2.1.1. Assessing Agent Use Cases for Automation, Analytics, and Decision-Making

Not every business process needs an AI agent. The architect's first job is to evaluate where agents create genuine value versus where simpler solutions — Power Automate flows, business rules, or standard reporting — would suffice. Agents add value when the task requires natural language understanding, contextual reasoning, or adaptive behavior. They add cost and complexity when the task is deterministic and rule-based.

The exam categorizes agent use cases into three pillars:

Task Automation — Agents that execute multi-step business processes. Unlike traditional automation (which follows fixed rules), agent-based automation adapts to variations in input. A task agent processing expense reports can handle different formats, extract variable fields, and flag anomalies — where a Power Automate flow would require rigid templates.

When to use agents for automation:

• The process involves unstructured inputs (free-text emails, varied document formats)
• Steps require judgment calls (is this expense reasonable? does this order need manager approval?)
• The workflow has too many edge cases for rule-based automation

When NOT to use agents:

• The process is fully deterministic (if X then Y, always)
• Inputs are structured and predictable (CSV imports, API-to-API data transfers)
• Speed and cost outweigh the need for flexibility

Data Analytics — Agents that analyze data and surface insights through natural language interaction. Instead of building dashboards that users must learn to navigate, analytics agents let users ask questions in plain language: "What were our top-performing products last quarter?" "How does this month's return rate compare to last year?"

Agent-based analytics adds value when:

• Users need ad-hoc analysis that dashboards don't cover
• Data literacy varies across the user base
• Multiple data sources need to be queried and correlated on the fly

Decision-Making — Agents that evaluate options and recommend or execute decisions. The Account Reconciliation Agent in D365 Finance exemplifies this: it analyzes transaction data, matches records against multiple criteria, evaluates confidence levels, and either auto-resolves matches or escalates low-confidence items for human review.

Decision-making agents require the highest level of guardrails because wrong decisions have direct business impact. The architect must define:

• Decision boundaries — what the agent can decide autonomously vs. what requires human approval
• Confidence thresholds — below what certainty level the agent must escalate
• Audit requirements — how decisions are logged for review and compliance

Exam Trap: When a scenario describes a simple, rule-based process (e.g., "route all invoices over $10,000 to the CFO"), don't recommend an AI agent. The correct answer is often Power Automate or a business rule. The exam rewards architects who know when NOT to use AI.

Reflection Question: A logistics company wants to automate shipment tracking notifications. Currently, they send templated emails when shipment status changes. Should this use an AI agent? What would change your answer?