My Background Before Starting
I went into the SAA-C03 with about 18 months of hands-on AWS experience as a backend developer. I had used EC2, S3, RDS, Lambda, and IAM in production, but I had significant gaps in networking (VPC internals, Transit Gateway, Direct Connect), high availability design, and cost optimization strategies.
I had already passed the AWS Cloud Practitioner, which gave me a broad vocabulary, but the SAA-C03 goes much deeper than the Cloud Practitioner on almost every service. Knowing what EC2 is does not prepare you for questions about placement groups, spot fleet configurations, and EC2 instance connect.
Total study time: 5 weeks, roughly 1.5 hours per day on weekdays and 3-4 hours on weekends. Call it approximately 50 hours total. I passed with a score of 784/1000 — comfortable but not exceptional. Here is exactly what I did.
The Study Plan Week by Week
Week 1 focused on the foundation: IAM and VPC. These are the two most tested areas on the exam and the ones where weak knowledge hurts you on questions in every other domain. I spent this entire week on nothing else.
For IAM I studied: the difference between identity-based and resource-based policies, permission boundaries, service control policies in AWS Organizations, cross-account role assumption, the IAM policy evaluation logic (explicit deny always wins, then organizations SCPs, then permission boundaries, then identity policies, then resource policies), and federation with SAML 2.0 and OIDC.
For VPC I studied: how subnets, route tables, and internet gateways connect, the difference between security groups and NACLs (stateful vs. stateless), NAT gateways vs. NAT instances, VPC peering vs. Transit Gateway vs. PrivateLink, VPC endpoints (gateway for S3/DynamoDB, interface for everything else), and the flow logs for troubleshooting.
Week 2 covered compute and storage. EC2 (all instance families, pricing models, placement groups, Auto Scaling lifecycle hooks), Lambda (concurrency, cold starts, execution role, deployment packages, event source mappings), ECS and EKS at a high level, S3 (all storage classes, lifecycle policies, replication, bucket policies, presigned URLs, Transfer Acceleration, multipart upload), EBS (volume types, snapshots, encryption), and EFS.
Week 3 covered databases and application services. RDS (Multi-AZ vs. Read Replicas, parameter groups, automated backups, manual snapshots, encryption at rest), Aurora (Aurora Serverless, global databases, Aurora vs. RDS MySQL performance), DynamoDB (partition key design, GSI vs. LSI, DynamoDB Streams, DAX, on-demand vs. provisioned, TTL), ElastiCache (Redis vs. Memcached use cases), SQS, SNS, EventBridge, and Step Functions.
Week 4 covered networking services and monitoring. CloudFront (origins, behaviors, signed URLs, field-level encryption), Route 53 (all routing policies: simple, weighted, latency, geolocation, geoproximity, failover, multivalue), Direct Connect vs. VPN, Global Accelerator, CloudWatch (metrics, alarms, dashboards, Logs Insights), CloudTrail, AWS Config, and Systems Manager.
Week 5 was pure exam preparation. Three full practice exams on CloudNinja, spaced two days apart, with thorough review of every wrong answer between exams. My scores: 74%, 81%, 86%. When I hit 86% consistently, I booked the exam for two days later.
Resources I Used
Stephane Maarek's SAA-C03 course on Udemy (paid, but always on sale for $15-20): This was my primary resource. Stephane's teaching is structured, clear, and accurately reflects the exam content. I watched every section at 1.25x speed and took notes on key concepts, service comparisons, and scenarios.
AWS documentation (free): I used the official docs for VPC, IAM, RDS, and DynamoDB specifically. When the video course mentioned a concept I did not fully understand, I went to the source documentation. The AWS User Guides are excellent reference material, though they can be dense.
CloudNinja free SAA-C03 practice exam (free): 65 questions with detailed explanations for every answer. I used this as my primary practice exam. The explanations taught me the "why" behind answers, not just the "what." The question style closely matched what I saw on the real exam.
AWS Skill Builder (free): The official practice exam from AWS. I saved this for the final week as a calibration tool. The questions felt slightly easier than my other practice resources, but the interface matches the real exam.
The Topics That Tripped Me Up Early
VPC routing: The most confusing topic early on. Specifically: which route table controls traffic from a subnet to the internet vs. between subnets, and how VPC peering routing works when you have more than two VPCs. I had to draw diagrams and work through specific scenarios before this clicked.
RDS Multi-AZ vs. Read Replicas: I kept confusing these. Multi-AZ is for high availability (automatic failover, synchronous replication, same region). Read Replicas are for performance (asynchronous replication, can be cross-region, must be manually promoted if needed for DR). The exam tests this distinction constantly.
S3 storage classes: The number of storage classes is intimidating — Standard, Intelligent-Tiering, Standard-IA, One Zone-IA, Glacier Instant Retrieval, Glacier Flexible Retrieval, Glacier Deep Archive. The key to remembering them is thinking about two dimensions: access frequency (how often do you read the data?) and retrieval time (how quickly do you need it when you do read it?).
IAM policy evaluation order: The logic for evaluating permissions across organizations, accounts, and principals is complex. Drawing the evaluation flowchart on paper and working through scenarios helped more than reading the docs.
CloudFront vs. Route 53 for latency: Both can improve latency for global users but in different ways. CloudFront caches content at edge locations (reduces origin requests). Route 53 latency routing sends users to the lowest-latency regional endpoint (reduces network distance). The exam tests knowing which to use for which scenario.
Topics Easier Than Expected
Pricing and cost optimization: I was worried about EC2 pricing models (on-demand, reserved, spot, savings plans), but the exam tests conceptual understanding not specific price numbers. If you understand that reserved instances are for steady workloads, spot for fault-tolerant batch jobs, and savings plans for flexible committed spending, you can answer most cost questions correctly.
Well-Architected Framework: The six pillars (Operational Excellence, Security, Reliability, Performance Efficiency, Cost Optimization, Sustainability) appear in questions but the pattern is clear — the question will describe a violation of one pillar and ask how to fix it. Once you internalize what each pillar prioritizes, these are straightforward.
CloudFormation basics: The exam tests high-level CloudFormation concepts, not syntax. You need to know what stacks, templates, change sets, and stack sets are, but you do not need to write CloudFormation YAML.
Exam Day Strategy
I took the exam at a testing center rather than online proctoring. I find it easier to focus without the stress of webcam monitoring and desk checks.
Strategy for the 130 minutes: I read each question carefully and marked any I was not confident about. On the first pass I answered the questions I knew well and flagged the hard ones. On the second pass I worked through the flagged questions. I finished with about 12 minutes to spare.
For hard questions where I was deciding between two plausible answers: I re-read the requirements and looked for the constraint that would eliminate one option. Usually the deciding factor is something like "minimize operational overhead" (eliminates self-managed options), "most cost-effective" (eliminates reserved for variable workloads), or "requires cross-region" (changes which replication/DR solution applies).
Do not change your first answer without a strong reason. I changed three answers in the review and got two of them wrong — my initial instinct was right. Only change an answer if you find a specific error in your reasoning.
What I Would Do Differently
Start with VPC and IAM, not at the end. I covered these in week 1 which was the right call. Do not push networking to the end because it touches every other domain.
Do not try to memorize all service limits and pricing numbers. The exam does not test specific numbers — it tests conceptual understanding of pricing tiers and relative costs. Time spent memorizing that S3 Standard-IA charges $0.0125 per GB is time wasted.
Take at least three full practice exams. I took three and each one revealed new weak spots. Two practice exams is probably not enough for most people.
Study DynamoDB data modeling. I underestimated this. The exam has several questions about when to use a GSI vs. LSI, how to design partition keys to avoid hot partition issues, and when DynamoDB is the right database choice. These are non-trivial if you have not thought carefully about NoSQL data modeling.