What is unique about the Semcasting analytical modeling process?
Semcasting employs a patented machine learning technology based on genetic algorithms that makes predictive model applicable and practical for every campaign. Models use all of our 750+ variables in an automated, iterative, learning process. We programmatically test each variable as an interaction with every other variable. 250 separate models per generation are created per generation. The most predictive interactions graduate to the next generation and are tested with 250 more variables. This process is repeated for 5,000 to 20,000 generations until the optimal combination of variables is identified. A total of at least 1.25 million separate models go into creating most models. The use of hundreds of variables rather than a half a dozen variables in the traditional regression modeling means that there is a higher likelihood that subtle predictors will be identified in variables and contribute to the overall accuracy. The process to build a campaign model is typically completed in less than a day.
Why is the Semcasting analytic process a better than the approach?
The benefit of modeling automation is that models can be mass produced for specific campaigns and local markets. Local level segment models can be generated to optimize for local geographic markets which are much more accurate and can be completed with no additional investment in time or resources. The Semcasting process means that separate DMA, State, County or zip code level models are created as part of the automation process. The averaging of scores effect of one-size fits-all national only models is avoided.
Is facebook microTargeting privacy friendly? Are we compliant with "Do Not Track, IAB and NAI standards?
As an organization Facebook is very concerned about protecting user privacy. As one of a select group of organizations that works with the Facebook Ads API we accept are also concerned about privacy and have in fact adopted policies which are compliant with industry standards and Facebook's internal policies.
IAB has provided the industry with guidelines on what is acceptable in term of cookie placement and tracking. Semcasting exceeds the requirement under this guideline. They establish that cookied browsers are required to be part of a "segment" of other like users who must remain unidentified. These unidentified users are tracked by cookie for as long as the cookie remains viable. Cookied segments of inferred interest are independent of location with longer the cookie being available, the more robust the profile of the individual. However, we believe the tracking of cookies as part of a segment is both inefficient and an inappropriate marketing methodology. Less than 40% cookie coverage and prospects qualified only by a inferred interest is structurally inefficient. The use of publically available external data is a more robust method for validating life stage, being in the market, and having the ability to pay. We qualify and rank locations, not individuals which means we are not stalking cookie traffic to derive interest.
Organizations and exchanges have popped up to profit by brokering these cookied segments – often without the knowledge or agreement of the cookied individuals themselves. Organizations with first party relationships with a customer (such as banks and catalogs) are now routinely brokering their unidentified, but prequalified cookie segments through third parties for a profit. This activity is under scrutiny by the FTC and industry groups and is, in part, the motivation of the "Do Not Track" advocates. Semcasting adamantly supports the regulation of third party data especially if the individual source has not had an easy mechanism for opt-out.
