6.1. Exam Strategy
The AI-200 delivers 40–60 questions in roughly 100–120 minutes — comfortable time if you don't donate it to case studies early. The formats each want a different tactic.
Single-answer multiple choice rewards constraint-hunting: read the last sentence first (the actual question), then scan the scenario for the keyword that eliminates services — "scale to zero," "kubectl," "guaranteed processing," "no cold starts." Phase 1.2's axis does the rest. Multiple response questions state how many answers to pick; each option is an independent true/false claim, so judge them separately — don't let one confident pick drag a neighbor in. Drag-and-drop / ordering questions are usually pipelines (KQL stations, deployment steps, rotation flows); anchor the endpoints first (table always leads a KQL query; push ends an ACR task) and the middle sorts itself. Case studies present one scenario feeding several questions: skim the requirements list once, note constraint keywords in the margin of your mind, and answer each question against stated requirements — not against what a sensible architect might add.
Three cross-cutting tactics. First, flag and return: any question consuming more than two minutes gets flagged; a later question often reminds you of the fact you were missing. Second, distractor patterns repeat — this exam's favorites are the plausible-but-wrong-species answer (Event Grid where Service Bus belongs), the works-but-wasteful answer (scale up instead of decouple; strong consistency "to be safe"), and the feature-that-doesn't-exist (auto-replaying DLQs, self-rotating secrets, multi-trigger functions). Third, when two answers both seem right, re-read for the cost or operations qualifier — "minimize RU consumption," "least operational overhead" — it always breaks the tie.
Score mechanics favor attempting everything: 700/1000 scaled, no penalty for wrong answers. Leave nothing blank.
