Retargeting is an important campaign strategy with many options available to marketers. But how are artificial intelligence (AI) technologies helping to make ‘smarter’ campaign decisions and change the retargeting game? Writing exclusively for ExchangeWire, Radoslaw Dobrolecki (pictured below), US business development director, RTB House, talks about AI, the advantages and differences between machine learning and deep learning, and the importance of considering deep learning technology an integral part of your marketing strategy.
With the rise of programmatic, ensuring you’ve got a competitive advantage in a noisy marketplace has become more difficult than ever. Artificial intelligence (AI) is key, but it is hard to understand what vendors really mean when they tell you their technology is ‘intelligent’.
Let’s start by admitting that AI is the core of competitive advantage in retargeting; you need only to throw a stone in any direction to hit someone promising you optimised results on account of their AI-powered solution. And the US has a wide variety of retargeting and programmatic vendors, all of whom claim to have great technology.
But when they tell you they’re AI-powered, what they’re really telling you is that, in most cases, they’re using ‘machine learning’.
What is AI?
Before we get into what machine learning is, let’s demystify AI. Like any other category, artificial intelligence at large includes great, innovative technology, but also older, less-powerful technology, when it comes to analysing colossal amounts of data.
Machine learning is essentially an AI subset in which algorithms are taught via data input. They absorb massive swathes of data and are trained, using more data and human-defined categories, to improve their predictions.
In short, there is a limit to how much it can learn – and, thus, to how much it can optimise. To truly do something new, it requires human input, either in the form of new data, or an updated algorithm.
You want deep learning
If you’re really trying to get the biggest bang for your retargeting buck, what you want to look for is deep learning. Deep learning draws its roots from artificial neural networks. It’s the newest technology to spring from everything we’ve learned about machine learning, and is based on how the human brain works, mimicking extremely high amounts of neural connections. The technology is so promising, that it is used in innovative solutions like Google self-driving cars and online translators.
Essentially, deep learning can learn to optimise on its own, using entirely new assumptions, making it far more efficient and rapid than machine learning can ever hope to be.
Beyond the tools, let’s take a look at a practical example. In chess, machine learning needs to know that there are three parts to the game: a beginning, middle, and end. It iterates how to win within each part, but will always constrain its decision-making within those first three components. It’s sticking to the beginning, middle, and end. Machine learning needs data to analyse the rules and how to learn. In essence, machine learning knows when a programmer will implement certain specifications – and will only activate against that standard.
Deep learning, on the other hand, has the ability to learn on its own and asks, why not create five steps, or fifteen? It will treat every new move its opponent makes like a new game altogether for the best strategy. Deep learning needs to know the rules of the game and will learn on its own, remember when Google AI beat Go world champion after AlphaGo made a mistake early in the game – can you imagine the implications for retargeting?
AI & advertising
In the ad space, you want to show the most relevant ad – the one with the highest probability of yielding a purchase for the person in front of it and driving them to your e-commerce store. Machine learning is restricted to the categories that humans have defined as relevant to purchasing – like price or product category. Those variables are manually plugged in, and must be changed if you want your AI to take new information into account.
Deep learning looks at the larger pool of products and the larger pool of that individual person’s behaviour. It’s going to make surprising assumptions. Imagine that you’ve forgotten about your friend’s birthday. With only two days left, there isn’t much time to search for a product, but still enough time to look for something special. In these cases, ultra-accurate personalisation will make the difference, and deep-learning models can begin to recognise that you’re enthusiastically looking for something. Whether it’s by a sudden change in behaviour, or a seemingly urgent spree, a typical model can be blind to these data points, but deep learning can make the connections.
In contrast to the traditional machine-learning approach, deep learning is able to find one user in an online crowd, a user who initially may appear to be acting in a disorderly fashion, but in fact may have the biggest potential to finalise the purchase. It’s possible because self-learning algorithms define every potential client, even the one who searches for a product slightly different from customary models. It references history, and sees that the person has changed his or her behaviour dynamically. Deep learning then delivers extremely precise conversion probabilities, learning not only from one user, but every user in the network. Think of the time and money you can save with an algorithm that practically intuits behaviour, even as it evolves – which it is always doing.
I’m not suggesting throwing the machine out with the bathwater. A digital or performance marketer will know to try both solutions simultaneously. The massive potential of performance-based solutions has already been seen in the market, according to Marin Software, nearly nine-in-ten marketers use some form of retargeting as part of their online strategy. In a time when almost all e-commerce players use retargeting as a way to multiply conversion and get superior ROI, finding the smartest approach for your own business becomes key.
And what does ‘smarter’ entail in this case? Well, when planning an online campaign, we have a number of providers and a million options to choose from, relating to machine learning and deep learning. Selecting the right one for your business without proper due diligence is virtually impossible and may lead to suboptimal results. This is the reason why multiple retargeting strategies and different techniques to expand access to selected users have been developed. Today, it is obvious that using a single retargeter is not always optimal. In fact, running two to three retargeting campaigns with different providers can bring much better results without any additional cost.
Technology changes everything! Don’t let your competitor be the first to act on that knowledge.