a new song, your program would be stumped because the new track wouldn’t match with any it has encountered before.

Sure there would be similarities, but you’d need to manually update your program every time Pearl Jam released a new record. Safe to say you probably wouldn’t want to use this method …

Let the Records Play

A human would learn to identify songs in a very different way: He would simply listen to a lot of Pearl Jam. (Seriously, this is the way to go, dudes and dudettes.)

After a while, he would recognize the sounds of Eddie Vedder’s folksy, baritone vocals or the blistering guitar interplay between Stone Gossard and Mike McCready as Pearl Jam’s sonic signature. With this knowledge, he could identify never-before-heard Pearl Jam tunes with excellent accuracy — and improve that accuracy as he absorbed more of the band’s music.

Machine learning emulates this human approach to learning and problem solving — but at a massive scale. It can rapidly analyze growing volumes of complex data, far larger than what humans can comprehend — and detect patterns in that data, far beyond what humans can see.

Instead of loading just Pearl Jam’s existing catalog to the program’s database, you could take the machine learning approach and play it songs by Pearl Jam and songs by other artists. Then you could apply a machine learning classification algorithm to spot commonalities in frequencies, tempo, pitch among the tracks labeled “Pearl Jam,” along with differences between those and the songs labeled “Not Pearl Jam.”

Machine learning can rapidly analyze growing volumes of complex data, far larger than what humans can comprehend — and detect patterns in that data, far beyond what humans can see.

Play a track it has never heard before at random, and the program mines its historical examples of recurring patterns to classify that song as “Pearl Jam” or “Not Pearl Jam.”

And just like a human, the computer’s accuracy constantly improves over time as it collects more and richer data, and uses that data to add greater detail to patterns and trends.

Do the Evolution

Machine learning has become so popular because it marries the raw analytical power of computers with the flexible decision-making abilities of the human brain. Email spam blockers. Credit card fraud detection. Google’s search engine and its self-driving cars. They’re all powered by machine learning.

Program a computer using traditional methods to perform a discrete task, like identify Pearl Jam songs or flag every email from a Nigerian prince as spam, and it will dutifully follow those instructions — but will struggle when confronted with an unknown variable.

When the problem you want to solve (inevitably) evolves or takes a new form, a program that has been developed to search for answers on its own will continue to offer new insights.

And because spammers are bound to flood inboxes with inventive new scams, any program designed to stop them must be constantly rewritten to account for these, too. This is cumbersome and wildly inefficient.

When the problem you want to solve (inevitably) evolves or takes a new form, a program that has been developed to search for answers on its own will continue to offer new insights, without needing to be rebuilt from scratch or manually updated.

Why Do We Use It?

This adaptability makes machine learning technologies ideal for tackling complex, data-heavy tasks that are also highly fluid. And as every e-commerce pro knows, marketing products online is definitely both complex and fluid.

Items go out of stock or out of style, seasons change, products go on sale — and these are just short-term challenges. Over time, organizational strategies shift, and marketers must reshape their tactics accordingly. This can mean adjusting ROAS or cost/sale goals, or placing a greater emphasis on profitability.

Adaptability makes machine learning technologies ideal for tackling complex, data-heavy tasks that are also highly fluid. And as every e-commerce pro knows, marketing products online is definitely both complex and fluid.

Similarly, the data points used by Sidecar’s models are constantly changing. AOV might be up one week, but down the next. As data changes, Sidecar’s technology reacts appropriately — taking actions like increasing or decreasing a product’s bid on Google Shopping.

Because we’ve built our models using machine learning, introducing new information actually makes them more accurate and more powerful. More data typically leads to more precise outcomes.

The Best of Both Worlds

To sum things up, machine learning means developing computer programs that can quickly and accurately analyze data and use it to make decisions in a way that mirrors the human learning process.

When you combine this human approach to decision-making with the awesome data-crunching power of computer processors, you unlock the best of both worlds. And we’ve built our optimization technology for product advertising to do just that.

Still curious about machine learning? Next to Pearl Jam, there’s nothing I like more than geeking out on this stuff. So leave any questions in the comments section below and I’ll do my best to quickly answer them.

Comment

There is no comment on this post. Be the first one.

Leave a comment