Domain 1 of 5 · Chapter 2 of 3

AI Use Cases

Unlock the complete study guide + 1,040 practice questions across 16 full exams.

Bundled into the existing AWS Certified AI Practitioner AIF-C01 premium course — no separate purchase.

Included in this chapter:

  • When AI/ML earns its cost vs. deterministic code
  • Naming the ML problem type from the output shape
  • Mapping the use case to the right AWS managed AI service
  • Exam-pattern recognition: stem keywords → service, and why distractors fail

Choosing an AWS AI/ML tier for a use case

TierWhat it isML expertise neededCustomizationBest-fit use case
Pre-trained AI servicesReady-made APIs: Comprehend, Rekognition, Transcribe, Translate, Polly, Textract, Lex, Personalize, Forecast, Fraud Detector, KendraNone - call an API, no model to trainLimited; some services support custom labels or domain tuningCommon tasks (sentiment, OCR, speech, translation, recommendations) where a generic capability is enough
Amazon Bedrock (foundation models)Serverless access to third-party and Amazon foundation models for generative AILow; prompt engineering, optional fine-tuning and RAG, no infrastructureMedium; in-context prompts, RAG, and fine-tuning without managing serversGenerative tasks: text/chat generation, summarization, code assistance, conversational agents
Amazon SageMaker AIFull platform to build, train, tune, and host custom ML modelsHigh; data science and ML engineering skills requiredFull; any algorithm, custom data, full control of training and hostingBespoke models where no managed service fits, or specialized accuracy and control needs

Decision tree

Generative task? draft / summarize / chat / code Yes Amazon Bedrock foundation models via API No Text or documents? sentiment, entities, scanned forms Yes Comprehend / Textract NLP insight / document extraction No Image or video? objects, faces, moderation Yes Amazon Rekognition image & video analysis No Speech in or out? audio to text, text to voice Yes Transcribe / Polly speech↔text No Personalized picks? product / content recommendations Yes Amazon Personalize real-time recommendations No Time-series forecast? demand / capacity planning Yes Amazon Forecast managed time-series forecasting No Amazon SageMaker AI build a custom model (last resort)
Cheat sheet

Cheat sheet

  • Use ML only when rules are learned, not written
  • Don't use ML when an exact guaranteed outcome is required
  • No representative data disqualifies ML
  • Mandated full explainability can disqualify ML
  • Cost-benefit can favor not using ML
  • Classify the ML problem from the output shape
  • Supervised (classification/regression) vs unsupervised (clustering)
  • Forecasting is time-series, not generic regression
  • Default to the highest-level AWS AI service
  • Comprehend analyzes text (NLP)
  • Textract extracts text from documents (OCR)
  • Rekognition analyzes images and video (CV)
  • Transcribe vs Polly are mirror services
  • Lex builds chatbots and voice assistants
  • Personalize delivers recommendations
  • Fraud Detector for managed online-fraud detection
  • Kendra for natural-language enterprise search
  • Bedrock for generative AI; Q for ready assistants
  • SageMaker AI for custom models only when needed
  • Three distractor archetypes in service-selection stems
  • Comprehend Targeted Sentiment is entity-level, not document-level
  • Comprehend entity recognition tags standard types out of the box
  • Custom entity recognition for domain-specific terms

Unlock with Premium — includes all practice exams and the complete study guide.

Also tested in

References

  1. Machine Learning on AWS
  2. Amazon Comprehend
  3. Amazon Textract
  4. Amazon Rekognition
  5. Amazon Transcribe
  6. Amazon Polly
  7. Amazon Translate
  8. Amazon Lex
  9. Amazon Personalize
  10. Amazon Fraud Detector
  11. Amazon Kendra
  12. Amazon Bedrock
  13. Amazon SageMaker AI