Our proprietary matching and modeling platform is designed to match, model and score billions of records a day while maintaining significant control for our data scientists to refine, tweak and customize models where appropriate, allowing for the perfect balance of quality (precision) and quantity (speed).
HIGH FIDELITY MATCHING
High precision data matching is a prerequisite for high precision modeling.
We deploy a strict, cascading match process matching the cleanest records first – minimizes false positives and ensures the highest fidelity data matches.
We deploy a comprehensive, proprietary multi-lingual first name derivations dictionary to accurately associate different versions of the same first name with the correct individual – enhances match rates while ensuring accuracy.
RIGOROUS AND ROBUST SAMPLING
Rigorous accuracy and robust customization yield superior sample data sets.
We leverage up-to-date US Census data (and full vote history when building political models) to accurately weight model samples on a national, state, congressional district and/or DMA level – ensures the most balanced and representative sample data sets.
We leverage cutting edge sampling techniques and tailor our post-stratification strategy to the geography and survey at hand, with a robust variety of attributes and characteristics at our disposal – enables maximum accuracy and customization of sample set creation.
We leverage address and cell phone append data that is updated monthly – ensures the greatest possible accuracy which yields the most accurate models.
STATE-OF-THE-ART, HIGH TRANSPARENCY MODELING
A client-first model is always custom and always transparent.
We use open source modeling software and the latest modeling algorithms – enables state-of-the-art custom modeling and complete model performance transparency via our trademark infographics.
We use feature sets with the most up-to-date data along with custom engineered features – enables superior model resolution.
We use a pipeline modeling approach, not a one-size-fits-all modeling algorithm – enables cross validated modeling with 3+ algorithms simultaneously, and identification and selection of the best performing model based on the client’s definition of success for production.