How machine learning can help us build effective contextual advertising
Yaroslav Kholod, director of programmatic operations at Admixer, explains what the industry must do to build a convincing contextual advertising infrastructure capable of taking the pressure off the soon-to-be dismantled third party cookie.
Across the digital marketing world, you hear the sound: advertisers are on the march towards a land beyond cookies, prompted by consumer demand for increased privacy, Google’s promised cookie shut-off and the efforts of companies such as Apple to curb excessive targeting.
What brands hope to find in this promised land are new ways to reach audiences. Contextual advertising – the privacy-first targeting method that displays relevant ads based on the content of a web page – is a source of particular excitement, and I share that feeling too.
Context, of course, was the means by which ads always used to be targeted, both in traditional media and in the early days of the web, before ID-driven targeting triumphantly – and, as it turns out, temporarily – rewrote the game.
In recent years, however, machine learning has dramatically boosted the potential of this old favorite, unlocking the opportunity to combine site context and user behaviour in order to deliver high-quality data segments for marketing purposes.
Contextual ads deploy machine learning to interpret pages, understand context and target the most relevant audiences. Properly trained, that technology might analyse not only the subject matter but the sentiment of the language on a particular page, and even the tone of images and video. AI can then potentially configure creative to match that context. And not only should contextual work for the web, but also for DOOH and CTV.
The latest marketing news and insights straight to your inbox.
Get the best of The Drum by choosing from a series of great email briefings, whether that’s daily news, weekly recaps or deep dives into media or creativity.Sign up
Contextual targeting is certainly rich with the promise of satisfied advertisers, happier consumers and better insights. But contextual is not a ready-made solution that will simply drop into place once the cookies are switched off.
Currently, context mapping adoption has serious flaws, and these require rapid action if we are to develop contextual targeting as an industry-wide approach:
The cost barrier
Contextual is costly and requires a heavy investment in the training of machine learning models. In particular, launch investment for companies seeking to design algorithms for context mapping and data collection can be huge. Publishers also face variable levels of expense on data storage, depending on their size, so they should add additional functionality to their content management systems. On the other hand, once this foundational work is done, costs start to decline.
The taxonomy problem
There is no standardized taxonomy for contextual – no single approach to the accuracy of placement of content and ad alignment. Different publishers and DSPs offer different taxonomies, which makes it hard to generate scale. And needless to say, as contextual becomes more popular, the problem will increase.
The attribution issue
Contextual, like any cookieless method of advertising, is hard to measure and attribute, requiring not just traffic metrics but the conversion and revenue data. These metrics may be hard to combine, though they ultimately offer a rich view of the funnel.
The danger behind these challenges, given the cost and sophistication of building machine learning with the ability to accurately interpret page sentiment and assign ads in real-time, is that small and medium publishers are cut off from the contextual opportunity.
What the industry needs, but currently lacks, is an open-source context mapping solution driven by IAB members, of the type that was successfully implemented for header bidding and ID solutions. The aim should be the development of a single approach to contextual targeting, combined with a set of rules for segmentation of content.
Contextual is an easy word to say, and its heritage means that everyone has some understanding of what it offers. But it is also remarkably complex to do well, and if it is to make good on its promise as a real-time, AI-driven index of content sentiment, we need to make sure that we as an industry, like all those hopeful advertisers, are marching in one direction.
Yaroslav Kholod is director of programmatic operations at Admixer