How do you measure streaming viewership?
We’ve received a number of questions from SV(M) users about the platform and our methodology. Here’s a big one:
The (M) in SV(M) stands for (Modeled). At this time, there is no reliable source for first-party streaming data. Unlike TV networks, streaming platforms do not release daily ratings.
Instead, Luminate applies our expertise in data science to the challenge of streaming measurement. We ingest streaming data from multiple sources and use mathematical modeling to calculate viewership. Published SV(M) viewing data has been derived from our model.
Read on for answers about the SV(M) model, our data and our approach to calculating streaming viewership. If you still have a question we haven’t answered, ask us!
What do you measure?
SV(M) measures U.S. streaming viewership, in minutes watched, for streaming TV shows and movies. We use a second model to calculate viewership demographics.
We track viewership for on-demand streaming platforms, including (but not limited to) AMC+, Apple TV+, Discovery+, Disney+, Hulu, Max, Netflix, Paramount+, Peacock, Prime Video and Tubi.
Streaming original viewership data in SV(M) goes back to 1/1/2022.
Data tracking for library titles may vary. We are continually adding new titles to our dataset.
No. Modeled streaming viewership data cannot be backfilled, restated or "added in later."
Automatic Content Recognition (ACR) panel data from over 3 million U.S. smart TVs
Proprietary search and page view data from 375+ million monthly users on PMC domains
Public-source page view data
Data provided by streamers, including some directly reported viewership numbers
Automatic Content Recognition (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.
Luminate receives ACR data from a panel of over 3 million U.S. smart TVs. This panel has been normalized from a larger pool to accurately sample the U.S. viewing population.
ACR panel data is anonymous. The smart TVs in our dataset are not linked to streaming accounts, users or locations. All TVs within a household are considered distinct and separate sources of viewing minutes.
No. SV(M) does not currently track mobile, tablet, laptop or other portable device viewing. Only smart TVs and connected devices are included in our ACR panel.
No. SV(M) only includes U.S. streaming data at this time.
Our data includes international titles, but all viewership is captured on smart TVs located within the United States.
Yes. We receive some viewership data directly from streaming platforms, which we use to validate and improve our model. When possible, we include these figures in our modeled viewership totals. We do not report direct data separately.
“Views” or “Total Views” estimates the number of times a title was viewed in its entirety, either during your selected time frame or as a cumulative figure (listed as ATD on Title Dashboards). We calculate Views based on duration, which can refer to the runtime of a movie, of a single TV episode or of a whole TV season.
There are four Views metrics:
Total Views (movies) = Minutes watched divided by the movie runtime.
Episode Views (TV) = Minutes watched divided by the episode runtime.
Season Views (TV) = Total minutes watched for the season divided by combined duration of available episodes within the time frame.
Series Views (TV) = Sum of Season Views within the time frame.
We calculate views based on duration (runtime), which we receive as metadata attached to a title. If we don’t receive duration data, Dashboards will display an error message in the Views box (“cannot be calculated” or “not available”).
Over 80% of titles in SV(M) have duration metadata attached; this issue is increasingly rare as we continue to improve our metadata coverage.
To calculate Total Views and Episode Views, we simply divide the runtime of a movie or episode by minutes watched.
Season Views are more complex, however. A “Season Views” metric estimates how many times all available episodes of a TV season were watched in a given time frame — more like Complete Season Views, in other words. It’s not a sum of Episode Views, because one-off episode-level viewership doesn’t necessarily equate to whole-season viewership.
To calculate Season Views, we divide season-level minutes watched by the duration (in minutes) of the season to date. For example, if 5 episodes totaling 140 minutes were available to stream during your selected week, Season Views = Minutes Watched divided by 140.
Our model is optimized for streaming originals: titles that premiere on streaming platforms and are exclusively available to watch on streaming. Titles that air on linear networks at the same time (i.e. linear TV shows with next-day streaming) may show inconsistent streaming data.
Live-on-OTT content is difficult to sample accurately. ACR panel sampling works best when it captures a large, broad dataset each day. The short window in which to record "live" views makes it difficult to model live viewership. Modeled viewership figures for titles that originally aired live on a streaming platform may be incomplete OR may combine live views with non-live views of the same title (i.e. after it was made available on streaming).
We map ACR panel data to anonymized household data linked to devices. Then, we use modeling to estimate which household member(s) are streaming a TV show or movie. Our methodology combines a Bayesian likelihood framework with historical consumption data, reputable open-source data and our own internal data.
The demographic panel comprises over 2 million smart TVs and covers 7 million U.S. adults across single- and multi-person households. Like the viewership panel, the demographic panel has been normalized from a larger pool of devices to accurately sample the U.S. viewing population.
Yes. Our demographic data comes from some of the same devices as our viewership ACR panel, but we developed a new mathematical model to calculate age and gender breakdowns.
No. Household demographics and individual user data are anonymized.
We report a confidence score that indicates the reliability and probable accuracy of our modeled metrics. Confidence level reflects the amount of data we receive for a title, amount of geographic coverage, consistency in daily modeled ratios (indicating model stability) and adjustment for households with children (since we don’t gather data on children).
High confidence for a title usually means that we have a significant amount of viewership data with a balanced geographic distribution resulting in consistent outputs for the selected time frame. Current and recent titles generally earn higher confidence scores, especially during the release window. When more people stream something around the same time, we receive more data points to stabilize and validate our model.
When fewer people watch something around the same time, it becomes more difficult to model demographic metrics. Low confidence for a title can mean that we did not receive enough raw data to provide a stable breakdown.
Our streaming data is attached to devices, not to people. We use single-person households to “anchor” our demographic modeling by providing baseline consumption patterns for an age group or gender. We then apply a Bayesian likelihood framework to estimate household members’ preferences for certain titles based on demographics.
For example, if a title is overwhelmingly watched by men aged 25-34 in single-person households, we can estimate that that title is being watched by men aged 25-34 in multi-person households, too.
No. We receive viewership data attached to devices, not to individuals. We then assign viewing minutes on a device to the likeliest demographic within a household based on our model. We don’t “double-count” or break out fractional demographics from potential multi-person viewing.
SV(M) includes access to Luminate’s Top 50 Movies (M) and Top 50 TV Shows (M) charts. These are weekly charts ranking the most-watched streaming original movies and TV shows.
Read more about our charts here.
The SV(M) charts rank streaming original content, which we define as follows:
TV series and movies identified as “originals” by the distributing streaming platform.
Ex. Glass Onion, Stranger Things, Andor
Films released theatrically that have a first-run SVOD streaming window within three (3) months of release on a vertically aligned streaming platform.
Ex. Aquaman and the Lost Kingdom, Wonka
Library titles are not eligible for our streaming charts, but they can be included in Ranking Reports and Streamer Dashboards. They may also appear as related titles elsewhere on the platform.
Luminate provides the modeled data and calculations used to create the Variety Streaming Originals Film and Television charts. Variety applies their own eligibility rules, which sometimes include or exclude specific titles.
Variety’s film chart consists strictly of streaming exclusive films, while the Luminate Top 50 Movies (M) chart includes films that premiered on streaming within 90 days of a theatrical release.