In the last few years, all major players within the ad tech community have noted the likely comeback of contextual advertising, caused by the privacy regulations and improvements in contextualized machine intelligence. Advertisers also find out that these ads can be more engaging for the audience because ad creatives closely match the surrounding content.
The contextualization available today can significantly improve the accuracy, relevancy, and usability of targeting models and ultimately drive greater ROI while ensuring brand safety and privacy of users. Let’s discuss how the untapped potential of these solutions can finally be utilized.
The second key pillar
Personalized advertising, which has been heavily embraced by digital marketers over the last decade, is based on the advertiser’s ability to deliver relevant, timely and engaging messages that resonate with users’ intents.
The main types of targeting within this approach are contextual and audience-based targeting. The former relies on matching ads with the page content, while with the latter, ads are displayed to particular users based on their online behaviour and/or socio-demographic data.
Up until now, the emphasis has been set on audience-based targeting via cookies. However, the privacy regulations, such as GDPR, CCPA, and phasing out of third-party cookies from the most popular browsers moved a scale towards contextual solutions. Contextual targeting uses first-party data and allows users to preserve data privacy.
By the means of web content categorization, contextual targeting is more likely to match with the current intents of users rather than on their past online behaviour. Ultimately, ads are served to the audience in the context that is particularly relevant and engaging for the brand.
Categorization and ML for better targeting
Contextual advertising existed long before the Internet; an example of this is seeing automotive ads when reading a car magazine. Today, due to advances in data science and computing, advertisers can move away from generic classification at the vertical level and dive deeper into the context.
From the technological standpoint, classification methods such as taxonomy are still at the heart of contextual targeting; however, their precision and granularity are significantly improved. For example, at MGID we use AI-based web content classification that detects nearly 500 unique content categories and can deploy additional options upon the request.
Sophisticated algorithms allow the classification of content types, topics, and other entities (people, products, etc.) and the sentiment analysis of the subjective information on the page. Thus, categorization and understanding of web content are achieved not just by keywords within the text information, but rather by high-level categories that take into account detail-level components and attributes. For example, advertisers can blacklist certain topics from the wheels of their campaign and target only pages with the most favourable sentiment.
Deep text classification and machine learning also allows one to discern patterns in past data, evaluate alternatives and make targeted recommendations for campaign optimization. To deliver relevant insights and make recommendations, data scientists can apply two basic models of prescriptive analytics, namely supervised and unsupervised machine learning.
In supervised learning, human expertise annotates source data, chooses labels and categories, which are then tuned via active learning and used by software to make predictions. In unsupervised learning, the software itself creates the system of categories from the source data and develops recommendations for the grouped cases. For now, the prevailing modelling choices of ad tech companies are to combine traditional classification methods such as taxonomy and supervised machine learning.
For now, both types of targeting, contextual and audience-based, are relevant in the advertising ecosystem. However, as it becomes easier to achieve a fine-grained classification of web content and ensure ads are accurately matched to the most relevant content, contextual is likely to challenge the primacy of cookie-based advertising.
Big tech and projects like IBM Watson, Google Data Studio, MonkeyLearn and MetaMind have developed the key resources needed for mass adoption – on-demand computing tools and open-source data. From here on after, the prime catalyst of contextual expansion is the ability to deliver purpose-suited control systems for campaign optimization and the advertising community’s desire to put new solutions in production.