Text decision tree

Adapted from Chapter 19, Microsoft AI and Copilot landscape, from Decision Trees: Frameworks for Technology Choices, reviewed October 27, 2026. Start at the top and choose the first layer that satisfies the requirement.

START
│
├─ Q0 GOVERNANCE GATE: Will agents act across systems (read/write to
│    email, CRM, files, tools) at more than pilot scale?
│      YES → Design Agent 365 + Entra Agent ID + Purview + Defender
│            BEFORE scaling any layer below. Then continue.
│      NO  → continue.
│
├─ Q1 Is the need met by AI inside Microsoft 365, Dynamics 365,
│    or Security apps the users already have?
│      YES → Embedded SaaS Copilot + agents (Option A). STOP.
│      NO  → continue.
│
├─ Q2 Can a business user assemble the agent from connectors,
│    knowledge, and natural-language instructions?
│      YES → Copilot Studio low-code agents (Option B). STOP.
│      NO  → continue.
│
├─ Q3 Does the solution need custom code, model choice, multi-agent
│    orchestration, or deep retrieval control?
│      YES → Microsoft Foundry pro-code (Option C). Continue to Q4–Q6.
│      NO  → re-check Q1/Q2.
│
├─ Q4 Is the task a well-scoped perception task (speech, vision,
│    OCR, translation, content safety)?
│      YES → Foundry Tools prebuilt APIs (Option D). STOP or combine.
│      NO  → continue.
│
├─ Q5 Is this classic ML — tabular prediction, forecasting, custom
│    training, MLOps — rather than generative AI?
│      YES → Azure Machine Learning / Fabric / Databricks (Option E).
│      NO  → continue.
│
└─ Q6 Must inference run on-device, disconnected, or in a sovereign
     site?
       YES → Foundry Local + Azure Local + Azure Arc (Option F).
       NO  → default to Option C (Foundry pro-code).

AI-generated illustration

An illustrative overview. Follow the video graph and text tree for exact decision routes.

The illustration is visually useful, but it does not preserve the text tree exactly:

  • It compresses Q0–Q6 into five broad questions.
  • It places Govern & fund at the last numbered step. In the text tree, the Q0 governance gate comes before scaling and then rejoins Q1.
  • It adds generic approved, caution and risk routes in place of the text tree's explicit A–F choices, stop points, continuations and return path.

Those changes make the image an interpretation of the topic. The graph in the video is drawn from explicit nodes and connections, so each branch can be checked against the text above. Use the text tree and the verified graph when following the decision logic.

Reference links

The tree follows the dated chapter version. These primary references describe the corresponding Microsoft products and capabilities.

Related decision trees