Signal measures the association between a correlation event and a goal event and quantifies the correlation between the two. This facilitates a deeper understanding of the behaviors that drive customer conversions, and can help guide product decisions.
Note: There are now more features available in Signal that are not highlighted in the video below. Refer to the Signal instructions to get current information on how to build Signal queries.
Signal Use Case
Using an a music sharing application as an example can highlight the value of quantifying correlations between events.
The music sharing app may want to understand the correlation between top events and users who purchase a song on the app. It is important to understand what the optimal actions that users take before purchasing a song are.
Building this query in Signal would involve selecting the target users, and how the "Song Purchased" goal event is correlated with the top events.
Values are returned after running the correlation. “Song Played” could have a strong positive correlation with purchasing a song. Most of the users who played a song later purchased a song.
This information can be used in future product decisions. By knowing that those who play songs are more likely to purchase songs, it is possible to build tools to encourage song plays. This could lead to a dramatic increase in the amount of users purchasing songs.
Signal Machine Learning Model
Signal calculates correlation using a well-known statistical algorithm called the phi coefficient. In addition to correlation, Mixpanel calculates what is called an opportunity score, and also presents a list of key findings.
The phi coefficient is a single number between -1 and 1 and it indicates how closely an event moves with your goal event. 1 means that 100% of all users that did the goal event also did the correlation event. -1 means that 0% of users that did the goal event also did the correlation event.
Opportunity is Mixpanel’s proprietary calculation that determines the level of importance of a given correlation. It pares down correlation results by eliminating false positives and highlights weak correlations. This additional assessment determines how much an event impacts conversion to the goal event. This reveals actionable opportunities to make product changes.
Signal also presents two key findings, rarity analysis results and conversion measurements. Rarity defines how common or uncommon it is for users to complete an individual event. The conversion measurements qualifies how likely a given conversion is to be helpful. For example, if all users are converting because the application forces this by design, then the conversion measurement will flag this as "unlikely to be useful".