# Enterprise AI operations

> How BROCS applies an operating model to AI already in use, covering operational maturity, production control, lifecycle management, and delivery speed.

Source: https://brocs.fyi/enterprise-ai-operations/
Framework: BROCS (Build, Run, Observe, Control, Secure), brocs.fyi. CC BY 4.0, attribute to brocs.fyi.

---


Enterprise AI operations cover the shared services, responsibilities, and controls behind
AI systems already in use. BROCS organizes that work under Build, Run, Observe, Control,
and Secure.

The same scopes can define requirements before deployment. The
[starting enterprise AI guide](/resources/starting-enterprise-ai/) uses BROCS as an
onramp for a first tool, pilot, supplier, or platform. After adoption is underway, the
same review applies to the active operating estate.

## AI can arrive before the operating model

An organization's first AI dependency may be purchased, embedded, configured, or built.
Microsoft treats these routes as separate adoption models with different degrees of
control and required skill.[^microsoft-models-operations] Deep model development is one route among
several.

Every route creates operating questions. Someone has to manage identity and data access,
understand runtime dependencies, record behavior, change configuration, contain cost,
and produce evidence after an incident. Those responsibilities exist even when a supplier
runs the model or the application arrives through an existing contract.

Build covers the sanctioned ways people configure assistants, connect data, create
agents, write prompts, evaluate behavior, publish work, train models, and write
application code.

## Adoption and operations answer different questions

AI adoption frameworks help an organization set ambition, choose use cases, develop
skills, and decide which technical paths to support. Google Cloud describes its adoption
framework as a map between current and desired AI capability.[^google-adoption-operations]

For an active-system review, BROCS uses current operating evidence. AI may already be
present in several parts of the organization while the support behind it varies by team,
supplier, or use case. The framework asks whether each active system has a supported
delivery path, an owned runtime, useful telemetry, safe change mechanisms, and enforced
requirements.

The approaches can be used together:

- Use an adoption or transformation framework to set direction, investment, and the
  desired organizational capability.
- Use BROCS to inspect the operating coverage behind current tools, production pilots,
  and deployed systems.
- Use the [BROCS assessment](/resources/assessment/) to record the gaps that can be
  demonstrated today.
- Use the [buyer questions](/resources/questions/) to test supplier responsibilities and
  the evidence available in production.

## Five operating scopes

| Scope | Operating question |
|---|---|
| Build | Do people have a supported way to create, configure, evaluate, and deliver AI work? |
| Run | Can the organization provision and operate the required runtime, state, secrets, and data placement? |
| Observe | Can teams measure service health, model behavior, quality, cost, and the actions an agent took? |
| Control | Can authorized people change routes, prompts, access, configuration, and rollback paths without rebuilding the system? |
| Secure | Are permissions, policy, spending limits, incident evidence, and compliance requirements enforced during operation? |

These scopes form a production AI operating model. Ownership may be distributed across
platform, application, infrastructure, security, data, finance, and compliance teams.
The interfaces and evidence between those teams have to remain clear.

## Operational excellence for AI

Google's current architecture guidance describes AI and ML operational excellence in
terms of deploying, managing, and governing production systems while reducing operating
complexity.[^google-operations] DORA's platform engineering guidance adds shared,
self-service delivery paths and warns that faster development can be absorbed by testing,
security, and deployment bottlenecks.[^dora-platform-operations]

BROCS applies those practices across the enterprise AI lifecycle. Use the
[operational coverage guide](/resources/operational-maturity/) to assess current evidence,
the [delivery guide](/resources/accelerating-ai-delivery/) to find delays, and the
[adjacent-practices guide](/resources/adjacent-practices/) for the source material behind
the framework.

[^microsoft-models-operations]: Microsoft, ["AI strategy - Guidance to set your organization's AI strategy"](https://learn.microsoft.com/en-us/azure/cloud-adoption-framework/ai/strategy), Cloud Adoption Framework, updated June 26, 2026.
[^google-adoption-operations]: Google Cloud, [*AI Adoption Framework*](https://services.google.com/fh/files/misc/ai_adoption_framework_whitepaper.pdf), 2020, accessed August 27, 2026.
[^google-operations]: Google Cloud, ["AI and ML perspective: Operational excellence"](https://docs.cloud.google.com/architecture/framework/perspectives/ai-ml/operational-excellence), accessed August 27, 2026.
[^dora-platform-operations]: DORA, ["Platform engineering"](https://dora.dev/capabilities/platform-engineering/), accessed August 27, 2026.