Best Practice for Modeling AI Assets in CMDB When AI Control Tower Is Not Available

andinos · 3 weeks ago

Hello community,

We are defining an AI inventory and governance model in ServiceNow, but we do not have AI Control Tower (AICT) licensed or installed in our environment.

After reviewing the CSDM 5 guidance and the AI Inventory data model article, we understand that the AICT approach relies on three layers:

Product Model
AI Digital Asset
AI Configuration Item (CI)

However, without AICT we do not have access to tables such as:

alm_ai_system_digital_asset
alm_ai_model_digital_asset
alm_ai_prompt_digital_asset
alm_ai_dataset_digital_asset

Therefore, we are trying to determine the most appropriate way to model AI assets using only CMDB and Product Model tables.

Our current thinking is:

Business Context

cmdb_ci_business_app
- Microsoft 365
- GitHub Enterprise
- AI Platform / Model Hub

AI Solutions

cmdb_ci_appl_ai_application
- Microsoft 365 Copilot
- GitHub Copilot
- Now Assist for EA
- Custom AI Agents

AI Models

cmdb_ai_model_product_model
- GPT-4o
- GPT-5
- Claude Sonnet
- Titan Embeddings

AI Services / Capabilities

cmdb_ci_function_ai
- OpenAI API
- Bedrock Text Generation
- Azure Document Intelligence
- Textract
- Translation Services
- Embedding Services

Technical Deployments

cmdb_ci_ai_model_deployment
- GPT-4o PROD
- GPT-4o TEST
- Claude Sonnet us-east-1

For governance attributes (owner, risk classification, Responsible AI status, privacy review, etc.), we are considering either:

Adding custom governance fields directly to cmdb_ci_appl_ai_application, or
Creating a custom governance table related to AI Applications.

One specific area where we would appreciate guidance is how to model AI platforms such as:

AWS Bedrock
Azure AI Foundry
Google Vertex AI

Would you model these as:

cmdb_ci_appl_ai_application
cmdb_ci_function_ai
cmdb_ci_business_app
or something else?

More broadly, what is the recommended CSDM-aligned approach for representing AI assets and governance when AI Control Tower is not available?

Any examples, implementation experiences, or ServiceNow recommendations would be greatly appreciated.

Thank you!

Vikram Reddy · 3 weeks ago

Hey @andinos,

This is actually the normal state for most orgs right now, not a workaround. cmdb_ci_function_ai and cmdb_ci_appl_ai_application are documented in the CSDM 5 white paper as the two CI classes that support AI, independent of AI Control Tower. AICT layers governance on top through the alm_ai_system_digital_asset, alm_ai_model_digital_asset, alm_ai_prompt_digital_asset and alm_ai_dataset_digital_asset tables, but the CI and Product Model layer you already listed, cmdb_ci_appl_ai_application, cmdb_ai_model_product_model, cmdb_ci_function_ai, cmdb_ci_ai_model_deployment, is there without it.

The distinction that decides your Bedrock/Foundry/Vertex question: cmdb_ci_appl_ai_application is meant for AI you develop and host yourself, cmdb_ci_function_ai is for externally developed and hosted AI you're consuming. That reshuffles your draft a bit:

AWS Bedrock, Azure AI Foundry, Vertex AI: cmdb_ci_function_ai, one CI per environment/region actually consumed
Microsoft 365 Copilot, GitHub Copilot, Now Assist: also cmdb_ci_function_ai, same logic, you're consuming someone else's hosted model
Custom in-house agents: cmdb_ci_appl_ai_application

For governance fields, resist bolting custom attributes onto the OOB CI tables directly. Extend the class or relate a governance table instead, that keeps you upgrade-safe for when you eventually license AICT and need to migrate onto the digital asset layer.

Thank you,
Vikram Karety
Octigo Solutions INC

andinos · 15 hours ago

Thank you, Vikram. That clarification was extremely helpful.

I have two additional questions regarding how ServiceNow expects these classes to be used when AI Control Tower is not available.

1. AI Product Models

Should cmdb_ai_model_product_model be used to represent foundation models and market models such as:

GPT-4o
GPT-5
Claude Sonnet
Llama
Titan

If so, is it acceptable to create and maintain those Product Model records manually?

My assumption is that, in an AI Control Tower implementation, many of these records would eventually be discovered and populated automatically through AI inventory and discovery integrations.

Without AICT, is manual registration and lifecycle management of Product Models considered the recommended approach?

2. Business Application vs AI Application

I am also trying to understand the intended relationship between cmdb_ci_business_app and cmdb_ci_appl_ai_application.

For example, assume we develop an internal AI agent called Corporate Assistant that:

supports a business capability,
is consumed by end users as a business solution,
uses multiple models (GPT-4o, an internal model, and an embedding model).

Would the recommended CSDM-aligned approach be:

Business Application

└─ Corporate Assistant

AI Application

└─ Corporate Assistant Agent

With relationships between both records,

or should the solution be represented only as an AI Application?

In other words, should cmdb_ci_appl_ai_application be viewed as:

a specialized type of Business Application,
a supporting technical AI component of a Business Application,
or a complete replacement for a Business Application in AI-centric solutions?

In other words, if an internally developed AI agent is considered an cmdb_ci_appl_ai_application, should the related cmdb_ci_business_app represent the business-facing product/service consumed by users while the AI Application represents only the AI implementation layer?

I am trying to understand the architectural role that ServiceNow expects cmdb_ci_appl_ai_application to play within CSDM.

Thank you very much!

Fernando Andinós

Petrobras

andinos · 15 hours ago