6.1. Exam Strategy and Time Management
You have 170 minutes for 65 questions—roughly 2 minutes 37 seconds per question. However, not all questions take equal time. Multiple choice questions with straightforward scenarios may take 1 minute, while case study questions with multi-paragraph scenarios may take 5+ minutes. Budget your time accordingly.
First Pass (90 minutes): Work through all 65 questions. Answer questions you're confident about immediately. For questions that require deliberation, eliminate obviously wrong answers, make your best guess, and flag the question for review. Never leave a question blank—there's no penalty for guessing.
Second Pass (60 minutes): Return to flagged questions. With the pressure of unanswered questions removed, you'll often see the answer more clearly. Re-read the scenario carefully—exam writers embed critical constraints in specific phrases.
Buffer (20 minutes): Reserve this for case study questions that require re-reading shared scenarios, and for final review of any questions where you changed your answer.
Question Approach Framework
For every question, follow this mental framework:
-
Identify the lifecycle stage. Is this about data prep (Domain 1), model development (Domain 2), deployment (Domain 3), or monitoring/security (Domain 4)?
-
Identify the primary constraint. What is the scenario optimizing for—cost, latency, security, compliance, scalability, or simplicity?
-
Eliminate the obviously wrong. Usually 2 of 4 choices can be eliminated immediately—they use the wrong service for the lifecycle stage or violate the stated constraint.
-
Choose between the remaining two. The last two options are usually both plausible—the correct one better satisfies the specific constraint in the question.
Reading the Question Stem
Exam questions embed critical information in specific phrases. Train yourself to spot these signals:
| Signal Phrase | What It's Testing | Likely Answer Direction |
|---|---|---|
| "Minimize operational overhead" | Managed vs. custom | Use managed services (AI services, built-in algorithms) |
| "Most cost-effective" | Pricing optimization | Spot for training, serverless for low traffic, Savings Plans for steady |
| "Minimize latency" | Real-time performance | Real-time endpoints, GPU, provisioned compute |
| "Ensure compliance" | Security/regulatory | VPC mode, KMS CMK, network isolation, CloudTrail |
| "Without retraining the model" | Production fix | Data pipeline fix, feature engineering, preprocessing alignment |
| "Gradually shift traffic" | Deployment strategy | Canary or linear deployment |
| "With the ability to roll back" | Safe deployment | Blue/green deployment |
| "Detect changes in data" | Monitoring | SageMaker Model Monitor (data quality) |
| "Detect bias" | Fairness | SageMaker Clarify |
| "Debug training issues" | Training convergence | SageMaker Debugger |
| "Automate the ML workflow" | Orchestration | SageMaker Pipelines |
| "With minimal code" | Low-code tooling | Data Wrangler, AutoML, AI Services |
Flag-and-Return Strategy
Flag questions when:
- The scenario is longer than 4 lines (case study—come back with fresh eyes)
- Two answers seem equally valid (you may find a disambiguating clue on re-read)
- The question references a service you're unsure about (don't waste 5 minutes; guess and move on)
Never flag more than 15 questions. If you're flagging more, you may be overthinking straightforward questions.
