This article is about guiding you through the creation of aggregated events.
What is an aggregated event?
In this article, we've learned about Aggregations and how to build aggregations.
To use these data points in a model, we want to know what is the optimal value of an aggregation.
Let's take an example: your common sense would tell you that the more a user logs into your app, the more it is likely to convert. But in reality, this is not a linear correlation. There is a threshold you will want to find that corresponds to the optimal value of times a user logs in per day/month/quarter, which indicates if they are more likely to convert.
- the Aggregation you'll start from: email_number_login_last_30_days.
Let's say we find the optimal value is 6.
Hmm wait, where does this 6 come from? From a frequency analysis you'll create when building the aggregated events (see next steps)!
- the Aggregated event you'll use in the model to assign points to: email_number_login_last_30_days >= 6
Uh wait, why can't I use a linear function in the model like:
score = email_number_login_last_30_days * points?
Precisely because this is not a linear correlation, which means there is a point at which logging in more than 6 days a month does not make a difference in terms of impact on conversion than just logging in 6 days a month. Therefore, no need to score them higher.
Aggregations cannot directly be used in the behavioral models because aggregations can take multiple values and have no timestamp, whereas the models use events with timestamps. Therefore the aggregation needs to be transformed into a boolean (has performed X or not), and with a timestamp (has performed X on that date). These are the aggregated events.
These aggregated events will be displayed in the "Event weights" and can be given a weight in the model.
- It's possible to create aggregated events based on email level aggregations only if the model you're working on is a PQL type of model.
- It's possible to create aggregated events based on account level aggregations only if the model you're working on is a MQA type of model.
How to build an Aggregated event?
- For Pro and Enterprise plan only
- You have an Admin or Architect role in MadKudu
- You know what a Aggregation is
- You have built an Aggregation
- You know what the Event mapping is
Before starting ...
Make sure you have followed this instruction to create the Aggregation.
Step 1: Create an Aggregated event
- Go to the Data Studio (studio.madkudu.com or through app.madkudu.com > Predictions > Data Studio)
- Click on Aggregations
- Click on Analyze from the Aggregation you will be using
- Select the Aggregated events to create based on the recommendation of the frequency analysis
- Frequency analysis
The frequency analysis graph helps to identify the optimal number(s) of occurrences that distinguishes the people who are more likely to convert than the ones who are less likely to convert, based on the aggregation you’ve selected. The recommendation will be to create aggregated event(s) for those optimal number(s) of occurrences. The optimal number(s) of occurrences have the highest correlation strength factor.
- The strength factor ranges from −1 to +1, where
- negative numbers (-1 < phi < 0) represent a negative relationship (red bars in chart)
- zero (phi = 0) represents no relationship
- positive numbers (0 < phi < 1) represent a positive relationship (green bars in chart)
- Frequency analysis
- Click on Save
Step 2: Deploy the Aggregated events
Now you should be back on the Aggregated events list page.
You have just created an aggregated event, which means you have added it to the library of aggregations.
Now to use the aggregated event in your model:
- Click on Deploy
- in the model you'd like to use the aggregation, in the Overview, click on Reload aggregations
Step 3: Use the Aggregated event in the Likelihood to Buy
See how to assign points to events when building a Likelihood to Buy model in this article.
- Email-level aggregated events can only be used in a Lead level LTB (PQL)
- Account-level aggregated events can only be used in an Account level LTB (MQA).