How I stopped worrying and learned to love big data and machine learning

In my pursuit of curiousity and knowledge, I decided to undergo a MOOC class on Coursera about Machine Learning. After few online readings, I found out that adding this skills in my personal toolkit could boost my digital marketing career.So let’s start…

Today marketing is an intensive task which requires marketers to crunch through large volumes of data, and most of which doesn’t really help them make impactful business decisions in the long run.

The digital universe is all set to grow by a factor of 300, from 130 exabytes to 40,000 exabytes by 2020. (Source EMC)

There are a lot people who believe “data (or big data) is everything”,but the truth is that what you learn from the data and what you do with it, is what actually matters.

Without Analytics, Big Data is Just Noise - Brian Solis

But with so much data around to consume, there is not much time to manually going through all the volumes and find insightful and actionable data. That’s where machine learning comes in.

What is machine learning ?

Before diving into machine learning and the benefits to your business,  let’s find out what machine learning really is.  Machine learning helps understanding data and statistics. It’s the process where computer algorithms find patterns in data, then predict the probable outcomes.  For instance, this is how your email software is able to scan the messages you receive and determine whether a particular email is SPAM or not depending on words in the subject line, the links included in the message or patterns identified by looking at a list of recipients.

What makes machine learning really useful is that the algorithm can “learn” and adapt its outputs based on new informations. For examples, when spammers change their tactics, the machine will quickly pick up on the new patterns and again correctly identify dubious messages as SPAM.


How big organizations use machine learning

At first, Email monitoring was just the beginning.Now, machine learning is everywhere. When you use Google Translate, there’s an algorithm translating what is said into actionable text. It’s same thing, when PayPal uses at least three different machine-learning models to judge whether users pose a risk of fraud. Facebook uses it too, for scanning images, looking for faces, then suggests members tag the people the algorithm finds in the picture. And what about us? Let’s say you  are watch a course about Photoshop and also one on InDesign. We can know you’re probably a Graphic Design, or are at least interested in desktop design, and we can use machine learning to recommend other courses you might want to watch .

However, machine learning goes well beyond what’s said above. In the real world, It can be used to predict transportation traffic patterns, weather forecasts, outbreaks of disease, stocks and commodities or hardware failures or spikes in web traffic—all of this so you and your organization can plan and react accordingly. The keyword in machine learning is : Prediction .

The challenges with machine learning

As exciting it might looks, there are some challenges when implementing machine learning in any organization. Fiirst, you need to understand what kind of algorithm to use for the problem you need to solve. Firstly, you can use a clustering algorithm which can help to classify a restaurant customer as more likely to dine in than take out, but it can’t be used to predict how raising menu prices would impact sales. In the other side, you can use a regression algorithm which can help to determine the effect of price changes on sales, but can’t predict one of a closed set of outcomes.

There’s a risk of “overfitting” the data, it happens when you train the machine to understand a set of data so well that it lacks the ability to generalize, learn, and make predictions based on new data. In this case, the models tend to make inconsistent predictions and finally become worthless.

In addition, some problems may not be solvable with machine learning. Unfortunately, you can’t always predict which can be solved, so the process of applying machine learning to the data never ends, leading an organization to chase the problem but never develop a functional model. In this case, the solution is knowing when to quit trying.


Should your organization adopt machine learning?

My answer is a big YES ! When implemented correctly, machine learning can help you solve enormous problems and predict user behavior in ways that will help your organization grow. As you know, organizations are drowning in data; from web analytics, customer demographics and usage information, purchase behavior to pricing, inventory and delivery systems—all of it impacts customer behavior and organizational growth.

Thank for us, the massive increases in cloud-based server capacity in the past 20 years make it possible for machines to analyze the data, make helpful predictions, then learn and adjust based on how accurate the initial predictions are. As more data is analyzed by the system, the prediction model improves and gets more precize.

So if machine learning can understand data and make predictions that will help any organization grow, why not do it? To make a good machine learning system for your business, you need four things:

  1. Understand of the machine learning process
  2. Understand the different algorithms available and the kinds of problems to which they can be applied
  3. Data (the more, the better)
  4. Patience