Do Algorithms Know You Better Than You Know Yourself?
Since the early days of digital advertising, the ambition for marketers has been to deliver the right message, to the right person, at the right time, through the right channel. Little has changed in these fundamentals, but at a time of near limitless data, today’s most advanced algorithms can predict user tastes like never before.
Increasingly sophisticated algorithms are able to display better mobile ads at the best possible moments in users’ journeys. And while the theory behind these techniques has existed for some time, the leaps taken in infrastructure over the last few years have made processing the reams of user data collected in near real-time a reality. The best machine learning algorithms enable the display of engaging ads that increase both short-term performance and long-term familiarity with brands. But what are the key elements that separate a killer algorithm from the pack?
Competing on quality data, from savvier consumers
Owning a smartphone is now a way of life for most of the developed world - with penetration up to 96% amongst UK 16-24 year olds - and with this comes plentiful opportunities to interpret data in innovate ways. Digital devices including phones and tablets leave contextual data with every interaction. In simple terms, with each click, hover, link followed, comment made, or even via powering up a device, data is created. When correctly collected and interpreted, these individualized and contextualized predictions can be used to understand or very likely infer something about those users through their behaviour.
The smartphone era has brought with it a higher level of expectation from users, who along with having heightened standards for the devices themselves, are now far more discerning about the advertising to which they are exposed. As such, advertisers are now inclined to pay more attention to the performance quality of interactions between people and their brand expositions. And this means going over and above the typical CPC and CPV metrics that have long become the industry standard.
Best Performing Algorithms Are Fed The Best Quality Data
However, writing the latest and greatest algorithms is just half the battle. In order to secure those coveted ‘Best in class’ results, a combination of both the most innovative, efficient algorithms and exhaustive, high-quality data is a necessity.
There’s an old saying in data interpretation, ‘garbage in, garbage out’, that has never been more appropriate in the current climate. As much as competing around algorithms will remain the most heated battleground for ad tech companies, excellent results all start with the quality of data. There’s no circumventing data input, and even the best machine learning algorithm will give deceptive results if only applied to low-quality data. It’s no coincidence that the runaway market leaders in machine learning for advertising, Google and Facebook, are also those with free-flowing taps of first-party, proprietary data.
More to the point, there’s no catch-all solution that will always be optimal in all circumstances; a variety of approaches is necessary to cater for differing sets of criteria. Each of these in turn come with decisions around which trade-offs are acceptable, with most of these typically around levels of speed vs accuracy. Like a top chef selecting the right ingredients for the right recipe, the most effective marketers will combine the most relevant data with the appropriate algorithm in order to deliver appetising results.
If you’re still planning to take part in the machine learning ad race, make sure your provider is not only collecting genuine first party data, but that it comes from future-proofed methods too. Otherwise remember; garbage in, garbage out.
Christophe Thibault, Chief Algorithm Officer, Ogury