Written by Hendrik Wistuba, Marketing Manager
How to approach brand risks in uncertain ad environments
Artificial intelligence has made its way into state-of-the-art brand safety & suitability solutions like Meetrics’. With these machine learning technologies, two fundamentally different approaches to deal with uncertainty emerge on the market. I’m going to discuss the philosophy behind each one and highlight the advantages of both of them. So, you know which best suits your advertising strategy.
But before that, a short overview of the best brand suitability practices for online advertising is needed. Known risks or desired media can be preselected with so-called block and target lists. Right after these, keywords which when found in an advertising environment identify this website or app as critical or unsafe are one of the most fundamental brand safety tools.
However, this only considers text-based content. Ad environments nowadays offer multiple forms of content at once: from written text over images to video or audio formats.
Ad verification vendors, hence, collect various signals on top of keywords prevelances to evaluate the brand-fit of websites and apps. Generally speaking, the more factors a brand suitability solution takes into account, the more sophisticated its overall assessment becomes. Here are four essentials, you should look for in a brand safety & suitability solution:
1. Semantic Analysis
Using machine learning models, single keyword silos can be expanded to recognise the multi-dimensional context of an article or content piece.
2. AI-based Image Classification
Artificial intelligence rapidly detects the imagery ad would display next to before they’re rendered in.
(3.) Sentiment Analysis
Computer systems may help to systematically understand the tonality of an article
4. Media and Journalistic Umbrella
To battle the recent spike in misformation, media experts, like NewsGuard, have begun categorising particularly critical or exemplary publications. As sentiment analysis is still in its infancy, this can be a much more reliable indicator of inciting rhetoric typical for unsafe ad environments
All these technologies add information to assess the brand suitability of a page. The interpretation of that data is where the two approaches, deterministic and probabilistic brand suitability, fundamentally differ from each other but more on that later.
Let’s look at an example of image classification first, to more easily understand the probabilistic nature of some technologies before going into greater detail on the advantages of each for online advertising.
The neural nets of leading ad verification providers such as Meetrics are trained to recognise other depictions. Images displayed in advertising environments are typically analysed by IAB or GARM categories, e.g. nudity, weapons, violence, drugs and pharmaceuticals, etc.
So what if an image is classified as the photo of a gun with 80% confidence?
The human eye can differentiate movie reviews and crime reports from legal safety violations within split seconds. Brand Suitability solutions automate this decision-making process with the help of other data points before the bid is placed or the ad is rendered on screen.
Therefore, AI-based image classification or any other technology mentioned above is by itself insufficient to determine the brand safety or suitability of an app or webpage. Other information must be taken into account. This is where a probability-based approach can make a difference.
Probabilistic brand suitability allows for a certain risk threshold in order to maximise advertising reach potentially harming the brand or supporting misleading or inciting publications.
Imagine the following case:
- Image classification states with 80% confidence that weapons are depicted
- Semantic analysis reports a high density of gun violence and crime-related keywords
- NewsGuard analysis recognises the website as a credible source of news content that among other things visibly differentiates between content and advert
Then, it’s safe to assume that the ad environment in question is that of a neutral and credible news report covering a violent crime. The potential risk to the advertising brand is undeniable but the additional reach may outweigh these concerns depending on the brand and its resilience.
Advertisers who like to take risks in order to address previously/otherwise unreached audiences could, although not recommended, opt for a probabilistic approach. The brand suitability criteria would then be softened beforehand allowing for the ad to run in such media.
The price for this increased advertising reach, however, could be immeasurable; especially in times in which a single screenshot can spread through social media like wildfire. Therefore, managers of particularly sensitive brands choose a deterministic approach instead. Any placement that could harm brand perception is then prevented even if the likelihood of negative backlash is slim to none.
In such cases, one or multiple factors would be used as knock-out criteria. Because every evaluation is based on multiple data points, the technology is highly customisable. The potential risk in reach can be minimised without putting your brand reputation at risk. In line with the earlier example, the consumer attention alongside tragic news content could still be utilised to communicate advertising messages in the context of objective reports without exposing the brand to any risks typically associated with more graphic news pieces. This deterministic brand suitability strategy has proven to be effective for many of our clients.
If you’d like to know more about opportunities like this and are interested in technologies like brand safety & suitability, subscribe to our blog via the link below.