Are ‘cutesy’ pictures of wild animals shared online fuelling the illegal wildlife trade and can marketers serve the increasing demand for social media animal marketing, but in an ethical and responsible way? Phil Aiston shows how working with animals – but with a conscience – can be a force for good.
So what is ‘deep learning’?
There are many different aspects of artificial intelligence (AI), but the field that is arguably having the biggest impact right now is machine learning. These algorithms can learn to recognise patterns and perform tasks simply from the data we feed them, without being explicitly programmed. There are a huge range of potential applications, because they are able to find the hidden insights in big data without even being told where to look.
Deep learning is a specific type of machine learning that employs multiple ‘layers’ of learning; decisions and data classifications are refined at each layer to produce a more accurate output. As you can imagine, the technical explanation goes much deeper (sorry) than this, but the important takeaways are the two main reasons why deep learning in particular is so powerful.
Firstly, it produces superior results with ‘unsupervised learning’ (i.e. the data we use to train the algorithms doesn’t have to be labelled), meaning we can make use of all that raw, unstructured information out there. Big data is a potential goldmine of marketing insights but a staggering 99.5% of all the data we have created is not yet analysed — unsupervised learning allows us to unlock those insights, because we can now automate the search. Secondly, what differentiates deep learning from other unsupervised learning techniques is that it can be adapted to lots of different solutions relatively easily.
One of those solutions is advanced machine vision. With deep learning we can now identify both text and objects in images and videos, and that is a hugely significant development for marketers because it allows us to analyse consumer activity on an unprecedented scale. Imagine being able to process all those social media posts and highlight when, where and by whom your products were being used, even when you are not explicitly tagged!
Not only does it allow us to garner raw stats on product visibility, but as we know, that kind of information is vital when it comes to designing effective campaigns. Who is your target demographic? What role do products and services like yours play in their lives? What text or images do they associate with them? Getting into the minds of consumers like this is key to creating a campaign that really resonates and, because the data is taken from a natural setting, the answers are more likely to be a true reflection of people’s thoughts.
We can take the idea a step further when we consider the emotions displayed in those posts. The gathering of opinions to inform strategy is nothing new in marketing, but again the scale and accuracy of what we can now collect is a game-changer. This ‘sentiment analysis’ is already being used to determine people’s feelings from the text they write, with deep learning used to automate and optimise the process. And, recently, companies like Affectiva (whose clients include Coca-Cola) have developed algorithms to recognise facial expressions through a webcam, so it’s not much of a stretch to see that technology being used to analyse our images and videos in the future.
But can we really detect the full range of human emotions accurately? Perhaps there is a feature of your product that customers are frustrated with — they might well express that frustration on social media, but that could be very difficult to identify if they use sarcasm or other subtle expressions. These indicators can vary widely across cultures and languages, too, making it practically impossible for us to manually program rules to spot them. So that’s where the real power of deep learning will become apparent; algorithms that are able to continuously improve through experience will eventually allow us to listen to our customers with human-like levels of emotional intelligence.
One of the most successful recent trends in marketing is personalisation — the ability to target consumers with timely choices and promotions, that interest them specifically, has proved invaluable to countless brands. Netflix for example estimates that 75% of viewer activity is recommendation-driven, and studies have shown that personalisation can increase the efficiency of marketing spend by 10 to 30 per cent. The underlying technology is known as ‘predictive analytics’, and it’s all about identifying patterns and trends to help us model the future.
As well as consumer demand, we can predict many other important factors such as campaign effectiveness and customer churn. However, in order to make those accurate predictions, we need to identify trends from huge amounts of historical data, often from multiple sources. And, as we covered earlier, that’s just where deep learning excels; automated algorithms that can process an entire dataset, not just a sample of it.
Deep learning and future possibilities
New applications for deep learning are being developed all the time and many of them could be beneficial to our industry in the future. It might sound a little Orwellian, but it has already been suggested that deep learning could be used to analyse surveillance footage, so what about the study of behaviour and facial expressions in stores to inform layout? Or even personalised advertising displays based on facial recognition technology?
The possibilities are almost endless, extending from financial analysis to better weather forecasting; perhaps one day we might see promotions that change depending on localised weather conditions, not just the month of the year we happen to be in.
Despite all the hype, we should still be aware of some limitations of deep learning — because we do need large amounts of data to learn from, and the learning process can be long and computationally expensive, it isn’t suitable for every application. Besides, your algorithm is only as good as your training data, which does mean that if, for example, we relied on biased social media feeds to produce a marketing campaign, there is the potential for it to produce something unpalatable.
But as long as it is employed correctly, deep learning has the potential to transform the way we do business, so it’s something we should all be looking at a little more closely.
Have an opinion on this article? Please join in the discussion: the GMA is a community of data driven marketers and YOUR opinion counts.
Please register below to unlock this article.
An email will be sent to you with your membership details.