Feature-Level Model Restriction in AI+ Studio allows administrators to control which AI models can be used for a specific Generative AI (GenAI) feature.

This capability is available within Feature Access Management, which provides a centralized interface to manage AI features, including enabling features, configuring inputs, and controlling access. 

With feature-level model restriction, you can:

• Restrict model usage for individual features

• Optimize performance and cost

• Enforce governance and compliance policies

Why Use Feature-Level Model Restrictions

Feature-level model restrictions are designed to address the following challenges:

• Cost Control: Some GenAI features operate at a large scale. Configuring expensive models for such features can significantly increase operational costs. Example: AI Sentiment Analysis processes a high volume of interactions and is optimized for Sprinklr’s in-house sentiment model. Allowing premium LLMs can lead to unnecessary cost escalation.

• Reliability and Performance: Not all models perform equally across use cases.

  Certain features are designed for specific models, tested for accuracy, and optimized for performance. Allowing unsupported models may result in:

• Reduced accuracy

• Inconsistent outputs

• Performance degradation

• Governance and Compliance: Organizations often require strict control over model usage to:

• Meet internal governance policies

• Ensure regulatory compliance

• Maintain security standards

Feature-level restrictions provide an additional layer of control beyond tenant-level configurations.

Prerequisites

Before configuring feature-level settings, ensure the following:

• You have access to AI+ Studio.

• Required permissions under Feature Access Management are enabled.

• You have Edit access to modify feature configurations.

Configure Feature-Level Model Restriction

Follow these steps to configure feture-level model restrictions under Feature Access Management.

​

• Go to AI+ Studio from Sprinklr launchpad.

• On the AI+ Studio screen, select the Feature Access Management card. The AI Feature Record Manager will open.

​

​

​

• Locate the required feature integrated with AI+ Studio.

• Click the vertical ellipses menu (⋮) and Select Edit.

​

​

​

• From the configuration screen, go to Feature-Level Settings screen. This screen allows you to define which AI models are available for a selected GenAI feature. Configure the following input fields under Allowed Models section of the screen:

  ​

  ​

  Field	Description
  Provider	Select the AI model provider from which models will be allowed.
  Models	Select one or more models from the chosen provider.
  + Allowed Models	Use this button to configure additional providers. Each row acts independently. You can define multiple provider–model combinations.

​

​

Click the 'Save' button to save the configured model restrictions and apply changes to the selected feature.

Configuration Behavior Summary

On-Top Allowed Models

AI+ Studio automatically includes a set of implicit models called on-top allowed models. These models are added to the feature configuration even if you do not explicitly select them. They ensure that existing workflows continue to function without disruption while expanding the available model options for a feature.

The following models are included as on-top allowed models:

• Partner default model
• Models from the default feature deployment

The effective model list is calculated as:

Allowed Models = Selected Models + On-top Allowed Models

This behavior ensures that the system maintains compatibility with existing configurations and avoids breaking changes. It also extends the allowed model set automatically without requiring additional configuration.

On-top models are not applied when a Sprinklr-managed restriction is enforced.

Types of Feature-Level Restrictions

Feature-level restrictions define how model usage is controlled for a feature. AI+ Studio supports two types of restrictions: customer-managed restrictions and Sprinklr-managed restrictions. These two approaches provide different levels of flexibility and control depending on your use case. 

1. Customer-Managed Restrictions

Customer-managed restrictions are configured by administrators through the Feature-Level Settings screen. These restrictions provide flexibility and allow you to customize model usage based on your requirements.

Characteristics

• Editable in Feature-Level Settings

• Supports:

  • Sprinklr-provided models
  • BYOK (Bring Your Own Key)
  • BYOM (Bring Your Own Model)

When you use customer-managed restrictions, only the models you select are allowed. On-top models are automatically included to maintain compatibility. You can update or modify these restrictions at any time as your requirements change.

Use cases

• Cost optimization
• Controlled testing
• Custom governance

2. Sprinklr-Managed Restrictions

Sprinklr-managed restrictions are defined and enforced by Sprinklr for features that require strict governance. These restrictions are not editable and are designed to ensure consistent performance and compliance.

Characteristics

• Read-only
• Not editable by customers

Under this model, only Sprinklr-approved models are allowed. On-top models are not included, and you cannot modify the selected Sprinklr models. However, you can still restrict models that you configure independently, such as BYOK or BYOM models.

Use cases

• Governance-sensitive workflows
• Performance-critical features
• High-scale operations

Example: AI Sentiment Analysis

AI Sentiment Analysis is an example of a feature that uses Sprinklr-managed restrictions. In this case, the feature is configured to allow only the Sprinklr in-house sentiment model.

Configuration

• Allowed model: Sprinklr in-house sentiment model

Because this is a Sprinklr-managed restriction, other Sprinklr models are not available and on-top models are not applied. You cannot modify the allowed Sprinklr models for this feature. However, you can still apply restrictions to models that you control.

Why On-Top Models are Excluded in Managed Restrictions

On-top models are excluded in Sprinklr-managed restrictions because the feature is already tightly controlled with a predefined set of approved models. In this scenario, backward compatibility is not required, and strict control ensures consistent and predictable performance.

As a result:

• On-top models are not added
• The allowed model set remains fixed

Best Practices

Follow these best practices when working with feature-level settings:

• Use customer-managed restrictions when you need flexibility
• Use Sprinklr-managed restrictions for critical or sensitive features
• Review model configurations regularly
• Align model selection with feature requirements
• Avoid assigning high-cost models to high-scale features
  

Feature-Level Settings give you granular control over how AI models are used across features. They help you balance cost, performance, and governance while supporting different configuration needs.

​

​

​

​

​

​

​

​

​

​