Mathieu Yarak, Group Senior Director of Data & Insights at Choueiri Group, shares his perspectives and insights on their use of AI and machine learning in advertising, life after third party cookies, and what DMS has in store for us for 2022.
We have been using AI and Machine Learning over the past year in different audience creation and activation strategies and one excellent example of their use is in the establishment of demo-led segments. At DMS, we represent more than 35+ publishers, the majority of whom are none-login or have shy login data, resulting in a demo audience offering that is quite limited and unscalable. Since the demand for these audiences by big CPG advertisers is quite high, we partnered and worked with 1PlusX, a Swiss German AI-powered DMP on their Demo AI models to create demo-led segments. Simply put, the model looks at the behavioral traits of users across publishers to determine, for instance, their gender. In 2021, we went through a testing period where we adapted their advanced AI models to our region and established an accuracy measurement system to measure its accuracy. As a result, we were able to increase our demo audiences from 2.5% to 33% (x16), with an accuracy of 72% for females and 75% for males (both higher than the benchmarks used in our markets). We can now proudly say that we have 33M addressable gender-based segments to be activated by our partners.
At DMS, we have been scoping and testing different solutions over the past year and recently onboarded IBM Watson contextual targeting as part of our partnership with Permutive, our DMP partner. IBM Watson is an AI contextual targeting solution based on NLP and works by crawling and classifying content based on different features such as categories, emotions, keywords, sentiment, and concept. The classifications are used to enrich the pageview events of users who have read articles on any given topic, which in turn are used to create targetable cohorts. Moreover, contextual targeting does not rely on 3rd party cookies or any personal information. The use of machine learning to improve the relevance of contextual segments ensures brand safety and brings us closer to brand suitability.
In addition, we have been in talks with different ID solutions, such as ID5, Neustar Fabrik ID, and UID 2.0 (through our partner TTD), each of which has different models and is based on different variables. Although Google seems to stand by its decision not to support or build alternative ID solutions to 3rd party cookies, the changes we are seeing from the king of data reveal that there is still much to discover moving forward.
We gather our users' behavioral data using Edge Computing technology, allowing us to process the data in real-time and most importantly, in a privacy compliant environment through Permutive. Diverse algorithms are used to create segments based on a four-tier granularity and the fun begins once we create an audience. Fueled by machine learning, we use audience discovery to uncover unique behaviors generated by the users within our ecosystem. For example, if we take the ‘Fashion Audience’, we can identify when they are active during the week, at what time of the day, their interests in terms of content, engagement level, content journey, and the list goes on. The granularity of insights is what allows us to optimize our activation from day 0 and thus drive superior outcomes.
2022 will be the year of data solutions for DMS, and we plan to focus on 3 solutions (alongside the AI-based Demo Audiences and AI-based Contextual Targeting mentioned earlier):
The objective of 0-party audiences is to create effective audience segments using data collected via surveys. This is done by bridging two state-of-the-art technologies: Qualtrics as a survey engine and Permutive as a DMP. The responses captured in the survey are moved through a bridge integration to our DMP where advanced analytics and LAL models are applied to scale the answers and create addressable audiences. We have tested this concept with Auto, Telco, F&B, and Tourism brands, where results showed an increase in both media and brand impact metrics. This year, our focus is to optimize this solution and activate it with our advertising partners.
The concept behind data clean rooms is mapping advertiser and publisher data in an encrypted, secure, and safe environment. We implemented our clean room ‘Permutive Vault' in Q2021, and initiated testing with a global automotive brand through their agency in Europe. This year, the focus is to provide advertisers with a safe passage for data partnerships.
Our top priority plan is building a data hub infrastructure fueled by advanced Analytics, AI, and Machine Learning. We have already started with ingesting 75TB worth of DMS users’ yearly level data, and overlaying it with various data streams such as DFP and GA. After testing one day’s worth of user level data (200GB), we recorded astonishing results: on a campaign optimization level, the performance increased by +450% during the last week where the learning was applied.