Crowd Science Audience Modelling AI utilizes a combination of audience sampling, advanced machine learning techniques, survey and behavioural data collection to create the first ever audience prediction engine. The result is a highly effective and accurate audience based targeting solution.
How it works
By sampling and collecting a multitude of direct user response and observed behavioral data, the Crowd Science CITRUS segmentation engine is able to identify the underlying behavioral patterns and attributes (independent variables) of respondents who have declared in a survey to be a certain profile (dependent variables); for example, women, age 25-29, play golf. Each attribute is carefully analyzed and modeled using a machine learning technique.
Machine learning as described by Wikipedia.org is:
a branch of artificial intelligence, . . . a scientific discipline concerned with the design and development of algorithms that allow computers to evolve behaviors based on empirical data, such as . . . databases. A learner can take advantage of examples (data) to capture characteristics of interest of their unknown underlying probability distribution. Data can be seen as examples that illustrate relations between observed variables. A major focus of machine learning research is to automatically learn to recognize complex patterns and make intelligent decisions based on data.
In the case of Crowd Science methodology, survey responses and associated behavioral, technographic and other attributes form the seed group to generate a training set. This training set when analyzed through machine learning creates a generalized description or signature of a defined audience, which when coupled with real-time pattern matching, can classify online individuals at a high degree of accuracy. These learner training sets or seeding respondents, are a key ingredient to be able to go from a small subset of individuals, in some cases as few as 500, to identify online individuals in the millions in a non-obtrusive and completely anonymous fashion.
Targeting performance is a process of defining a prediction model that delivers the most effective allocation of ad impressions. In the most raw form, no targeting is the rate in which an audience exists on a web property. This rate is referred to as run-of-site or an index of 100. As targeting models are defined, they may take any number of degrees or efficiencies that provide a tradeoff with number of required impressions and true audience. As illustrated below, an index of 422, or similarly a lift of 322%, in plain English reads as for every impression where no targeting is being used, approximately 8 of every 10 impressions do not reach the intended audience; whereas, a lift of 322% in this scenario is successful in reaching the intended audience at least 8 out of 10 times.
Audience modelling provides a few unparalleled benefits for the publisher and advertiser:
- The only truly first-party data source
- Modelling derived from fully opt-in data collection
- Cookieless targeting capabilities