Home Analysis Contextual targeting as a replacement for audience-based targeting

Contextual targeting as a replacement for audience-based targeting

Photo by Nicolas J Leclerq on Unsplash
Share on

Predictive Audiences from Comscore is an advanced contextual targeting solution for connected TV and other digital video that is designed to match the effectiveness of audience-based targeting. It does not rely on third-party cookies, is GDPR-compliant and can be used by advertisers to find their target audience by first understanding, in great detail, what ‘lookalike’ consumers are reading online or watching on television.

Predictive Audiences is being made available in Europe, with the UK and Germany already launched. ACR-based viewing insights from connected TVs are one of the data inputs, initially harnessing the partnership with Samba TV that was announced in September. Videonet recently hosted a webcast that explores the solution in detail, which you can listen to here.

Predictive Audiences can be used for digital video across all devices, including tablets and smartphones, but is being hailed as a particular gain for the connected TV marketplace. It can be used when buying directly from a television/premium video sales house or programmatically (through private marketplaces that can be set up in DSPs and SSPs, and via an open exchange using pre-bid). The solution can be applied to advertising that will be inserted into live or on-demand content.

This new product draws upon large, opted-in Comscore panels where the desktop and mobile behaviours of consumers can be tracked, showing the kind of household they are, thanks to lifestyles, purchase decisions and buying intent, etc. Thus, a heavy international traveller can be identified by their regular purchase of airline tickets and hotel stays, for example. Comscore has around 180,000 digital panellists in the EU, including approx. 60,000 digital panellists and 8,000 mobile panellists in Germany.

Comscore also has access to viewing and advertising exposure data from around 900,000 connected TV homes in the EU, as well, thanks to the partnership with Samba TV. That is made up of approx. 500,000 UK homes and 400,000 in Germany.

The viewing is measured using automatic content recognition (ACR), the technology that identifies viewing that appears on the television glass, no matter what input source is being used (e.g. set-top box, streamer box or direct from the Smart TV itself, which also means the viewing could be broadcast or streamed). Spain will be added to the connected TV samples, with Q1 2021 as the target date. Other European markets will follow later.

When advertisers want to find their target audience via connected TV (or other digital video), Predictive Audiences takes the advertiser’s target audience characteristics and finds those kinds of people from within the Comscore (opted-in) panels. In effect, it finds lookalikes to the audience the advertiser seeks.

Because Comscore can also track the media consumption of the digital panellists (who are all opted-in) they can see the kind of content that a given target audience consumes. Predictive Audiences can therefore cross reference a lookalike group to their media consumption. Rachel Gantz, General Manager, Activation Solutions at Comscore, explains on the webcast: “The content relationships could be quite random – perhaps consumption of personal finance is a predictor of someone being a heavy traveller.”

Crucially, this media consumption can be identified and understood at a very granular level thanks to the use of proprietary Comscore technologies. It is the contextual detail, sub-divided into an astonishing 350,000 content categories, that enables the accurate cookie-free targeting.

For text-based web content, Comscore’s web crawler analyses all words on a page, including within sidebars, as well as any metadata associated with the web page. For television/digital video viewing, Comscore uses metadata from the broadcasters/publishers but also advanced visual recognition and then AI-based video analysis tools that provide information about the subject within shows down to video frame level.

Predictive Audiences understands the context of programming in such detail that an advertiser can look beyond blanket labels like ‘news’ to see that a programme under the news genre actually contains regular cooking and fashion segments that might be suitable inventory.

Brand safety is one of the big wins with Predictive Audiences, Comscore believes. As Gantz notes on the webcast: “One of the approaches to brand-safety in connected TV is to judge it at app-level but that will not tell you much about the content itself, and there is so much content within an app.” Broadcasters can provide metadata about shows and title-level insights about the content, but there is nothing like the detail Comscore claims for predictive Audiences. Title-based targeting is not available via DSPs as a general rule, Gantz adds.

Referring to the news show with a cooking and fashion segment, she says, “These are advertising opportunities, like for a retailer, within content that would be missed if the content was classified as a news show.”

This deep contextual understanding means that Covid-related content can be divided into sub-categories, with advertisers able to distinguish between infection counts and quarantine cooking tips, for example. It allows advertisers to take a more nuanced approach to brand-safety. “If you are only using brand safety ‘on’ or ‘off’ then you could miss opportunities,” Gantz suggests.

“This puts more control in the hands of the advertiser when deciding how and where they want their ads to appear. It helps the buy-side get visibility into exactly what they are buying.”

Metadata alone is not sufficient for detailed contextual targeting, according to Comscore. “From a connected TV standpoint, the metadata available via SSPs and DSPs is ‘spotty’ at best, sometimes because publishers do not want to share too much information about the content, Gantz declares. “Publishers can classify content differently,” she adds, highlighting a practical barrier to relying on metadata when comparing content.

“There are lots of holes in metadata. You cannot rely on metadata for a fully accurate presentation of the content,” Gantz argues.

Once you know the content that indexes very high for the target group, the next step is to buy it. This can be done programmatically via DSP integrations.

Comscore has already declared Predictive Audiences a success. “It is pretty robust lookalike modelling. We know that contextual categories are a predictor of an observed behaviour. The key is the very granular indexing; that is why this solution can be cookie-free,” Gantz explains. The accuracy of Predictive Audiences has been validated during testing by Comscore. “That has been a very important focus for our team: we have tested and tested.”

Gantz confirms that the sample sizes on the Comscore panels are sufficiently large that they remain representative even given the granularity of the contextual categorization (350,000 sub-divisions). Pointing out that Comscore is a measurement company, she declares: “We take lots of pride in our methodology and its robustness. Everything we do needs samples that are sufficiently large that they are accurate and representative.”

Gantz believes Predictive Audiences is good for the sell-side too, because it helps them monetise more inventory, including with direct buys. Sales houses can demonstrate what the content is in more detail. The cooking and fashion segments in what would otherwise be considered a news show are an obvious example.

Gantz calls it a win-win solution, and Comscore is talking with major media owners where the new solution is being made available. “Publishers can ingest this data directly and offer that to advertisers as a service,” she says.

The insights enabled by Predictive Audiences can also be used for first-party cookie enrichment at a time when first-party cookies are about to become more important, due to the demise of third-party cookies.

Comscore provided a set of audience-based targeting solutions before GDPR effectively forced the company to remove its audience insights from the European market. Predictive Audiences is the replacement. Rather like classic television planning, Predictive Audiences uses content as a predictor of where the target consumer will be, only with more contextual knowledge. This contrasts with audience-based targeting, where the buyer knows that their target consumer is definitely watching a particular piece of content right now, and where the content itself (the context) is not necessarily relevant to the trade.

“We are really excited to be able to target connected TV viewers in a GDPR-compliant manner,” Gantz confirmed on the webcast. With connected TV one of the winners from the lockdowns, and advertiser interest in this channel increasing, this product arrives at a good time. “This is designed to be as accurate and as strong as the audience-based targeting it is designed to replicate,” Gantz concludes.

More information

The Predictive Audiences webcast includes a 25 minute presentation, with slide-deck, followed by a comprehensive audience Q&A including questions from senior buy-side executives. You can listen (free) by registering here. The speakers are:

  • The Rachel Gantz, General Manager, Activation Solutions, Comscore
  • Guido Fambach, EVP, EMEA & APAC, Comscore

Share on