Everyday Data Science In Everyday Language

At this point, we are almost two years into the COVID-19 pandemic. To survive, businesses need to go online wherever possible. As a result, we must learn to work with the immense amounts of data that an e-commerce business can capture.

If you’re doing marketing online, you are working with data science every day. This is because every ad platform uses some kind of algorithm to manage its traffic. Hence, even if you are not an analyst or data scientist, you need to understand machine learning in plain terms. Otherwise, you might believe that predictions work like magic; you put in a target, and automatically your performance metrics will fall in line. Unfortunately, that is often far from the truth!

Machine learning is not magic!

How does machine learning work? First of all, the algorithm’s task is to find a maximum or a minimum point – usually a maximum, because both your business and the ad platform benefit from more traffic and more sales. Next, you give it a constraint (or several) to work within. Perhaps you have a hard budget limit. Or else, you might only want to pay a certain amount per click, or per customer order, or per dollar of revenue.

With these information, the algorithm must find the maximum amount of business that it can send your way every day. To find a starting point, it might begin with driving as many clicks as possible to gather information. A small step at a time, it will make adjustments, comparing results from each try to see if it is moving in the right direction.

In other words, machine learning is hardly magic – rather, it’s a slow process of getting things slightly less wrong each time. And most of us don’t have the hard skills (the theoretical knowledge) to write algorithms. Yet, we suffer financially when the algorithm doesn’t perform as we expect. And as business people, we need the soft skill of sensing when our marketing investment is on track or not. This involves knowing the mechanics of the process, and getting enough experience (trial and error) to get a good sense of when it’s working as intended – or not.

Exploring hills and valleys

Think of your earnings at every possible combination of clicks and unit cost as a landscape with hills and valleys. Every decision that the algorithm makes will move you by one step along the landscape. The goal is to find the top of the highest hill – if you want to maximize revenue; or the bottom of the lowest valley, if you want to minimize cost.

But unfortunately, you can get stuck on top of a small hill and miss a taller one somewhere else in the landscape; or similarly, find the bottom of a shallow valley. In these cases, you have reached “a local maximum / minimum”, instead of the “global” one.

Navigating thoughtfully

Just as you would use a map to choose which roads to take when on a trip, machine learning involves choosing a model to predict the outcome. This model then determines how you move along the landscape based on orientating yourself around different landmarks, or features. Have you ever gotten lost in an unfamiliar city? Similarly, an algorithm can “get lost” within the landscape if you don’t give it the right landmarks to find its way.

For example, suppose you have a business selling two products – paddle boards and snowboards. Therefore, you need your algorithm to predict how many of each item you will sell in a given week. Let’s say that you decide to use the previous 3 weeks’ sales as the only feature to predict future sales. Furthermore, you have a 20% end-of-summer discount on your paddle boards in September. Then, your algorithm will predict that you’ll sell even more paddle boards in October, and you may end up stocking up on paddle boards at a time when most people are looking for snowboards.

If this example makes you slap your head in frustration at how stupid it is, that’s the idea. Machine learning cannot work unless the humans who set up the model use their common sense and business knowledge to give the machine the right roadmap to use. In this case, if you just add one more feature – the season – you would get a much more sensible outcome.

Machines are only as smart as we are

Hopefully, this post debunks the idea that we use machine learning because machines are smarter than us. In fact, quite the opposite! Machines only have more patience to keep repeating the same steps thousands of times, with small but fixed changes each time around. So, don’t feel intimidated by machine learning. When you’re not getting what you want out of an algorithm, you should question why it is not working. Picking a different target, changing the starting point, or choosing another roadmap can all make a big difference – you’re in control!

Leave a Reply

Your email address will not be published.

This site uses Akismet to reduce spam. Learn how your comment data is processed.