Creating Look-alike Models with Oracle Modeling 360

You can use look-alike models in the Oracle Data Cloud platform UI to identify high-value users who behave similarly to your best customers and converters so that you can increase the reach and precision of your target audience.

About Oracle Modeling 360

Oracle Data Cloud provides an automated, self-serve modeling solution, Oracle Modeling 360, for sending model requests to a modeling vendor. A model request includes:

• The users you want modeled (the signal audience)
• The group of categories used to rank the users in your signal audience (the profile input)
• The instructions for tailoring your model request, such as the granularity of the data you want returned (for example, the top 0-1%, 1-5%, and 6-10% of users in your custom look-alike model)

After your modeling vendor receives your model request, they will score and stack rank users according to who best matches the attributes of your signal audience and they will add the top percentage of look-alikes as new categories to your taxonomy. You can then deliver your look-alikes across multiple media execution platforms for targeting, analysis, and optimization.

Look-alike modeling provides the following benefits:

• Automated workflow: The Oracle Data Cloud platform automates the process of sending model requests to look-alike modeling partners.
• Custom models with a large selection of data: Create models from your first-party data, second-party data shared with you, and third-party data from the Oracle Data Marketplace.
• Multivariate models: Multivariate look-alike models consider all the attributes of the users, including the frequency and recency in which they have been tagged with the attributes.
• Rapid model creation and activation: Create a model request in minutes; receive your look-alikes within a week.
• Automatic refresh: Your look-alike model is refreshed weekly so you always have the latest scores for your best users.

To create and activate a look-alike model :

1. Install the Oracle Modeling 360 look-alike modeling app.
2. Create a model request.
3. Monitor your model request.
4. Activate your look-alike models.

Note: Your look-alike rate will be set in the campaign workflow. This will typically be a CPM on impressions set in the central rate card and will be displayed on your model category when you create an audience.

Install Oracle Modeling 360 app

The app catalog includes the Oracle Modeling 360 app for creating look-alike models. This app does not require any ramp time.

Important: You cannot delete a look-alike app. Delete permissions are disabled to protect the flow of data from the look-alike partner to the platform.

To add the Modeling 360 app:

1. Contact My Oracle Support (MOS) and request access to look-alike modeling with Oracle Modeling 360.
2. Log on to partner.bluekai.com, and then select Apps > Install Apps.
3. Click App Catalog.
4. Click Look-alikes campaign solution type.
5. Select Oracle Modeling 360.
   
6. In the App Name box, enter a name that identifies the look-alike app configuration.
7. (Optional) In the App Profiles section, enter an descriptive name for a custom profile in the App Profile Name box, select the categories to be included, and then click Add App Profile. You can add multiple custom profiles.

   Tip: The best models include the largest possible data set. The default profile includes all the categories that are available to you and will generate the best models.

8. Click Save . The look-alike app is added to the Install Apps page.

9. The platform creates an audience named "Audience for Oracle Modeling 360 Lookalike - [Partner ID YourPartnerID + Vendor Name M360 - Default Lookalike Model App]" in your partner seat and shares it with Oracle Modeling 360. It includes the following categories:

   • All your first-party categories under the private node in your taxonomy.
   • Your self-classification node, which contains all the first-party categories you created.
   • All second-party categories that have been whitelisted to you by another DMP client with modeling, targeting, and analytics permissioning.

10. The platform automatically creates a server data transfer (SDT) campaign in the look-alike partner's seat to deliver your first- and second-party categories in an hourly batch file. The look-alike partner begins ramping your first-party data. All the third-party data in the Oracle Data Marketplace has already been sent in a separate SDT batch delivery and has been ramped.

11. You can send your model requests to your look-alike partner and your custom app profiles are ready to be used in your model requests.

Creating a model request

After you have added a modeling app, you can create a maximum of 50 model requests in your Oracle Data Cloud partner seat.

To create a model request:

1. Select Manage > Models. The Models page is displayed.
   
2. Click Create Model. The Create Model Request dialog opens.
   
3. In the Model Name box, enter a descriptive name for your model request that makes it easy to identify later. The model name must not include any special characters.
4. In the Signal Audience list, select the audience containing the users you want to model (for example, an audience containing your converters or purchasers). Enter two or more characters in the audience name to filter and select the audience.
   When you select the audience, the Reach field displays the number of users seen in the selected audience over the last 30 days (not the projected reach of the model).
   

Tip: You can select your signal audience directly from the Audiences page by selecting its check box and then selecting Create > Create Model.

5. In the State Date and End Date boxes, enter when you want the model to start and stop running. By default, the Start Date is set to tomorrow's date, and the End Date is set to 6 months later.

6. In the Model Categories Format list, select the top groups of look-alikes in your model that you want classified into new categories in your private taxonomy. The following table summarizes the model category format options:
   

   Option	Categories created	Description
   Default	0-1%, 1-5%, and 5-10%	Includes three categories representing the top 0-1%, 1-5%, and 5-10% of users in your model to provide a balance of performance and scale
   Extreme Performance	. 0-.01%, .01-.1%, .1-1%, 1-2%, 2-5%, 5-10%	High-performance model with increased granularity and precision. Useful for targeting the highest of the top-ranked look-alikes
   Scale	0-1%, 1-5%, 5-10%, 10-15%, 15-20%, 20-30%	Expanded reach useful for targeting a large number of look-alikes. This format is based on the default model, but it has additional categories for lower percentages of look-alikes to increase scale.
   Custom	User-specified	Custom look-alike model. Enter a comma-separated list of ranges, which may contain whole numbers and decimals (for example, 0.1-0.5%, 0.5-1%, 1-2%)

7. In the Country box, select look-alikes from which countries to include in your model categories. By default, look-alikes from only the United States are included.

   You can also add look-alikes from Argentina, Australia, Austria, Belgium, Brazil, Canada, Chile, Colombia, Denmark, Finland, France, Germany, Iceland, India, Indonesia, Ireland, Italy, Japan, Luxembourg, Malaysia, Mexico, Netherlands, New Zealand, Norway, Peru, Philippines, Poland, Portugal, Russian Federation, Saudi Arabia, Singapore, South Africa, South Korea, Spain, Sweden, Switzerland, Thailand, Turkey, United Arab Emirates, United Kingdom, United States, and Vietnam.

8. Under Predictors, select which groups of categories are used to rank the look-alikes in your model (1st-party data and/or 3rd-party data). The default is 3rd-party data only,

9. (Optional). In the Notes box, enter any comments related to your model request.
10. Click Save. Your model request is added to the Manage > Models page.

Monitoring your model requests

You can use the Models page to monitor and manage your model requests.

This page lists the status of your model request, which may be on of the following (listed in order of the model creation process):

Status		Description
	Sent	Your model request has been delivered to the look-alike modeler.
	Ramping	This model uses first-party data that is currently being processed by the modeler. When your data has finished ramping, the status will switch to the Creating state. Subsequent model requests containing your first-party data will not require this ramp time and will be processed within the standard 1-week time frame. Ramping is not applicable for model requests that include only third-party data.
	Creating	The vendor has received your model request, and the model is now in the process of being created. The vendor does the following to onboard your look-alike modeling data: After the look-alike partner finishes creating your model, they create an offline file that maps your look-alike users' online profiles to the modeled categories you specified in the model request. The look-alike partner drops the offline file onto Oracle Data Cloud upload servers. The platform then onboard your look-alike users into new modeled categories in the look-alike partner's taxonomy within 24 to 48 hours. Rules written by Oracle Data Cloud map the users in your offline file to the new modeled categories. Your new modeled categories are then whitelisted into your taxonomy.
	Active	Categories representing your look-alikes have been added to a Oracle Modeling 360- Private node in your taxonomy. Your model will continue to be updated weekly.
	Pending Info	The model is pending additional information before it can be activated. Typically, this means that your model request does not include enough data. For more information, contact My Oracle Support (MOS).
	Disabled	The model has been disabled by a user in your Oracle Data Cloud partner seat.

The Models page also lists the number of models you have created out of the 50 you are allotted. It also includes columns listing the ID, name, profile, categories, labels of your model request, the user who created the model, and the dates when the model was created and last updated. You can sort your models using these columns and filter them based on their name and status.

Disabling models

You can disable an active model to stop the continuous training of the model and the updating of the users in your look-alike categories. You may want to disable an active model if you want to use a different model that better fits your business goals.

To disable models, select the check boxes for the models to be disabled, click Disable, and then click Disable in the confirmation dialog.

Activating disabled models

To activate a disabled model and begin re-training and updating the model, select its check box and then click Enable. Your re-activated model will re-train on the current data set. This means that you will have to wait for a new model to be created before it is in the Active state and your look-alike categories are ready for use.

Viewing model details

You can click a model to view a detailed summary of the model that includes information on the model categories being created, the model vendor, and the signal audience.

Activating your look-alike models

After your look-alike categories have been added to your taxonomy, you can add them to your target audiences, and deliver them across multiple media execution platforms. You can find your look-alike categories in a Look-alikePartnerName- Private node in your taxonomy. For more details, see activating data.

Related

Installing an app

Creating an audience

datacratic, Oracle Modeling 360, look-alike models, profile