The Behavioral Genome is a product of our thesis that businesses don’t require raw behavioral data to optimize performance or ROI. One of Yobi’s goals is to reduce the proliferation of identifiable consumer data. Rather than broker raw data, Yobi commercializes a non-identifiable, machine-readable representation of behavior. This representation maintains statistical equivalence with raw data without ever revealing personal information. As a result, businesses can integrate Yobi’s Behavioral Genome into all data acquisition and customer science capabilities.Request demo
Clean and Aggregate Data
For each of our datasets, we use machine learning to reduce the raw behavioral data to a representation of each identifier as a vector of numbers.
Capture Signal & Preserve Privacy
These numbers contain the statistical information needed to make predictions about future behavior, while revealing none of the details about past behavior.
Produce Safe Machine-Readable Data
This approach results in a representation that can be used in the models that we or our clients develop, but doesn’t make sense to a human and can only be interpreted in the context of other vectors generated by the same process.
Further Ethical Modeling
Our vectors can also be tuned to improve prediction of specific behaviors, and to remove information about protected characteristics to reduce risks of violating privacy or supporting discrimination.
We use custom probabilistic models to build connections across datasets. These models capture patterns in behavior that allow us to identify relationships between individuals and devices across datasets. However, rather than trying to resolve identity to a single person, we use these probabilistic models to maintain uncertainty over individual identities. This allows us to report aggregate statistics for a set of identifiers across datasets or produce representations for individual identifiers that draw on multiple datasets without explicitly reidentifying individuals.
We build predictive models of individual behavior using techniques from machine learning combined with insights from behavioral science. The resulting models are able to generate high-fidelity predictions about specific behaviors.