The advent of artificial intelligence in consumer devices is creating higher expectations from customers. We are in an age where everything has to be relevant, requiring today's marketers be "more." More efficient. More effective. More innovative.
In turn, marketers are looking to machine learning and programmatic as the media solution to meet expectations and deliver relevant brand messages to the right people, in the right context, and at the right moment that matters most. Luckily, the flood of smart devices in-market today are generating an enormous body of data that can power the machine learning needed to develop truly adaptive media solutions.
Here at Oath, our passion is developing new ways to enhance our platforms and arm our partners with the tools they need to efficiently and effectively hit business goals while uncovering true advertiser intelligence. That's why we are excited to announce the highly-anticipated addition of our machine learning-based optimization and forecasting system, AdLearn and Predictive Audiences in the BrightRoll DSP. The combination of these two features, along with more than 165B daily data points of highly accurate Oath data, will help turbocharge the performance of your programmatic campaigns.
What is AdLearn?
Celebrating its 20th anniversary, AdLearn has remained at the forefront of innovation. Now in BrightRoll, it has evolved to become the most advanced advertising optimization and bid management system in the industry. Fueled by a state-of-the-art machine learning system with patented Value Maximization technology, AdLearn is the prime optimization engine combining demand and supply data with predictive performance algorithms to connect the best ad with the right user and placement, at scale.
How does it work?
All of the above components involve critical machine learning algorithms, many of which are patented, that are combined in a unique and proprietary configuration that works to ensure every impression is a smart impression.
With the integration of AdLearn into the BrightRoll DSP, we set out to improve performance - and we did it! Initial results show a 48% increase in conversion and a 38% decrease in cost-per-click. Additionally, we saw a decrease in both eCPM and vCPM goal types by 20-25%, demonstrating AdLearn's ability to help brands meet upper and lower funnel goals.
The heart of AdLearn consists of four main components that adapt to the campaign as it runs. These components work together to take into account over 2,500 different factors—including demographics, user behavior and other visitation and performance statistics—to determine the best strategy and optimal bidding tactics to deliver against campaign goals.
Performance Prediction – Estimates the KPI rates (CTR, CVR, IVR, etc.) per impression.
- Control System – Maximizes ROI while meeting pacing and performance constraints by computing campaign-level bid adjustments.
- Forecasting – Estimates properties of the campaign's price-volume curve, which is used to turbo-charge the control system to maximize efficiency.
- Bidding – Combines performance predictions and information from the forecasting system to enable optimal bidding
What are Predictive Audiences?
For advertisers with performance goals, understanding how to leverage data to reach the audiences more likely to convert can be a daunting task. While remarketing and lookalike segmentation are powerful tools, they are focused on what a person has done in the past. This can be helpful for finding specific users or expanding qualified reach, but these tactics often fall flat when it comes to measurable performance. That's where predictive audiences come in.
Leveraging key algorithms from AdLearn and insights on over 1B Oath users, Predictive Audiences use machine learning and probabilistic equations to build a quantitative model that analyzes your converting users in over 1 million dimensions to find the most accurate predictors of purchase and models the rest of Oath's 1B users against those predictors. The result is high performing, customizable audience segments that tie directly into the AdLearn algorithm.
In consolidating Predictive Audiences into the BrightRoll DSP, our engineers took the data science principles from ONE by AOL, incorporated them into a new machine learning platform, and added the full breadth and depth of Oath's data signals. Predictive Audiences combine industry-leading performance with the ease of Audience Builder's self-service flexibility, so that clients can convert the exact type of consumers they need to reach direct-response goals.
The results across beta campaigns have been impressive. When compared to line items using other audience modeling, Predictive Audiences had 4-8X lower eCPA across platforms and verticals.
How do they work?
Predictive Audiences are built using a seed of consumers who have taken a required action. Our model then examines over 1M attributes within that seed to create a predictive audience. Our machine learning engine finds the top correlated data points which predict conversion. We then score each Oath user individually against these "predicted" activities which lead to conversion. For users who score positively, we bucket them into eight tiers, each tier corresponding to their probability to convert on a specific event in the near future, whether that's downloading an app, filling out a form, adding an item to the cart or making a purchase. This tiered model approach allows advertisers to calibrate the right precision versus scale to reach campaign goals.
In order to account for the changing needs of consumers, segment scoring is refreshed daily. That way, a person who just bought a pair of running shoes is not served shoe ads for weeks to come.
During a performance campaign using Predictive Audiences, BrightRoll will leverage the identified users within the tiers, and calculate the optimal bid within the bidding parameters set for that line item.
With AdLearn working in conjunction with Predictive Audiences, we have seen exceptional performance enhancements for our clients across multiple verticals.
At Oath, we understand that every brand has campaign goals specific to their business. We are committed to developing tools that blend great data with intelligent algorithms to help you achieve the objectives that matter most.