This chapter covers the following topics:
Data Mining is the process of discovering patterns and relationships in large amounts of data. These patterns and behaviors are used to predict the behavior or customers and prospects.
A marketer uses the data by building “models” which act as a set of rules used to predict the value of a specific customer attribute. The prediction is based on the known values of other attributes. Oracle Marketing is integrated out-of-the-box with Oracle10g R2 Data Mining to build predictive models of customer behavior.
After installing Oracle Marketing, several implementation and administrative tasks must be performed to ensure that ODM is configured to fit your business needs.
A license for Oracle10g Data Mining Release 2 is a prerequisite for using data mining functionality within Oracle Marketing. Upon installation of Oracle Marketing, data mining functionality is automatically implemented. In other words, there are no additional required implementation steps.
The implementation steps listed in this section are optional and only necessary if additional functionality is required by specific business processes. This section assumes that Oracle Marketing and Oracle Data Mining are installed on the same database instance. Additionally, this section assumes the following:
Oracle Marketing is implemented
Oracle10g Data Mining Release 2 is implemented
The seeded data sources reference information stored within:
Oracle Customer Model (Trading Community Architecture or TCA)
Interaction History
Order Management
Customer Intelligence
To perform optional implementation steps, see the following sections:
Registering the Oracle Data Mining (ODM) schema with Applications Foundation allows ODM to programmatically fetch the ODM username and encrypted password, thus avoiding the security risk of maintaining this information in a property file.
During the installation process the ODM schema is registered as an external schema with Oracle Applications and a default database username (ODM) and password is assigned. For security reasons, you should change this password.
To change the password, log in as a user with System Administrator responsibility.
Navigation: Security > ORACLE > Register
Notes
Search for the ODM database user name, and change and confirm the password.
A concurrent manager is required exclusively for running data mining concurrent requests in order to avoid deadlock when multiple models or scoring runs are generated in parallel.
To activate the data mining concurrent manager, log in to Oracle Applications as a user with the System Administrator responsibility.
Navigation: Concurrent > Manager > Administer
Notes
Activate Marketing Data Mining Manager
Set the following data mining system profiles as mandated by your business requirements.
| Option | Required | Level | Setting | Effect/Limitation |
|---|---|---|---|---|
| AMS : Data Mining ODM Debug Enabled | Optional | User | Default value is No. If set to Yes - debug is turned on If set to No - debug is turned off |
Enables debugging information to be written in the concurrent request log. This is optional because turning on debug is optional. |
| AMS: Default Fixed Cost for Optimal Targeting | Required | Site | 0 | Defaults the fixed cost in the Optimal Targeting page. |
| AMS: Default Margin per Order for Optimal Targeting | Required | Site | 2 | This profile enables users to set a default value for the margin per order. The profile value will be used for default calculations. Users can change the values as per their need. |
| AMS: Default Cost per Contact for Optimal Targeting | Required | Site | 1 | This profile enables users to set a default value for the cost per Contact. The profile value will be used for default calculations. Users can change the values as per their needs. |
| AMS: Default Conversion to Order Percentage for Optimal Targeting | Required | Site | 5 | This profile enables users to set a default value for order conversion rate. The profile value will be used for default calculations. Users can change the values as per their needs. |
Additional Information: If the application and ODM are in the same database instance, set the profile AMS_ODM_DBLINK as NULL..
The data mining concurrent programs can be accessed by either logging into applications as System Administrator and navigating to Concurrent > Requests or by logging into Oracle Marketing as the Marketing Super User and navigating to Administration > Marketing > Predictive Analytics > Concurrent Requests.
Use the table below to run concurrent programs as mandated by your business requirements.
| Concurrent Manager | Required | Responsibility | Description |
|---|---|---|---|
| Workflow Background Process | Yes | System Administrator | Parameters Form:
|
| AMS: Expire Data Mining Models | Optional | Oracle Marketing Administrator | Background process monitoring data mining models for expiration. A frequency of once a day should be sufficient. |
| AMS: Update Data Mining Party Details | Optional | Oracle Marketing Administrator | Updates detail information collected on parties. The frequency should be the same as the Customer Intelligence Model (BIC) Summary Extraction process. |
After implementing data mining, depending on your business requirements, several administrative procedures may need to be performed.
For more information see:
Oracle Marketing ships with seeded models. However, custom models can also be created. See Creating Custom Model Types for more information on custom models. For more information on using the seeded model types, see the Oracle Marketing User Guide.
The following model types are seeded for data mining:
| Seeded Model Type | Purpose |
|---|---|
| Response | Used to predict customers/prospects that are most likely to respond to your e-mail, telemarketing, or direct mail campaign. |
| Loyalty/Retention | Used to predict individuals or businesses that are most likely to defect from your organization. This type of model assists you in converting them to loyal customers. |
| Custom | Used to predict the value of any customer attribute in TCA or in a user-defined data source. For example, you can set up a custom model that predicts financial risk based on a credit score attribute or a custom model to predict likelihood of high credit card balances vs. low balances. |
| Product Affinity | Predicts the likelihood of purchasing a particular product. You can also use this model to predict propensity to buy within a given product category based on an analysis of all products in that category. |
| Customer Profitability | Predicts the likelihood of a customer relationship being profitable. |
Oracle Marketing uses “predictive models” to predict future customer behavior. They are built using the Naive Bayes algorithm, which can predict binary or multi-class outcomes. Currently in Oracle Marketing, this algorithm is used to predict the binary outcomes of a target field.
For example, you might want to predict which customers are likely to respond to a direct marketing campaign or predict which customers are likely to buy a particular product. The possible values for the target field would be yes or no (i.e., a binary outcome).
A custom model type can be created to predict the value of any customer data attribute in the Oracle TCA customer data model (or in any user-defined data source) for customer information. This feature enables users to model any binary (Yes/No) customer behavior.
For example, marketers can set up a custom model to predict financial risk (based on a credit score attribute or a custom model) to predict likelihood of high credit card balances.
To create custom model types, you need to:
Add New Lookup Code
Restart HTTP Server
Verify New Model Type
To create a new lookup code for the custom Model type, log in to Oracle Forms with Oracle Marketing Administrator Responsibility.
Navigation: Setups > Lookups
Notes
Add the new model type to the Lookup Type: AMS_DM_MODEL_TYPE.
Meaning: This will be the display name (in the Oracle Marketing drop-down menu) for the custom model type.
Lookup types are cached in the HTTP Server. Therefore, after creating a new lookup type, you must restart the HTTP server to refresh the data. The exact procedures for this step will vary depending on your environment.
In order to begin using a new model type, you will need to setup the target field that the new model type will be predicting.
To verify the new model type, log in to Oracle Marketing as a user with super user responsibilities. Navigate to Administration > Marketing > Analytics > Targets and create a target. The custom model type you created must be visible in the Model Type drop-down list.
Oracle Marketing ships with seeded data sources that are used by both list generation and data mining functionality. The seeded data sources are editable. By default, all attributes (or columns) associated with the seeded data source are active. As the administrator, you must deactivate fields you wish to exclude for list generation and data mining analysis.
To define the data source attribute details for data mining, you should do the following:
Select the Data Mining Details view within the Attributes mid-tab of a data source.
Check the "Use for Data Mining" column to select attributes to be used for data mining.
To create a data sources using Predictive Analytics, login as an Oracle Marketing Administrator.
Navigation: Administration > Marketing > Predictive Analytics > Data Sources > Create Data Source
For more information on creating Data Sources, see Creating Data Sources in the Audience Administration Dashboard.
When marketers create predictive models, they are required to choose the model type and the target. As the administrator, you will need to set up the data sources and targets for the marketer. Terms are defined as follows:
Data Source: A database table (or view) which is the source of data
Target: Associates the data source, model type, and the target field. The target field is a column within the data source whose value you are trying to predict.
For example, you might define a target called “Laser Printer Indicator”. This target associates a specific model type (say, Product Affinity), a user-defined data source (B2B: Orders Data Mart), and a target field (a column within the corresponding data source). When a marketer chooses the Product Affinity model type and the “Laser Printer Indicator,” they can then build a predictive model to determine who is likely to purchase a laser printer by mining data stored within the corresponding data source (in this case the “B2B: Orders Data Mart”).
Oracle Marketing currently supports binary outcomes for the target field. As such, the possible values for a target field must either be binary or must be mapped to a binary outcome. For example, if you are trying to predict high-income households, then you will set up the “Income” target field such that values greater than 100000 is defined as the positive target (yes) and the other values are non-positive (no).
Navigation: Administration > Marketing > Predictive Analytics > Targets > Create
Notes
Name: Choose a name that logically represents the customer behavior being predicted. For example, if creating a target to predict customers who will purchase a laser printer, then name the target as “Laser Printer Indicator”.
Model Type: Select the Model Type that will use this target.
Target Field: The target field selector LOV displays the attributes of parent data source as well as any selected child data source. Choose the target field within the corresponding parent or child data source.
Condition: Use this column to select a condition operator for the target field.
Value: Use this column to indicate a value for the condition selected. Certain conditions allow certain value inputs. For example, if the condition “=” is selected for the target field “Response”, then an appropriate value could be “Yes”. The text input is case sensitive and must match the value in the database exactly. Note: In this example, if the value for “Response” in the database is “yes”, then the positive value of “Yes” will not match the database value.
Upper Value: Use this column to indicate a upper value when “BETWEEN” is chosen as the condition operator. Only the BETWEEN operator requires a value in this column. For example, if using the target “household income” with the operator “BETWEEN” and a Value of “100000” and an Upper Value of “150000”, then this would consider all household incomes between 100,000 and 150,000 to be positive target values. All other income values are considered non-positive. To build a valid predictive model, the data must contain positive and non-positive values.
Value Description: Use this column to give your values a text description.
The following targets are seeded :
B2C: Direct Mail Responders
B2B: Direct Mail Responders
B2C: E-mail Responders
B2B: E-mail Responders
B2C: Telemarketing Responders
B2B: Telemarketing Responders
B2C: Loyalty/Retention
B2B: Loyalty/Retention
B2C: Profitability
B2B: Profitability
B2C: Product Affinity
B2B: Product Affinity