How Can We Help?

Predict which Users will Click on Ads

In this use case example, we will be walking through how to predict the future behavior of users using a Machine Learning Pipeline. Specifically, we’ll cover how to predict which users are most likely to engage with ads.  This use case is relevant to any company that shows ads, and is especially useful if your ad sales teams would like to move beyond selling ad displays to selling ad clicks to brands. Typically user cohorts based on who is likely to engage with an ad will demand higher price-points than display-based cohorts.

What data do I need for this prediction?

The actions available for you to predict with a Future Events pipeline are based on the Event data that you are sending into Cortex. This sample prediction requires:

  • Ad Event Types: events representing when users engage with ads, e.g. impression, click, etc. Note that ad platforms like DFP typically make it relatively easy to export clicks and shown events.
  • Ad Event Categories: if the goal is to predict which users will click on specific types or categories of ads, that information should be sent over as an attribute to the corresponding ad event.

While the above are the only pieces of information required to set up our sample prediction, more data will typically lead to better predictive performance. Other information that can be used to build features for our pipeline include:

  • Additional User Behaviors (e.g. logins, clicks, pageviews, adds to cart, etc.) with additional metadata (e.g. What device is the user on? Where was the user referred from? etc.) which give more detail to the event. As an example of additional metadata for an event, an item category can passed along with the event in order to differentiate which purchases occurred specifically for the category of interest. Alternatively, item details can be sent in a separate feed as long as the item is identified within your purchase events with a shared ID.
  • User Attributes (e.g. demographics, loyalty status, etc.)

How do I predict Which Users will Click on Ads?

Step 1​: Choose Pipeline Type

Select ‘Create New Pipeline’ from within your Cortex account, and choose the Future Events pipeline type.

Step 2: Define Event

Future Events pipelines are used to predict the probability that some event happens in the future for each of your data points. Any future events prediction can be stated as a sentence as follows:

(A) happens (B) or more times where (C) within (D) days.

For our prediction, that reads:

An Ad Click happens 1 time where (optional) Category=”XYZ” within 7 days.

Here is how that will look in Cortex

Step 3: Define Groups

Sometimes you are looking to make a prediction for every user in your system, while other times you want a prediction only for a certain group of users. This prediction will be made for All Users, so nothing needs to be done for this step.

Step 4: Specify Settings

Settings is where you give a Name to your pipeline as well as the option to add Tags. Name and Tags are the two main ways to find your pipeline within Cortex after it has been created, so best practice is to use descriptive Names and Tags specific to each prediction.

In this example, the pipeline name is “Predict Ad Clicks” and the Tag “Ads” has been added as well.

Additionally, you can choose to run this pipeline only once, or have it rerun weekly. In this example, the pipeline is set to re-run every Sunday. This means that your pipeline will use the latest available data to retrain and re-generate up-to-date predictions on a weekly basis.

Step 5: Review

The final step is to review your pipeline and ensure all settings look accurate! If anything needs updated, simply go ‘Back’ in the workflow and update any step. Otherwise, click ‘Start Training’ and sit back while Cortex generates the predictions.

Those are all the steps necessary to create a future events pipeline that predicts which Users will click ads.

Related Links

Still have questions? Reach out to support@mparticle.com for more info!

Tags:
Table of Contents