Hordes of demon droids capable of emotion and ready to sacrifice humanity to achieve world domination. No, it’s not the next Will Smith blockbuster – it’s what many people imagine when they hear the words ‘machine learning’.
In fact, it’s a branch of statistics – the Fresh Prince of Bel-Air gives way to your accountant. But in reality, it’s a bit more complicated than that. Yes, it’s statistics, but on an industrial scale. Algorithms designed to process the quintillions of bytes of data that we produce every day – in short, to adapt to the Big Data age. As simple as it might seem, this technology offers never-before-seen possibilities for companies, and will have a huge impact on our daily life.
What is machine learning?
Machine learning refers to the act of identifying statistics in millions of pieces of data. The real innovation lies in the quantity of data that is collected and processed. With the Big Data boom, companies have more information at their fingertips about their customers, consumers, businesses and processes than ever before. They also benefit from significantly faster and cheaper storage and processing capacity.
Data is fed into the algorithm, which learns from existing information. It observes, detects weak signals and establishes correlations that would be imperceptible to even the most expert human eye. Companies hope to be able to identify different customer behaviours and the series of actions that result in this behaviour – quickly and objectively. This allows companies to anticipate and predict the behaviour of future customers through data comparison. The ultimate aim is to facilitate decision-making.
Machine learning in action
Essentially, these algorithms identify repeated data series – patterns of a sort. Once these patterns have been revealed, the algorithms can detect them in other similar data sets.
Imagine an e-commerce site that collects data from its customers in an Excel table. Each row of the table corresponds to a customer, and each column corresponds to a variable: name, surname, sex, age, email address, purchasing history, date of their last visit to the site, number of emails opened, pages visited, etc.
In the age of Big Data, the astronomical quantity of data makes this Excel table unreadable – hence the idea of developing an algorithm that analyses and reveals precise patterns. The program could detect that 90% of visitors aged between 25 and 30 who live near Paris and who bought product A over the past 3 months also bought product B. When a new line is created in the table (a new customer), the system can analyse it, compare the variables to other customers’ variables, and predict the new customer’s behaviour. And continue to learn.
Every new customer who meets these criteria will automatically receive personalized suggestions, encouraging them to purchase product B.
And this is just one of many possible applications.
Machine learning – what could it be used for?
We’ve seen that Machine learning is highly effective when it comes to e-commerce – whether for predicting or trying to influence a purchase, making personalized suggestions (cross-selling) or knowing when a customer is going to abandon a basket or unsubscribe from a newsletter.
Generally, machine learning is a way of automating the process of categorizing prospects. By observing past customer behaviour, a company can determine where a prospect lies in the conversion funnel. And this means that it’s possible to send the right content at the right time so the prospect passes through the various stages of the funnel, ultimately reaching conversion.
As well as e-commerce, machine learning applies to a range of fields.
This is perhaps the first application that comes to mind. When they were first created, chatbots were simply databases of questions and matching answers. With the arrival of machine learning, the software has undergone major improvements, and can now analyse sequences of words and understand complex requests.
As they deal with similar requests, their performance improves, ultimately giving the illusion of talking with a human.
Banking and finance
There are also a number of very clear uses in the banking and finance sectors. The most obvious is as a way of quickly assessing a customer’s financial health and the risks associated with granting a loan. To do this, the customer’s situation and variables are compared with information from customers who did not repay their loans. Banks have also put machine learning systems in place to identify customers who are about to close their accounts.
In trading, machine learning can also be used to predict fluctuations in share prices in the long term. The number of financial indicators means that it’s difficult for analysts to carry out objective evaluation on their own – hence the value of algorithms of this kind.
How could a computer recognize a face? Take the DeepFace program, which Facebook has been working on for a few years now. It can determine the 3D appearance of a face based on a photo – allowing it to obtain images of the same face from multiple angles.
Thanks to machine learning, the program then looks for similarities between this face and other photos it is given. The angle of a photo, light levels, hairstyle and other variables are no longer a barrier to face recognition.
You’ve probably heard about Watson, the AI developed by IBM. The American giant claims it can help hospitals and doctors to improve the speed and accuracy of medical diagnoses. Essentially, Watson is Dr House’s electronic twin. Or even a quicker and more objective version.
Here’s how Watson works. The system starts by taking in hundreds of medical textbooks and as many specialist journals. Once it has accumulated this knowledge, health professionals can simply submit the results of medical analyses to the system to find a credible diagnosis.
The limits of Machine Learning
Machine learning offers unrivalled possibilities for businesses, regardless of their sector. We’ve mentioned e-commerce, finance and medicine, but that’s not all. Vehicle manufacturers also use machine learning technology for their self-driving cars.
What are its limits? If we go back to our example of the e-commerce site, it allows the business to predict consumers’ preferences and offer them personalized recommendations. But it doesn’t give the slightest indication on what makes customers prefer one product to another, or how to influence or develop their tastes.