Tracking "AI" applications

MEBoatman · 3 weeks ago

I am the process owner/lead for our Enterprise Architecture team's Application Portfolio Management practice. We use ServiceNow's Enterprise Architecture module for this. We have a relatively mature application inventory - to the point that lots of people across the org are leveraging it for information about applications. Definitely happy about that. The challenge comes in when people look to our app inventory for information that it does not currently hold.

The latest ask is to help internal audit identify "AI applications". I am currently trying to distill some requirements as I'm sure we can all agree that what constitutes an "AI application" is certainly subject to debate.

That having been said I'm wondering if others have run into similar requests? Is there a "ServiceNow way" to track this kind of thing?

Mathew Hillyard · 3 weeks ago

Hi @MEBoatman

I'm sure I've seen something about this in the many ServiceNow webinars on AI. I believe there are plans to enable AI categorisation in EA as a future feature.

A quick AI tool query brings up the following - I've done my best to edit out the AI-generated errors but some may still remain. It seems about as good advice as I've seen to date:

1. The AI CSDM 5.0 Blueprint

ServiceNow maps AI components across four main structural boundaries of the CSDM domain:

 ┌────────────────────────────────────────────────────────┐
 │        Design Domain (Application Portfolio / EA)       
 │  [cmdb_business_app] (e.g., Enterprise Copilot)        
 └───────────┬────────────────────────────────────────────┘
             │ Consumes (for now)/ Uses (once the platform is fixed for CSDM 5.0)
 ┌───────────▼────────────────────────────────────────────┐
 │                  Manage Technical Services             
 │  [cmdb_ci_service_discovered]
 │  (Application Service - e.g., Now Assist Production)   
 └───────────┬────────────────────────────────────────────┘
             │ Runs On / Composed Of
 ┌───────────▼────────────────────────────────────────────┐
 │               Operational Domain (CMDB CIs)            
 │  [cmdb_ci_appl_ai_application] OR [cmdb_ci_function_ai]
 └───────────┬────────────────────────────────────────────┘
             │ Hosted On
 ┌───────────▼────────────────────────────────────────────┐
 │                Foundation / Infrastructure             
 │  [cmdb_ci_compute_element] (Containers, K8s, GPUs)     
 └────────────────────────────────────────────────────────┘

2. Core CI Classes for AI Architecture

To model AI payloads effectively, map your architecture to these specific CI classes:

AI Business Application (cmdb_business_app): The strategic, conceptual asset managed in Enterprise Architecture. Use this to track governance, total cost of ownership (TCO), business capabilities, risk compliance, and whether it is a managed/unmanaged AI asset.
Service Instance (cmdb_ci_service_auto) (multiple child tables): Instantiation of the Business Application in a specific environment; represents all the software/hardware CIs related to it.
AI Application (cmdb_ci_appl_ai_application): Used for self-hosted or privately managed AI deployments. If your enterprise hosts open-source Large Language Models (LLMs) like Llama inside Kubernetes clusters, Docker containers, or private Linux/Windows runtimes, model them here.
AI Function (cmdb_ci_function_ai): Used to represent SaaS or public cloud platform AI capabilities. This tracks model endpoints consumed over APIs (e.g., Azure OpenAI service, Anthropic Claude API, or Google Gemini cloud workloads).

Note on Latest CSDM Guidance: While early drafts treated these strictly as hosting buckets, formal guidance from AI Control Tower (AICT) structures them sequentially. The AI Application acts as the installed software package layer, while the AI Function acts as the operational runtime/skill deployment underneath it.

3. Step-by-Step EA & CSDM Modeling Workflow

Step 1: Define Strategic Design (EA Layer)

Navigate to the Enterprise Architecture Workspace > Application Portfolio > Business Applications.
Create your Business Application record (e.g., "Customer Service AI Triage").
Link the Business Application to the specific business capabilities it satisfies (e.g., Customer Engagement, Automated Response Management).

Step 2: Establish the Service Instance (Logical Operational Layer)

Instantiate a Service Instance(e.g., "Customer Service AI Triage - Prod") to represent the environments.
Build a Consumes::Consumed by relationship from the conceptual Business Application to this Service Instance

Step 3: Populate the Infrastructure & AI Engine (Physical Layer)

For External Cloud LLMs: Create an entry in cmdb_ci_function_ai for the specific API endpoint. Map it to the Application Service using a Runs on::Hosts dependency.
For Internally Hosted LLMs: Create an entry in cmdb_ci_appl_ai_application representing the runtime container image/software.
Map Underlying Hardware: Link the AI Application CI directly to the compute engines or cluster layers—including specific GPU hardware elements (cmdb_ci_gpu) which drive model processing. [1, 2, 3, 4]

4. Enterprise Architecture Matrix

Architectural Layer CSDM Domain Primary ServiceNow Table AI Example Mapping

Business Strategy	Design	cmdb_business_app	Enterprise GenAI Assistant Platform
Logical Runtime	Manage Technical Services	cmdb_alloc_service_app	Now Assist API Integration - Production
AI Software/Engine	Sell/Consume & Build	cmdb_ci_appl_ai_application	Self-hosted Llama-3 70B Orchestrator
AI Cloud Endpoint	Sell/Consume & Build	cmdb_ci_function_ai	Azure OpenAI GPT-4o API Service
Compute Hardware	Foundation	cmdb_ci_gpu / cmdb_ci_server	Nvidia H100 GPU Cluster Instance

I hope this helps!
Mat

yogesh41 · 2 weeks ago

Hi,

The AI Control Tower is the single pane of glass discover, and manage organization's AI capabilities such as AI System, Models, or Datasets. Enterprise Architecture has integration with AI Control Tower to track and manage relationship between AI System and Business Application.

It allows managing inventory of AI System that are supported by Business Application. This will help the users know which existing or new Business Application has AI capabilities , which models it uses etc.

As posted by someone oneelse as well, AI CSDM 5.0 Blueprint is way to discover and track for hosted AI services.

We would like to know which type of associattion is more suitable in your use case ?

Yogesh