I85PREP
⌘ K
AB-100 study materials
AB-100 study resources

AB-100: Plan AI-Powered Business Solutions

AB-100 · Work through your exam, one topic at a time.

All guide sections

Plan AI-Powered Business Solutions

Start from the business outcome, not the model

Planning questions test whether you can decide whether and where an agent belongs before anything is designed. Expect scenarios that hand you a process description and ask for the correct next step, the right success metric, or the right target workload.

Key concepts:

  • Value assessment: baseline the current process (cycle time, cost per transaction, containment/deflection rate, error rate) before defining targets. A plan without a baseline cannot prove value later.
  • Use-case triage: score candidates on business impact, data readiness, technical feasibility, and risk. High-volume, high-variance, language-heavy work is agent-shaped; deterministic, rule-based, structured work belongs in classic automation (Power Automate, RPA) and should stay there.
  • Data readiness: name the systems of record, confirm that grounding content exists and is current, understand its permissions model, and identify the owner.
  • Responsible AI and governance: classify data sensitivity, map regulatory obligations (residency, retention), define the human-in-the-loop boundary, and record all of it in the plan.
  • Cost and licensing: agent capacity/message consumption, Microsoft 365 Copilot licensing, and Azure consumption for custom models - model these before committing.
  • Build vs. extend vs. buy: extend a first-party copilot, compose in Copilot Studio, or build custom on Azure AI Foundry.

Traps: recommending an agent because the technology is available rather than because the process fits; skipping the data and permissions assessment and discovering the knowledge source is unusable; treating a pilot backlog as the roadmap; quoting benefits with no baseline.

Rule of thumb - the determinism test: if you can write the process as a flow chart with no judgment calls, do not build an agent.

Worked example: a service desk gets 10,000 tickets a month. 45% are password resets with a fixed procedure - automate with a flow. 35% are "how do I…" questions answered in SharePoint - an agent grounded on that content with an escalation path. 20% are complex incidents - human handling with agent-assisted summarization. Three technologies, one baseline.