Why create overrides?

A predictive model is based on historical data: it analyzes past data to predict the future. 

In the Likelihood to Buy model, the segment of an Account (very high, high, medium, low) is defined by the total activity of its contacts. This activity is weighted based on historical analysis. However, you may want to add override the model to force the segment based on a specific behavioral trait. 

Example: 

• You are selling a seat-based product with a basic plan at 10 seats max. You want to make sure you surface to your Sales team any account with 10 active users as it means they are reaching the limit of their plan and could be interested in the higher plan. To do so you can add an override in the model like "If number of active users in the account in the last 30 days is >= 10 then score very high". 
• You already have a PQL motion in place where individual users with a lot of engagement are surfaced to your sales team via a PQL score. You may want to limit to high or medium only the score of an account if only 1 user is highly active. To do so you can add an override in the model like "if number of active users in the account is = 1 then score medium" 

How to add an override to a live model?

Pre-requisites 

• You have the permissions of the Architect or Admin role
• You know what an Aggregation is 

Step 1: Duplicate the live model

1. Go to the Data Studio (studio.madkudu.com) 
2. Duplicate the model marked as "live" (live models cannot be edited directly)
3. Name the duplicated model how you want

Step 2: Create the override

1. 1. In the model,Model >Overrides
   2. Click on Create new rule
   3. Select the Aggregation, condition and value 
   4. Click onSave.

Step 3: Assess the impact of the override

You would want to make sure this override is not boosting or penalizing too many accounts in a specific segment degrading the performance of the model.  

To compare the impact of the override you just added on the training dataset

Click on the Overrides impact analysis button to look at the performance on the trainingdataset. The difference in performance between the tab Thresholds and this tab is the impact of all the overrides on the performance.

To compare the impact of the override you just added on the validation dataset

Open the live model on your browser, head to the Review > Performance tab, and compare the performance graph you see to the Performance tab of the model you added to override.

Step 4: Deploy override

Keep these considerations in mind when deploying overrides:

1. All your Accounts impacted by this override and that are usually scored by MadKudu will get rescored with the next batch scoring within the next 4-12 hours (see when the next Analysis then Sync process should run in the Processes Page).
2. If you automated workflows in your CRM based on the Likelihood to Buy score, these may therefore get triggered.

When you are ready, go to the Deploy tab to deploy the model

Well done!

F.A.Q 

If an account falls in an override "should be low" and another "should be very high", which override is applied?

Overrides penalizing the score of an account have priority over an override boosting the score.

Ex: If there is

• an override "IF number of active users last 30 days >= 20  THEN should be very high "
• and an override "IF number of models deployed last 30 days = 0 THEN should be medium"

Then the 2nd rule will prevail and the account with 20 active users who haven't deployed a model will be scored medium. 

Where are the aggregations in the picklist coming from?

The picklist contains Aggregations which you can create or edit in the Aggregation section of the Data Studio (left navigation bar "Aggregation").