Machine learning: Fiction, fact and the future of marketing

Olivia Calafat, Fabio Ercolani / May 2019

Machine learning is often heralded as ‘the future of marketing’, but can advertisers make practical use of this cutting-edge technology today?

Much of digital marketing is about recognising patterns, spotting the difference between meaningful signal and background noise. But with vast amounts of data to process, it can be difficult to pinpoint important information; and, once you have it, to use it productively.

Janusz Moneta, Senior Ads Marketing Director at Google, explains: “Marketers need to be able to see the whole picture, from the most obvious details down to the faintest patterns. Thanks to machine learning, this super-charged vision is now possible.”

Harnessing the power of ML

In Google’s B2B marketing team, we utilised testing and experimentation to explore whether ML can help us uncover insights more quickly across 40+ countries and 20+ languages. We applied cutting-edge ML to every step in our growth funnel to transform our marketing strategies. This article examines the key contributions made by adopting this technology to our campaigns – along with their results.

Capture the feeling – ML and sentiment analysis

It can sound counterintuitive to talk about ‘sentiment’ in relation to machine learning. Indeed, B2B marketing has previously favoured rational messaging, ignoring or sidelining the fact that business owners are human beings, with human drives. By failing to appeal to these fundamental needs and instincts, have marketers been missing a trick? 

Natural Language Processing (NLP) allows marketers to analyse the sentiment of their messaging, with the aim of understanding why particular messages resonate. Using NLP analysis allowed us to tap into an extensive dataset of emotional connotations and use ML to generate new messaging variations at scale  – and in any European language – tailored to specific registers.

Increasing the levels of emotional acuity and empathy in our marketing and using controlled experiments across the UK, Italy, Germany, and France throughout 2018 to identify the best messages, resulted in an impressive increase in performance, in some cases even doubling the conversion rate of our ads.

The same techniques can be used to optimise landing page performance, mapping the connotations of page elements and testing different ML-generated alternatives. Again, we saw impressive outcomes, with the best variation performing up to 80% better than the original.

Crucially, the data collected allowed us to dig down into the nuance of each creative: in one case we found out that 81% of a banner’s impact was driven by a single phrase that inspired gratification.

Foreseeing the future – ML and predictive optimisation

B2B marketing often involves long purchase cycles, and the true value of a signup may not become clear for up to six months. Using a wait-and-see approach for optimisation on this kind of timescale is simply not viable.

However we found that by analysing historical campaign data, the ML system could learn to predict the potential long-term value of an ad click. Within just two days of a customer signing up, it was possible to accurately estimate their adoption of our solutions over the next three months.

Pre-post analysis of campaigns that were optimised for long term impact based on this prediction revealed up to a  33% improvement in ROI. The prediction of customer lifetime value also has the potential to streamline the sales process by ensuring that each new customer is assigned a level of support appropriate to their needs.

What can ML do for you – now?

It may be more straightforward to get the most out of current ML solutions if you have access to data scientists and engineers, but even these limitations are disappearing. The technology is advancing so quickly that off-the-shelf solutions are starting to emerge, and the tools to enable sentiment analysis and predictive optimisation are ready and waiting. 

With all the hype, it’s easy to assume that machine learning is a far-flung futuristic concept with limited application in the here and now, or even that it’s a threat to marketers’ own creativity and vision. However, by implementing this technology in current marketing campaigns and measuring the results, we’ve seen first-hand the key support role ML can play.

Using these cutting-edge tools to enhance and augment our own expertise, modern marketers now have the power to identify and understand customer needs like never before.

5 rules of play to automate for your customer’s needs — and serve them better