- Past Event Highlights
- Article
Updating the TelevisaUnivision Household Graph for the Future
Sergey Fogelson, Ph.D. – VP, Head of Data Science, TelevisaUnivision
Pouya Tehrani, Ph.D. – Director of Data Science, TelevisaUnivision
Sergey Fogelson and Pouya Tehrani of TelevisaUnivision presented a household-level identity graph designed to more accurately identify and target U.S. Spanish-language media consumers, whose total addressable market (TAM) differs significantly from the general population. The system integrates extensive first-party data (e.g., direct-to-consumer touchpoints) with 10–15 third-party datasets (including demographic, behavioral and ad exposure data) to detect signals of Spanish-language media affinity (e.g., language settings, site visitation, cultural content consumption). The graph links identifiers (e.g., IPs, hashed emails, device IDs) to individuals and households, enabling cross-device and cross-platform targeting and measurement. The team rebuilt their graph using a Bayesian inference framework to unify disparate data sources more rigorously. Instead of sequentially adding data, they now combine all sources simultaneously, weighting inputs (e.g., prioritizing first-party data) to estimate the probability that identifiers belong to the same household. A key innovation is the creation of a continuous “Spanish-language media affinity score” (0–1) assigned to every U.S. household, allowing dynamic audience definition via thresholding (e.g., selecting households above a given likelihood). The system also incorporates large-scale behavioral datasets (covering ~90% of U.S. households across 35,000 domains) and expands identity resolution through improved IP matching (including IPv6), resulting in more flexible targeting, better coverage and enhanced campaign measurement and optimization. Key Takeaways:- Spanish-language media TAM is distinct and measurable, with ~60 million U.S. individuals of Hispanic descent (including ~40 million adults across ~20 million households) representing the core audience.
- Identity graphs enable cross-device household targeting, linking identifiers such as IP addresses, hashed emails and device IDs to individuals and households for activation and measurement.
- Bayesian data fusion improves accuracy, replacing sequential data integration with probabilistic weighting of multiple data sources to better resolve identity and household structure.
- Affinity scoring enables flexible audience sizing, with a 0–1 Spanish-language media affinity score, allowing marketers to dynamically expand or contract target audiences based on likelihood thresholds.
- Massive behavioral datasets enhance signal detection, including third-party data covering ~90% of U.S. households and activity across 35,000+ web domains.
- Improved identity resolution increases scale and precision, with expanded coverage of IP addresses (including IPv6) and hashed email identifiers improving match rates and programmatic activation.
- The graph supports full funnel use cases, including audience targeting, personalization, campaign measurement and post-campaign analysis comparing intended vs. actual reach.
