Luminate Film & TV Streaming Viewership (M) models streaming consumption data.
Our streaming viewership modeling breaks new ground. In a fragmented entertainment landscape with no comprehensive source for first-party OTT viewership or demographic data, we’ve applied our expertise in data science to the challenge of streaming measurement.
In brief: SV(M) collects a proprietary, multi-source combination of data. Then, we apply a linear optimization model to calculate streaming viewership. We apply a second model to estimate demographic splits. We leverage Luminate Film & TV’s library of entertainment metadata to enhance our modeling and build a database of streaming titles. Our methodology allows us to lead the industry in reporting streaming ratings and publishing film and TV charts.
Read on to learn more about our data, our model and the methodology that brings it together.
SV(M) measures U.S.viewership for streaming TV shows and movies. Our metric is minutes watched per title: the total amount of time that viewers in our dataset spent streaming a movie, TV season or episode on an OTT streamer.
Our standard unit of viewing time is minutes watched. SV(M) permits users to toggle between Minutes and Hours as the display metric. (Hours watched = minutes watched divided by 60).
In addition to time watched, SV(M) estimates the total number of views for TV shows and movies by dividing minutes watched by the runtime. For TV shows, the Season Views metric estimates the number of views for the entire season.
In addition to minutes streamed, SV(M) models age and gender audience demographics. Title Dashboards include male/female percentage splits as well as viewership by age bracket.
Read more: SV(M) Model FAQs: Streaming demographic data
SV(M) publishes weekly charts that rank the Top 50 Movies and TV Shows based on minutes watched. Our charts help set a benchmark for streaming platform “ratings” as the industry continues to evolve.
Read more: Charts: Streaming film and TV rankings
We also track metadata—i.e. director/showrunner, production companies, release date, runtime and other identifying information—for the thousands of titles on the SV(M) platform.
Read more: Title Dashboards: Film and TV data at a glance
SV(M) incorporates multiple data sources into our consumption and demographic modeling: ACR panel reporting, proprietary web traffic and search data, internal Luminate metadata, and information sourced from vendors and partners across the entertainment industry.
Our most important source of consumption data is Automatic Content Recognition (ACR) data from smart TVs in the United States. ACR technology can identify the content being played on a smart TV or connected device (Apple TV, Roku, etc.) by analyzing the audio and video on screen. It works automatically to collect viewing information while keeping individual viewers anonymous.
Both SV(M) models — viewership in minutes and viewership by demographic —draw from panels of millions of ACR-enabled U.S. smart TVs. These panels have been normalized from a larger pool to accurately sample the U.S. population. The streaming viewership ACR panel comprises over 3 million devices. The demographic ACR panel comprises over 2 million devices representing 7 million people in a mix of single- and multi-person households.
ACR technology is platform-agnostic. The ACR panel data we receive includes viewing activity on all major U.S. on-demand streaming providers. SV(M) viewership data covers the 10 largest streamers by audience share, plus dozens of other OTT and FAST channels. We receive direct first-party viewership data from select streaming platforms, which we use to validate and enhance our modeled figures.
SV(M) also compiles aggregate viewership data and metadata for top streaming providers.
Read more: Streamers: Provider data at a glance
SV(M) automatically processes incoming data and matches it to titles in our database. Then, we implement a proprietary linear optimization model to calculate consumption by minutes watched. We use a second model, also proprietary, to estimate viewership demographics. Both models were developed and are maintained in-house by Luminate data scientists.
SV(M) automatically identifies content and matches entities in our dataset with Luminate Film & TV metadata. ID unification attaches metadata to each title in the SV(M) database and groups entities together (like TV seasons and episodes). This process builds our database and helps ensure that reported viewership data is correctly attributed to each movie, TV show and episode.
After ingesting and processing our dataset, we use mathematical modeling to predict and optimize viewership figures. The SV(M) model extrapolates total viewership from ACR panel data while correcting for anomalies, biases and gaps in reporting. The model is designed to normalize and address disparities in reported viewership. In some cases, it applies different weights to different sources.
We validate modeled figures against directly-reported viewing figures and carefully selected reference data (see below).
The SV(M) model augments and balances ACR panel data with alternative signals related to audience engagement—many of which are proprietary to Luminate.We track page view and search behavior from 378 million monthly active users across the Penske Media Corporation (PMC) portfolio, which includes Variety, The Hollywood Reporter, Deadline, IndieWire, Rolling Stone, Billboard, the Golden Globes, SXSW and dozens of other brands. We also collect web traffic, search, and reported viewership from publicly available sources.
We compare the above with actual reported consumption and, if needed, rebalance our model to reduce disparities. We re-train and adjust the model about once a month.
Our demographic model combines Bayesian statistics with historical consumption data, reputable open-source data and our own internal data. It draws from some of the same sources as the consumption model, but works differently.
We map ACR panel data to anonymized household data linked to connected TVs. Single-person households help “anchor” our demographic modeling by providing baseline consumption patterns for an age group or gender. We then apply a Bayesian likelihood framework to extrapolate consumption patterns within multi-person households.
More to know
Charts: Streaming film and TV rankings
SV(M) User Guide FAQs: Help with platform features
About SV(M): Platform overview and updates
Read next: SV(M) Model FAQs: Viewership, data sources and chart eligibility