Data-powered marketing has been all the rage for some time now. Well more than the rage; well-established is a better description. After all, ADMA Data Day celebrated its 15th birthday this year!
REA Group is also well-established in this space. That is not to say we don’t have a long way to go, because we do, but wherever we can we provide our consumers personalised, relevant, timely and contextual marketing messages based on data.
We understand and define people’s needs through behavioural models, profiling and classification algorithms. And we determine the right content via recommendation systems to provide data-based content that is relevant to them.
A great example is the REA Group property purchaser model, which uses machine learning to identify the consumers most likely to have purchased a sold property. A message is sent to recipiants that is personalised to both the individual and the property, giving them timely information that will help them on the next stage of their property journey – whether that’s understanding their new property market, finding moving services or even utilities connections.
Data-powered marketing has been driven by the explosion of the data available to us through digitalisation. This includes both behavioural data on digital platforms, and content data, particularly in the form of text, images and videos. At the same time, there have been vast improvements in computing power, and access to it, through the likes of GPUs and cloud-computing, to help capture and process all this data.
Combined, these have led to a surge in new ways to analyse data and build predictive models (and re-opened the door on some old ones). Right now is a dynamic and exciting time to be a data analyst or data scientist.
But it can also be a dangerous one…
With a vast and growing collection of techniques at our disposal, it can be easy to focus on testing various options, trialing increasingly complicated techniques and fine-tuning them. This is fun after all! And don’t get me wrong, there are certainly times when this is well worthwhile, but it can be easy to fall into the trap of ‘because it is so complex, it must be right’.
In practice, Occam's Razor often applies; the simplest solution is almost always the best. They’re typically easier to develop, explain, implement and update. More often than not, greater rewards are reaped when effort is focused on the data rather the way to analyse it; be it ensuring you have all the right data, or how that data is transformed prior to analysis.
The property purchaser model is one of REA Group’s most effective. It is based on a key selection of rich features which are fed into a relatively simple classification algorithm. All the performance improvements we gained were from improving the data that fed the model, rather than improving the algorithm or testing alternatives. This is just one example of many – the theme has been consistent.
At the end of the day, it doesn’t matter how fantastic your algorithm is – if you feed it bad data, you’ll get a bad output, receive bad insights, and then make a bad decision. My advice; data-driven marketing can be incredibly powerful… just don’t forget about the data.
By Glenn Bunker,
Data Science Manager, REA Group