Decision Trees — When to Use What
This page consolidates all the "when to use what" guidance into a single reference. Use these decision tables when you need to quickly identify the right AWS service for a given ML scenario.
Data Preparation
| Scenario | Recommended Service |
|---|---|
| No-code data prep (non-technical users) | Glue DataBrew |
| Visual data prep feeding into SageMaker | SageMaker Data Wrangler |
| Feature importance scores | Data Wrangler Quick Model |
| Batch ETL at scale (PySpark) | AWS Glue ETL |
| Schema discovery from S3 | Glue Crawlers → Glue Data Catalog |
| Fuzzy matching / deduplication | Glue FindMatches |
| Format conversion: CSV → Parquet (batch) | Glue ETL job |
| Format conversion: JSON → Parquet (streaming) | Kinesis Firehose (native conversion via Glue Catalog) |
| Format conversion: images → RecordIO | im2rec (MXNet utility) |
Data Ingestion
| Scenario | Recommended Service |
|---|---|
| Stream to S3 with transforms (least effort) | Kinesis Data Firehose + Lambda |
| Real-time custom processing with replay | Kinesis Data Streams |
| SQL on streaming data | Kinesis Data Analytics |
| Real-time anomaly detection | KDA + Random Cut Forest |
| On-premises → S3 with scheduling | AWS DataSync |
| Petabytes + slow network | Snowball Edge |
| RDS/DynamoDB → S3 for ML | Glue / Data Pipeline / DMS |
Model Training
| Scenario | Recommended Service |
|---|---|
| Standard ML training | SageMaker Training Jobs (data from S3) |
| AutoML with no expertise | SageMaker Autopilot or Canvas |
| Reduce training cost | Spot Instances + Checkpointing |
| Faster data loading | Pipe mode (RecordIO) or FastFile mode |
| Distributed training | Horovod (data parallel) or SageMaker distributed |
| Custom framework or algorithm | ECR Docker → SageMaker BYOC |
| Hyperparameter tuning | SageMaker Automatic Model Tuning |
Model Deployment
| Scenario | Recommended Service |
|---|---|
| Steady real-time traffic | SageMaker Real-time Endpoint |
| Intermittent / unpredictable traffic | SageMaker Serverless Inference |
| Periodic bulk predictions | SageMaker Batch Transform |
| Large payloads (up to 1 GB) | SageMaker Async Inference |
| A/B testing models | Production Variants on a single endpoint |
| Edge deployment (no internet) | Neo → IoT Greengrass |
| Choose best instance type | SageMaker Inference Recommender |
Monitoring and MLOps
| Scenario | Recommended Service |
|---|---|
| Data drift / model quality monitoring | SageMaker Model Monitor |
| Training-time debugging | SageMaker Debugger |
| Model explainability | SageMaker Clarify (SHAP values) |
| ML workflow automation | SageMaker Pipelines |
| Model versioning / approval | SageMaker Model Registry |
| Human review of predictions | Amazon A2I |
| Automated retraining on drift | Model Monitor → EventBridge → Pipelines |
Security
| Scenario | Recommended Service |
|---|---|
| SageMaker access control | IAM execution roles (never access keys) |
| No internet for SageMaker | VPC + S3 Gateway Endpoint + VPC Interface Endpoints |
| Encrypt at rest | KMS CMK |
| Data lake governance | Lake Formation (column-level access) |
| Audit API calls | CloudTrail |
| Monitor resource metrics | CloudWatch |
AI Service Selection (No ML Expertise Needed)
| Task | Recommended Service |
|---|---|
| Text classification | Comprehend Custom Classifier |
| Sentiment / entities / key phrases | Comprehend |
| Image classification (custom) | Rekognition Custom Labels |
| Face detection / matching | Rekognition |
| Extract text from documents | Textract |
| Speech to text | Transcribe |
| Text to speech | Polly |
| Language translation | Translate |
| Chatbots / voice bots | Lex |
| Time-series forecasting (managed) | Forecast |
| Product recommendations | Personalize |
| Fraud detection | Fraud Detector |
| Enterprise search | Kendra |
| Generative AI / foundation models | Bedrock |
| Equipment anomaly detection | Lookout for Equipment |
Flashcards
1 / 10
Question
You need to detect anomalies in a real-time data stream. What AWS services do you use?
Click to revealAnswer
Kinesis Data Analytics with its built-in Random Cut Forest (RCF) function. This is the standard pattern for streaming anomaly detection.
Pro Tip
When evaluating which service to use, consider the effort spectrum. Managed AI services (Comprehend, Rekognition, etc.) require the least effort. SageMaker built-in algorithms are next. Custom SageMaker training (BYOC) requires the most effort but gives the most control. Match the level of effort to the complexity of your problem.