Machine learning: Behind the hype
The study and application of machine learning and artificial intelligence have been around for decades, yet recent hype in the advertising industry presents this as an entirely new and disruptive trend.
Machine learning
With so much industry noise, how could this conversation about the hype play out between marketers and vendors?
Headlines used to be all about big data. Now they are about machine learning. Is there a relationship between these terms?
Machine learning and big data are symbiotic. You need a lot of data to use machine learning and, it has become increasingly obvious, you need machine learning to discover insights and take repeatable actions from that data.
Take for example the evolution of facial recognition within Facebook. Initially, users had to frame a face in a photo and then tag it. Then Facebook was able to detect faces automatically. Now it can recommend who the person is (depending on settings) with its DeepFace approach “closely approaching human level performance”. How? By using increasingly sophisticated machine learning on the huge volume of photos Facebook allows you to store for free.
As computing storage and processing costs came down, 'big data' simply went mainstream in business language before 'machine learning'. But it’s the increasing commercial innovation based on machine learning and the promise of general intelligence that is capturing the imagination... and generating the noise.
I feel pressure with the application of machine learning advancing at a rapid pace. Within advertising, where has machine learning come from and how is it being used today?
Chances are that you, or one of your vendors, are already using machine learning in your advertising and have been for some time.
For example, credit card companies and customer relationship management companies like Experian and Acxiom have been clustering and segmenting for decades. Fast forward to today and there are multiple applications that are dependent on machine learning techniques, including lookalike modeling, campaign delivery optimization, product recommendations, and probabilistic device matching.
I have some hesitations that machines will replace humans and am fearful of relinquishing control.
Machine learning will automate most of your campaign planning and delivery efforts. Eventually.
It’s a complicated and manual process today to design and deliver a targeting strategy. With increasingly digital paths to conversion and connected devices, machine learning can be used to automatically optimize delivery against potential customers as their underlying interests, needs, and brand interactions evolve over time.
However, humans are not completely out of the equation. Firstly, as “The Godfather of Ad Tech” Dr. Boris Mouzykantskii, chief executive and chief scientist at IPONWEB, noted at a recent ExchangeWire event, “You still need to watch, very carefully, at the business level what your machine learning is doing for you.”
Humans must police the controls, set the alerts, and review the results because machines will do what they are told, even when you introduce bugs in the code. What you don’t need to do is watch the rapid iterations of test and learn. Secondly, the time saved on planning and delivery can be spent by humans on curating first-party data, creative execution, and consumer experience projects.
Will there be an obsolete adoption moment such as broadband internet replacing dial up?
It is unlikely that machines will take over completely. Broadband is great but there are still people that find dial up valuable. AOL had more than two million dial up subscribers last year! Similarly, there will always be a human touch in advertising. It’s the repeatable processes that are ripe for automation. Closer to home, programmatic advertising has disrupted the insertion order business over the last few years but that also will not shrink to zero, as not all orders fit neatly in simple interfaces.
What does the future look like for marketers given increasing automation?
Gartner recently released its 2016 Hype Cycle for Emerging Technologies, placing machine learning at the 'peak of inflated expectations'. This means that we’ve yet to hit the 'trough of disillusionment' (where augmented reality sits today) never mind reach the 'slope of enlightenment', when emerging technologies have true mainstream adoption.
So, be prepared to face increasing pressure to apply machine learning and then for some of those efforts not to work out initially or as easily as promised. Automating long-standing industry processes is not easy, but just like dozens of non-advertising industries have learned to trust and then embrace automation through approaches like machine learning, so will the advertising industry.
Doug Conely, chief strategy officer, Exponential
+001 510 250 5500
@exponentialinc