There has been a lot of interest, if not hype, around artificial intelligence (AI) in recent years. Quite simply AI is about creating intelligent machines; machines that can do things as well as humans, if not better in terms of speed, efficiency and objectivity. While there are degrees of AI and factions that differ in their view of what true artificial intelligence is, there is a subset of AI –machine learning (ML) – that is enhancing all aspects of our daily lives. Machine learning is all around us, used in things we take for granted, such as search engines, recommendations on our favourite online stores or media streaming apps; it is also a driving force in autonomous vehicles and robotics. Digital marketers have also taken for granted the machine learning algorithms that the likes of Google and Facebook apply in their programmatic advertising platforms for bidding strategies or finding lookalike audiences.
Theoretically, anything that requires a decision engine can make use of ML. Let us define ML as simply programs that access data to learn for themselves and improve their decision making abilities. ML algorithms have been around for decades, but have become more prevalent in recent times – not only due to the accessibility of vast amounts of data, but also the technological enhancements in storing and processing such data, both in speed and cost. There are algorithms that allow us to train models to recognise objects (classification) or predict numerical values (regression); both of these are examples of supervised learning. There are also algorithms that allow unsupervised learning, where we can create models to group objects or people based on similarities inherent in the data (clustering) or discover hidden links between objects (association).
Another phrase that has become mainstream in the past year is deep learning, a subset of machine learning that takes inspiration from the human brain in creating algorithms known as artificial neural networks (ANN). Again, the concept of this algorithm existed decades ago, but has once again become popular due to the advancement in computing power, and many believe that ANNs are our best chance of achieving real AI. While other algorithms plateau in performance, deep-learning algorithms scale and perform better the more data they are fed.
While academia and the tech behemoths are pioneering the evolution of all paths in ML, some even striving for ‘the master algorithm’, there is a more pragmatic side, in the application of algorithms to answer everyday questions – questions that are typically answered with human analysis. This can be time consuming, using troves of data, or can introduce bias, no matter how objective we humans can be. So, where we have data, we have the potential to build useful models to improve decision making in terms of speed, accuracy and objectivity.
We can apply some ML algorithms in marketing without much knowledge of the maths behind the algorithms, though understanding the intuition behind it would certainly help. Agencies and clients alike are already applying a form of machine learning, namely regression, for market mix modelling, which has been around for quite some time. Another application of ML in media that has come to the fore in recent years, through the classification algorithms, is digital attribution. There are other areas that are ripe for ML in marketing. For example, an existing customer base can be segmented through clustering algorithms and the best segments used to seed lookalike audiences. This can improve the efficiency of our marketing dollars and increase the customer base with better quality candidates in terms of lifetime value, loyalty or any other metrics important to a given business. Another example could make use of collaborative or content filtering algorithms with existing customer base to determine recommendations for future products to purchase, based on associations with other customers or products from user ratings or product features. The recommendation engine can then fuel a remarketing campaign using dynamic creative.
These are just two applications for machine learning in marketing, but where we have data and have decisions to be made, we can become quite creative in finding use cases for ML algorithms.
Getting started with ML requires data analysts with programming expertise in usually either R or Python, and scaling your ML solution may require distributed computing power and data engineers with big data or cloud computing expertise. All this may sound daunting, but it actually doesn’t
take a lot to get started, given the endless resources available online. Machine learning is a valuable tool that will become indispensable in the modern marketer’s toolkit, so now is the time to get up to speed.