Improve Targeting with Predictive Models
Analytical Targeting is a feature soon to come to SAS Customer Intelligence 360 to apply machine learning methods to help decide the targeting of customers. I would like to walk you through a real-life example on how this feature creates value, but first, let me summarize the logic behind this feature.
The goal with Analytical Targeting is to target people who have the highest likelihood to convert based on the primary metric defined in a task (which is click-through by default). Targeting users who are most likely to convert improves the task conversation rate by excluding the members of the target audience who are unlikely to convert – and thereby freeing them up for other more relevant messages. The predictive models used in Analytical Targeting are built and deployed automatically with minimal input from the user.
There are two simple steps to deploying Analytical Targeting:
- Enabling the modelling: Calculating the model scores starts. The model is not used yet.
- The "model fitness" symbol will indicate the performance of the model so the users can decide whether to deploy the model or not.
- Also, the "cumulative gains chart" is populated indicating the lift in conversions of the current model over the "baseline". The cumulative gains chart can help users select a cutoff threshold that corresponds to a desirable gain.
- Active targeting: A cut-off is set, and the model scores are used to target users who are the most appropriate audience – i.e., more likely to convert.
The first step is to select the task that is likely to meet the prerequisites of Analytical Targeting feature which are:
- Single experience task (standard web tasks and standard mobile tasks),
- More than 50 conversions (less than this would not allow for a good enough model),
- More than 5000 unique users who observed the task, and had previously visited within the last 28 days (to give behavioral data to build the model)
Now, let us see an example:
In our case, we see from the Performance tab of a web task that we have 19970 unique users in the last 25 days with 690 conversions. This gives us enough data to build the model:
After the feature is enabled from the Targeting task for the chosen task, once the prerequisites are met, the model will be built. Enabling the task is a simple button click:
Results of the model appear once the prerequisites are met. Then we can click on the “Edit” button to select the cutoff and activate (deploy) the model.
Activating the model is another simple exercise; clicking on the “Target users most likely to convert” button.
For example, here, the cut off is set for 30 percent and the model is automatically deployed. This means, only future visitors that score above the cut off will see this task.
Of course, the model performance can improve, and in this example, after 1 week, the model did, and we saw an increase lift from 119% to 140%.
As populations change, and visitors perhaps get more exposed to a task, or perhaps as tasks become less relevant, the model, and the model performance will change again.
Analytical Targeting is a feature in “test mode” at present, though it is available for suitable users that wish to test the feature for themselves. If you are interested learning more about it, here is a great article. Please get in touch with me, or your usual SAS Customer Intelligence 360 contact if you would like to know more about the feature.
