Contextual Targeting Has Evolved, but Advertisers Haven’t Caught Up

Tim Cross 26 September, 2019 

The much debated ‘death’ of third party cookie targeting is predicted by some to give a new lease of life to contextual targeting, which doesn’t require third-party cookies to be effective. But Christian Dankl, co-founder and chairman of contextual advertising business Precise TV, says that since most of the industry has been so focussed on audience targeting for the past decade, many have misconceptions around what contextual targeting is actually capable of nowadays. In this Q&A, Dankl discusses what’s actually possible with contextual advertising, and explains his company’s ‘contextual DMP’.

Dankl will be on-stage at New Video Frontiers 2019 on October 16th discussing the topic: Context & Consent: Can Digital Advertising Cope Without the Cookie?

Do you think we’re going to see the death of the third-party cookie?

Yes, absolutely, although it has been slowly dying – we’ve seen Apple’s Intelligent Tracking Prevention (ITP) and Firefox’s Enhanced Tracking Protection (ETP) to block third party cookies by default and it is expected that Google Chrome is moving in that direction too. In addition to technical blocks, we have seen regulatory changes, such as GDPR and the upcoming California Consumer Privacy Act (CCPA) affecting the way we will target users going forward.

I think the reason the third-party cookie hasn’t disappeared yet is because the industry as a whole is so heavily reliant on it for getting targeting right, and the limited alternatives to third party cookies such as advanced contextual targeting.

How has contextual targeting evolved since the early days of digital advertising?

The problem I see is that because the industry was so reliant for the last fifteen years on audience-based targeting matching all sorts of third-party data with you first-party data and doing a lot of modelling – all of the innovation has been focussed on audience buying. There are some exceptions of course, such as the acquisition of contextual pioneer Grapeshot by Oracle for up to $400M last year.

If you think back to what contextual was like fifteen years ago, you were essentially buying broad categories, things like ‘sports’, and that was pretty much it. So you’d be buying everything from cricket to golf, to swimming, to NASCAR racing all together, and as you know that doesn’t represent a coherent audience group at all.

We really have to innovate and think differently, and our way of thinking differently is not just breaking it down into content taxonomy, but going deeper than that. It starts by essentially understanding the moment and the signals that are coming in. So as an example, think about coffee. If you’re just targeting ‘hot beverages’ as a category within ‘food and beverage’, that doesn’t really help you. If you want to target a barista type audience for example, that’s interested in artisan coffee and might buy a £2000 coffee machine, you’re not going to find that audience using the content taxonomy of coffee & hot beverages that’s out there. So you have to go deeper on page level or video level, and then create a predictive model that ranks the probability of that specific contextual content to the interests and affinities of your target audience.

For brands this means a lot of work and data in terms of getting the model right. You won’t see an impact if you start tomorrow doing contextual targeting based around the old taxonomy, there’s endless wastage there. That’s really the challenge, firstly to get the model right and secondly to keep optimising the deep learning algorithm. That’s where we come in and help our clients.

Precise has released a ‘Contextual DMP’ – what does that mean, and how similar is it to a standard DMP?

The first difference is that our contextual DMP is a non-PII DMP (it doesn’t contain any personally identifiable information such as cookies). That’s very important because we do a lot of work with toy and pharmaceutical brands, and take privacy extremely seriously, as a result we took the decision to open up to scrutiny by being third party audited for both COPPA (Children’s Online Privacy Protection Rule) and HIPAA (Health Insurance Portability and Accountability Act ) compliance. Under COPPA for example, you’re legally not allowed to store any personally identifiable information, or use it for modelling or targeting, on children’s content.

So what is in there? Essentially, you’re breaking down content into the most granular data you can get hold of. So with the coffee example, it’s not just food and beverage, you’re identifying content that resonates with coffee connoisseurs, and people who are into artisan coffee. You need to understand which correlations of content resonates with the audience. Maybe there are other interests, are they into a specific sport, or do they read specific blogs. So there is, for example, a correlation between people who are amateur high performance cyclists, show above average disposable income and have an interest in artisan coffee making, particularly around what beans and coffee machines can extract the right dose of caffeine, which based on scientific studies can boost your race performance by five percent on average.

What is predictive contextual targeting, and how is it different from traditional contextual targeting?

Traditional contextual marketing essentially used what you’ve learned about the past to help you make decisions about the presence. Predictive contextual targeting instead looks at the current moment and uses AI to predict the who is behind the screen. In the absence of behavioural data from a third party cookie, we have to use all the data we can get in that moment like context and live signals on who is likely to be behind the screen and what their thinking might be like – it’s all probabilistic, but with a very high confidence that it will resonate.

To train your model you need a feedback channel. Skippable ads or click-to-play video ads are perfect because if people don’t like it, they skip it, so you instantly know if you’ve got contextual targeting right or wrong. With skippable ads, if say forty percent choose to skip the ad, it means sixty percent of the audience is giving you valuable feedback. That makes a huge difference in our AI modelling.

YouTube was recently fined by the FTC for the handling of children’s data on its platform – is contextual targeting a solution to this problem?

Absolutely, Precise is certified as KidSafe COPPA compliant since 2017, so advertisers working with us can be confident they’re not using any PII relating to children for ad targeting. As part of the FTC settlement Google has agreed that from 2020 onwards, behavioural targeting such as keyword or affinity won’t be available on YouTube on any content identified as kids content. From 2020 onwards kids advertisers will have two main options for compliant YouTube ads, either contextual channel level targeting, or the more granular video-level contextual targeting, like Precise TV. From our experience the issue with channel-level targeting is that it’s too broad and you can’t be selective about which videos you align with – so a lot of the videos you end up running ads on may not actually resonate with your target audience.


About the Author:

Tim Cross is Assistant Editor at VideoWeek.
Go to Top