There is a lot to learn looking at user-level behavioral data. Lytics currently studies behavioral data automatically in real-time and reports six scores for each user1.
These six scores each represent a distinct behavioral quality and can be composed to build rich audiences.
Scores as seen on a user profile.
Note: If you are already a customer of Lytics, you can see the following explanation of each score alongside your own data on the Learn Scoring page.
Quantity measures a user's cumulative activity over their lifetime of brand engagement. The more activity the user registers, the higher the score. This score measures the user relative to all other users.
It is a common tactic to target a user based on the number of times the user has visited a website, or has performed some other behavior. This becomes increasingly difficult as marketers add more data sources to their stack, and the user continues to engage over time.
Quantity takes into account a user's behavior on all data sources and measures that relative to how the most and least active users are engaging — so a user's score will always be between 0 and 100. You can think of this as a test score.
What is the point of scoring between 0 and 100? This is how we can ensure that any audience created with a score will always stay relevant. Perhaps 1,000 page visits seems like a lot for a user now, but the number will only grow larger as your site grows older and the amount of content you have increases. Another benefit of having a bounded score range is the ability to see the complete distribution of all users.
This is an example of how scores look like across an entire audience. The x-axis is the score (ranging from 0 to 100; 5 to 95 in the example for clarity) and the y-axis is the number of people who have that value as their score.
Recency measures how recently the user's general interaction has been. More recent activity means a higher score. This score measures the user's recent activity relative to the user's past activity.
Without scoring, this would be achieved by looking at the last time a user visited. Although better than nothing, that approach is kind of crude. Maybe an at-risk user opened your email by accident? It'd be an expensive oversight to assume that the user had recent activity and didn't need any nurturing.
Here is the distribution of users by their recency. If this distribution is bimodal, it means your users have all either visited recently or they have abandoned completely.
Frequency measures how consistent a user is over time in interacting with your brand. More frequent interactions means a higher score. This score measures the user relative to all other users.
This serves as a measurement of user regularity. Do they visit once a week? Once a day? Once a lifetime?
Since this score is relative to all your users, you can easily target your most frequent users, rather than something like "users who visited in the last week", which will vary wildly in size.
Again, this score has a fixed range of 0 to 100. All the scores are like this. It is how we can continuously update user profiles without having to update audience definitions.
Intensity measures the depth of a user's typical interaction with your brand. More sustained intense/deep usage means a higher score. This score measures the user relative to all other users.
The behavior a user exhibits during a single session is very telling of them as a consumer. If they have high interaction in a session (high intensity) they are more likely to be a deeper researcher or more curious. If they have low interaction in a session (low intensity) they are more likely to be casually browsing or engaged with a certain piece of content, but not your overall brand.
Momentum measures the rate at which users are interacting with your brand. Users who are interacting more than usual with your brand will have a higher score. This score measures the user's recent activity relative to the user's past activity.
It's easy to confuse how momentum and recency differ, but they are actually very different. Universally speaking, we've found them to have a 5% correlation. Recency measures absolute recency of activity, but just because a user has recent activity, doesn't mean they're not at risk of churning.
If a user maintains a constant rate of activity, their momentum score will be 50. If they are more active than they used to be, their momentum will be greater than 50, and might warrant a loyalty offer. If they are less active than they used to be, their momentum will be less than 50, and might warrant a win-back campaign.
Propensity predicts how likely a user is to return with subsequent activity. Users exhibiting positive interaction patterns are more likely to return and have higher scores. This score measures the user relative to all other users.
There are many reasons why users churn — changing interests, competition from competitors, bad experience, etc. — but from a data perspective, attrition of any kind starts to look similar.
Propensity employs an ensemble of statistical models to identify any patterns it can find for detecting how and when attrition starts to occur. With time, it's able to find more patterns in your data and become increasingly accurate in identifying when users start to exhibit those behaviors.
Each Lytics score is accessible as a Custom Rule in the Segment Builder. They can be added to any segment definition as an intelligent filter when the size of the segment is larger than desired.
Example: When crafting a segment to be used to buy ads against, the size of the segment is critical. The size can be arbitrarily shrunk by taking 10% of the matching users, or it can be intelligently shrunk by creating a threshold with a Lytics score such as Propensity or Momentum. This way, the best fit users remain.
Interfacing with scores directly can be difficult. They are low-level building blocks that require expertise to use to their fullest. In addition to these scores, Lytics offers out-of-the-box behavioral segments that use scoring under the hood.
Customizing Data Sources
Lytics scores work best on behavioral data — that is, data that was generated by a user, like a web view or email open, rather than on list imports or other non-behavioral data.
Every supported 3rd party integration is already configured to include only behavioral data in user scores. By default, any custom integration is assumed to be non-behavioral. If you're using a custom integration that you want to contribute to behavioral scores, you'll need to update the
scoring_whitelist property in your account settings, which you can do directly via the Account API, or contact your account representative to update the setting.
- Users must have behavioral data to make confident measurements before they have scores. If a user was added to Lytics via email upload, for instance, they would have no scores.