How is Audiense able to extract insights?
Our approach is to compliantly use publicly available information, using our own algorithms and third party sources to enrich that information to surface deep audience insights.
From what sources?
We gather data from a variety of social networks (Facebook, Instagram, Youtube,
Twitter) and public censuses on a granular and aggregate level. However as anyone who truly understands the Social Data industry Twitter is the largest open platform. For instance more than 200M instagram, facebook and youtube posts are tweeted every month.
Social Analytics vs Social Data
It is very common for Social Analytics to be used interchangeably with Social Data. Social Analytics is a means to provide performance analytics of brands on social. A few of the most common metrics are the number of fans, followers, impressions, etc. which are easily accessible regardless of network (including Facebook and Instagram).
Social Data on the other hand focuses on surfacing community insights, exploring conversations and/or campaigns. For example Unmetric is considered a social analytics tool exclusively whereas Audiense would be considered a Social Data tool, and then there are platforms such as Pulsar that do both.
Within Social Data we are able to differentiate between the type of social data under scrutiny whether it be Conversation Insights (Pulsar, Crimson or Brandwatch) and Audience Segmentation and Insights (Audiense).
Are your findings statistically significant if they are predominantly derived from Social Data?
Think of Audiense Insights as a survey targeting the most relevant people in your audience. If you are new to statistics you may be surprised to learn from a sample of 17,000 people you attain a margin error of less that 1% and confident level of 99% for any audience size. To learn more about statically significant samples we recommend you go to http://www.raosoft.com/samplesize.html. When you create audiences with Audiense Insights they usually made up of 250,000 people depending on the filters and the criteria applied which is far more than the required sample of 17,000.
It’s no secret that the use of social is more prominent within the younger generations. As a result when analyzing an Audience of over 60s it is essential to be diligent and cross-verify the insights extracted from a variety of sources.
It is important to take into account that the insights we provide are always compared to a baseline that is relevant to your audience. This comparison increases relevancy as any biased (social data) influences both datasets suggesting any over or under indexing attributes are highly specific and relevant insights for your audience.
Our audience coverage varies geographically but there is almost always a statistically significant sample of data to extract valuable insights from. We have developed functionalities that allow you to access all the available data in order to find a significant sample of your target audience.
The basis for generating insights
Generally, through machine learning, Audiense infers audience insights according to public interactions (tweets, follows, likes, etc.), the public information available within profiles of the users (name, profile photo, time zone, location), other public aggregated information (demographic census for specific segments based on surveys and other digital interactions) or a combination.
We/our partners (IBM Watson) train models with real examples. For instance, my profile bio or my handle can indicate what age range I fall into. After training the models with thousands of real examples we look out for correlations with other characteristics (it could be photos, relationships, names, public census, etc. ) in order to attribute age to individuals that haven’t been explicit. This is the basis for machine learning and those correlations can sometimes be common sense or explicit derivations whilst at other times be indirect correlations. Learn more by watching:
How each tab is calculated
There are certain variables which we/our partners train their algorithms with which are considered to be highly weighted but not exclusive to:
• Segmentation: relationships between people (ie: follow, reply, like)
• Auto-naming: profiles self description
• Demographics: photo profile, explicit locations, self descriptions, surveys, name, relationships, timezone, social posts.
• SocioEconomics: age, interests, self descriptions, public censuses
• Interests: social posts, relationships
• Content: social posts
• Personality: posts, psycholinguistics
• Buying Behaviour: social posts, psycholinguistics
• Online Habits: photo profile, explicit locations, self descriptions, surveys, name, relationships, timezone, social posts.
• Multi-channel - Activation: social posts, relationships
Other FAQ’s
How do you link profiles to different networks? The only compliant ways to do
this is
1) Gather information directly from the user with their consent
2) Utilise public data