Interviews, insight & analysis on digital media & marketing

The contextual hype cycle: sorting the pioneers from the pretenders

By GumGum EMEA MD Peter Wallace.

If Goldman Sachs injects $75m (£55m) into your business, it’s a pretty good signal that you’re on the right track. The investment bank’s vote of confidence in our machine learning-powered contextual ad targeting platform means we can scale up operations, helping even more advertisers ramp up the effectiveness of their online media buying.

But it’s also a sign that contextual advertising is on a distinct journey of its own, one which neatly mirrors Gartner’s Hype Cycle, a method of categorising the stages of innovation. It’s a process that has led to a lot of bandwagon-jumping amongst the ad tech community, with every vendor and his dog lately claiming to be able to offer contextual – the practice of aligning ads with surrounding content.

But what do they actually mean? My hunch is that there are a lot of very different definitions of this tech currently circulating in the market and it’s important for advertisers to understand the differences.

We’ve been developing this technology for twelve years so we’ve seen contextual already move through the first two Hype Cycle phases – the initial Innovation Trigger and the period of Peak Inflated Expectations that has followed. There are certainly great expectations of contextual, because it promises an effective way of targeting consumers with zero impact on their privacy; an enticing combination in an era when cookie-use is becoming ever more limited.

Not that contextual should just be seen as a handy way of getting around new privacy regulations. This is a technology that has been shown to be more effective than behavioural targeting – the current practice of serving ads to consumers based on their previous internet browsing. We believe that there is a great virtuous circle being created – contextual is user-centric because it is inherently respectful of users’ privacy, and is aligned with what they are viewing and engaged with in the moment.

Just one word

So far so promising. But contextual isn’t yet a golden land of opportunity, principally because its early incarnations have been shown to be simplistic at best. I’m talking here about the practice of analysing content based solely on tracking keywords in text. Once upon a time it seemed like a great way of auditing copy at scale. But just looking at single words in isolation is a very limited approach.

What happens, for example, when the tech surveys articles about basketball or photography, both very likely to contain the word ‘shot’? As keyword targeting has been seized on in recent years by agencies who want to guarantee brand safety to their clients, an over-zealous system would remove these articles from an inventory, erring on the side of caution in case the articles were referring to gun crime.

Clients lose out on perfectly decent ad spots, while publishers shed revenue. This became painfully apparent over the past year when content featuring words related to the pandemic was avoided by advertisers to such a degree that UK culture minister Oliver Dowden had to appeal to businesses to stop the practice to support the news media.

Keyword blocking has also effectively amped up discrimination across all forms of media as cautious advertisers attempt to steer their brands away from any topics deemed remotely ‘controversial’, including discussion of LGBT issues and Black Lives Matter protests. In a year when safety and risk became a primary concern for everybody, it seems that keyword blocking has allowed some in the ad community to take the desire to stay safe too far.

Next-gen contextual

Thankfully there is a way forward for contextual technology. Integrating AI and deep neural networks into the programme means that online content can be analysed in a more nuanced fashion. To take the example above, we can analyse the patterns of language to detect that an article containing the word ‘shot’ is actually about basketball rather than murder, potentially releasing it back into available inventory.

Alongside this ability to analyse text, GumGum’s contextual targeting solution, Verity, uses computer vision to understand the images and videos within a web page as well. Many contextual solutions would only look at the metadata behind them, which will often tell you very little about what it contains. Being able to accurately audit the visual element of content is only going to become more important as streaming services surge in popularity.

The upshot of this sophisticated level of detection is that contextual intelligence is effectively mirroring the way a human would scan content and make judgements from it, but at scale.

The Plateau of Productivity

Time and again in ad tech, we’ve seen trends that have gone through the Hype Cycle. At the peak of expectation, the market is saturated by new entrants, all claiming to be the next best thing. What follows is the realisation that platforms are not created equally, leading to consolidation and fallout of the companies which don’t create value.

This has happened with behavioural ad networks, audience targeting and now we’re seeing it with contextual, ID solutions and I’ve no doubt Connected TV (CTV) will follow suit.

The hype around contextual is warranted, the proof points are there in performance, scale and stability in a market which has been tossed and turned by changing data privacy. What we need now is more transparency and understanding of the technology that each new entrant offers, along with verification such as the Media Rating Council (MRC) that gives an independent validation of quality.

Once this has happened we can quickly move into the Plateau of Productivity and continue to develop contextual as the market-leading targeting solution that it is.