Artificial intelligence in Oracle Analytics Cloud relies on semantic metadata to interpret user questions accurately. AI agents use the subject area definition, not the raw data source, to generate responses. When business attributes are ambiguous, such as non-unique customer names, AI agent results may become inaccurate if the semantic model doesn’t clearly define entity grain. Proper semantic modeling ensures accurate aggregation, filtering, and entity resolution. This article focuses on a specific and highly effective tuning technique that uses the Descriptor ID property in the semantic model to improve AI agents’ aggregation accuracy, entity resolution, grouping behavior, and filtering, even when business attributes are non-unique.

Understand the Business Challenge of Ambiguous Customer Attributes

Consider a customer domain with the following attributes:

  • Customer ID (unique identifier)
  • First Name
  • Last Name
  • Preferred Name (non-unique display attribute)
  • Age
  • Country

In real-world enterprise data, names are rarely unique. For example, Preferred Name is not guaranteed to be unique. Multiple records may share the same Preferred Name, while each record maintains a distinct Customer ID.

If the semantic model doesn’t define how Preferred Name relates to Customer ID, AI agents may group by name instead of by identifier. This can cause aggregation and entity resolution errors in AI-generated queries.

Why Agent Tuning Begins in the Semantic Layer

AI agents rely on:

  • Logical keys
  • Aggregation rules
  • Logical table grain
  • Column relationships
  • Descriptor mappings

If a descriptive column (such as Preferred Name) is not uniquely mapped to its identifier (Customer ID), AI agents may:

  • Aggregate at the wrong level
  • Collapse duplicate names
  • Miscalculate distinct counts
  • Apply filters inconsistently

The accuracy of AI agents is directly tied to semantic precision.

Steps to Configure the Descriptor ID to Control Aggregation and Display Behavior

You can use the Descriptor ID property to map the Preferred Name to Customer ID:

  1. Define Customer ID as the logical identifier.
  2. Configure Preferred Name as the descriptor column.
  3. Set the Descriptor ID property to link Preferred Name to Customer ID.

With this configuration, aggregation and filtering operate at the identifier-level, while Preferred Name remains the display attribute.

Example

The following example illustrates how ambiguous customer names can impact AI aggregation behavior.

1. Review data in the physical table

The physical table contains duplicate Preferred Name values associated with different Customer IDs. The names Vikas Sharma and Karan Singh each appear more than once with different Customer IDs. This creates ambiguity when aggregation occurs at the name level.

Sample physical table showing duplicate Preferred Name values with different Customer IDs
Sample physical table showing duplicate Preferred Name values with different Customer IDs

2. Configure Descriptor ID in the semantic model

Open the Preferred Name logical column properties and set Customer ID as the Descriptor ID column. This links the display attribute to its unique identifier.

Configure Descriptor ID to link Preferred Name with Customer ID
Configure Descriptor ID to link Preferred Name with Customer ID

3. Reinforce behavior in AI agents (optional)

Optionally, provide custom instructions in the AI agent configuration panel to reinforce identifier-based aggregation logic.

Optional custom instruction to reinforce identifier-level aggregation
Optional custom instruction to reinforce identifier-level aggregation

4. Validate results in a workbook

After deploying the semantic model, test AI queries in a workbook to confirm the correct aggregation and filtering behavior.

AI agent correctly aggregates at the Customer ID grain after Descriptor ID configuration
AI agent correctly aggregates at the Customer ID grain after Descriptor ID configuration
Aggregate filter correctly applied using Customer ID grain
Aggregate filter correctly applied using Customer ID grain
Accurate distinct count using Customer ID
Accurate distinct count using Customer ID
Correct average age calculation by country
Correct average age calculation by country

Observe how AI agent behavior improves after Descriptor ID configuration

  • Correct aggregations at the Customer ID grain
  • Accurate distinct counts without duplicate-name collapse
  • Reliable aggregate filters
  • Proper entity resolution

Summary

Tuning AI agents in Oracle Analytics Cloud is about strengthening the semantic foundation. The Descriptor ID property ensures identifier-level aggregation, accurate grouping, and reliable AI responses at scale.

Call to Action

Review your semantic model and validate identifier-to-descriptor mappings in your subject areas. Visit the Oracle Analytics Community to share your scenario or consult the Oracle Analytics Cloud documentation for additional guidance.