Skip to content

Task Statement 1.2: Identify Practical Use Cases for AI

AWS AIF-C01 Exam Focus: Task 1.2 tests your ability to evaluate business problems and select the right AI/ML technology or AWS managed service. You must recognize when AI/ML provides tangible value versus when traditional rule-based software is better, choose the correct ML technique (classification, regression, clustering), map real-world workloads to managed AWS services, and decide between traditional ML and Foundation Models (FMs).

1. Business Value Drivers & When AI/ML is NOT Appropriate

Primary Value Drivers of AI/ML

  • Automating Repetitive Tasks: Replacing manual data entry, image categorization, or document routing with automated models to increase throughput and reduce operational cost.
  • Assisting Human Decision-Making (Augmented Intelligence): Providing predictive scores or draft content to human experts (Human-in-the-Loop) for higher quality and faster decisions (e.g., assisting radiologists with medical image flags or loan officers with risk scores).
  • Scaling Operations: Handling massive increases in request volume without linearly increasing headcount (e.g., processing millions of customer support tickets automatically).
  • Hyper-Personalization: Adapting real-time user experiences, product recommendations, and search results based on historical behavior patterns.

When AI/ML is NOT Appropriate (Exam Scenarios)

  1. Deterministic Outcomes Required: When rules are 100% fixed, regulated, or require exact mathematical outcomes without probabilistic variation (e.g., calculating sales tax, standard accounting ledger updates). Use traditional, rule-based software (if/else logic).
  2. Simple Rule-Based Systems Suffice: If a business process can be solved with a few explicit rules or standard SQL queries, introducing ML adds unnecessary complexity, cost, and maintenance overhead.
  3. Insufficient or Poor Quality Data: ML models require representative data to learn. If data is unavailable, unlabelled (when labels are required), or severely corrupted, ML will fail ("garbage in, garbage out").
  4. Unfavorable Cost-Benefit Ratio: When the cost of building, training, hosting, and maintaining an AI model outweighs the incremental business value generated.
  5. Zero-Tolerance for Errors or Inexplicable Logic: In environments where absolute 100% accuracy is required and probabilistic errors/hallucinations are unacceptable, probabilistic models (especially GenAI) should not be used autonomously.

2. Selecting the Right ML Technique for Specific Use Cases

When traditional ML is required, you must map the business question to the correct mathematical technique.

Business Question ML Technique Target Output Type Typical Real-World Scenario
"Is this transaction fraudulent or legitimate?" Binary Classification Discrete Category (Class 0 or 1) Credit card fraud detection, Spam vs. Non-Spam filtering
"Which category does this ticket belong to?" Multiclass Classification Discrete Category ($>2$ classes) Customer support ticket routing (Billing, Technical, Account)
"What will the price of a house be next month?" Regression Continuous Numerical Value House price estimation, revenue projection, temperature prediction
"How can we group our un-categorized customers?" Clustering Group / Cluster Assignment Market basket segmentation, targeted ad audience creation
"Is this network traffic pattern unusual?" Anomaly Detection Binary / Anomaly Score Cybersecurity intrusion detection, manufacturing equipment failure detection
"How many units of Product X will sell next week?" Time-Series Forecasting Ordered Temporal Numeric Values Retail inventory demand forecasting, server load projection

3. Real-World AI Application Domains

1. Computer Vision (CV)

Extracting spatial boundaries, visual information, and semantic context from images or video streams.

  • Examples: Automated quality inspection in manufacturing, facial identification, document text extraction (OCR), autonomous driving obstacle detection.

2. Natural Language Processing (NLP) & Speech

Analyzing, understanding, translating, and generating human spoken or written language.

  • Examples: Sentiment analysis on social media reviews, real-time audio transcription, multi-language document translation, conversational AI chatbots.

3. Recommendation Systems

Analyzing past user behavior, demographic similarity, and item metadata to predict future preference.

  • Examples: E-commerce "frequently bought together" suggestions, media streaming content recommendations.

4. Fraud Detection & Risk Modeling

Analyzing transactional metadata in real-time to identify anomalous behavior or assess risk probability.

  • Examples: Flagging stolen credit card transactions, automated insurance claims assessment.

5. Enterprise Knowledge Bases (RAG)

Connecting Generative AI models to private internal data stores so users can search and query company documents using natural language with source citations.

  • Examples: Internal HR policy search tools, legal contract analysis platforms.

6. Agentic AI

Deploying multi-step autonomous workflows where AI plans tasks, calls external systems via APIs, updates databases, and completes complex end-to-end business operations without continuous human intervention.

4. AWS Managed AI/ML Services Capabilities

AWS provides pre-trained managed AI services that allow developers to add intelligence to applications via simple API calls without building or training custom models.

Vision & Document Processing

  • Amazon Rekognition: Pre-trained computer vision service for image and video analysis. Identifies objects, text, faces, celebrities, and inappropriate content (moderation).
  • Amazon Textract: Specialized Intelligent Document Processing (IDP) service. Automatically extracts printed text, handwriting, fields from forms, and tables from scanned PDF documents.

Language & Speech

  • Amazon Comprehend: Natural Language Processing (NLP) service that extracts insights, sentiment, key phrases, language, entities, and personally identifiable information (PII) from text.
  • Amazon Transcribe: Automatic Speech Recognition (ASR) service that converts spoken audio into accurate written text.
  • Amazon Polly: Text-to-Speech (TTS) service that converts written text into lifelike spoken audio.
  • Amazon Translate: Neural machine translation service providing fast, high-quality language translation.
  • Amazon Lex: Service for building conversational interfaces (chatbots and voicebots) using voice and text (powers Amazon Alexa).

Specialized Industry ML Services

  • Amazon Personalize: Managed recommendation engine based on the same technology used by Amazon.com for real-time personalized recommendations.
  • Amazon Fraud Detector: Fully managed service that uses ML to identify potentially fraudulent online activities (e.g., fake account creation, payment fraud).
  • Amazon Forecast: Managed time-series forecasting service that uses ML to deliver accurate business forecasts (inventory, supply chain).

Platform & Generative AI Managed Solutions

  • Amazon SageMaker AI: Full platform for building, training, tuning, deploying, and monitoring custom traditional ML models and hosted FMs.
  • Amazon Bedrock Knowledge Bases: Fully managed RAG workflow that automatically chunks documents, generates embeddings, stores them in a vector database, and grounds FM responses in enterprise data.
  • Amazon Q: Generative AI-powered assistant for business (Amazon Q Business) and software development (Amazon Q Developer).

5. Traditional ML vs. Foundation Models (FMs): Decision Framework

Selecting between Traditional ML models and Foundation Models is a core decision evaluated on the AIF-C01 exam.

Dimension Traditional ML Models Foundation Models (GenAI / FMs)
Model Scope Single, narrow task (e.g., churn prediction) Broad, general-purpose (e.g., text, code, images)
Primary Output Numbers, probabilities, categorical tags Synthetic text, code, audio, images, multi-step actions
Data Requirements Requires large, high-quality labeled dataset Pre-trained; requires zero or few examples (In-Context Learning)
Explainability High (Decision trees, regression coefficients, feature importance via SageMaker Clarify) Low ("Black box" neural networks; subject to hallucinations)
Cost & Latency Low compute cost; low latency (milliseconds) Higher compute cost (token pricing); higher latency (seconds)
Customization Retraining / Hyperparameter Tuning Prompt Engineering, RAG, Fine-Tuning

Key Selection Criteria

1. Regulatory & Explainability Requirements

  • Choose Traditional ML if strict regulations require clear explanation of why a decision was made (e.g., credit card loan denial explanation required by law).
  • Avoid FMs in strict white-box audit scenarios due to lower interpretability and potential hallucinations.

2. Task Modality & Flexibility

  • Choose Foundation Models if the requirement involves freeform text synthesis, multi-lingual summarization, unstructured document reasoning, or multi-modal inputs.
  • Choose Traditional ML if the input is structured tabular data in SQL/CSV and the output is a discrete numerical prediction.

3. Operational Constraints & Cost

  • Choose Traditional ML for high-throughput, low-latency, low-cost microservice tasks (e.g., evaluating 10,000 requests per second for simple churn probability).
  • Choose FMs when user interaction value offsets higher latency and token-based pricing (e.g., draft email responses for customer service agents).

6. Exam Decision Rules & Common Pitfalls

High-Yield Decision Rules

  1. Select Amazon Comprehend if the scenario mentions analyzing customer review sentiment, extracting entities, or redacting PII from text.
  2. Select Amazon Textract if the scenario mentions extracting structured forms, tables, or handwriting from scanned PDF documents (do NOT choose Rekognition for document forms).
  3. Select Amazon Rekognition if the scenario mentions detecting objects, faces, or content moderation in raw photos or videos.
  4. Select Amazon Transcribe if the requirement is converting recorded call-center audio into text transcripts.
  5. Select Amazon Polly if the requirement is converting written news articles into audio streams.
  6. Select Amazon Lex if the scenario mentions building a conversational chatbot or IVR voice response system.
  7. Select Traditional ML (SageMaker) over FMs if the requirement demands 100% explainable credit scoring on tabular data with low latency.
  8. Select Rule-Based Software (No ML) if the business problem involves fixed, deterministic calculations (e.g., tax computation, static discount matrix).

Common Exam Traps

  • Trap 1: Over-Engineering with Custom Models
  • If a scenario asks to extract sentiment or transcribe audio with minimal operational overhead, do NOT choose "Train a custom model in SageMaker". Choose the pre-trained managed service (Amazon Comprehend or Amazon Transcribe).
  • Trap 2: Textract vs. Rekognition Misallocation
  • Rekognition is for general computer vision (objects, faces). Textract is explicitly built for document processing (tables, forms, text in PDFs).
  • Trap 3: Forcing ML on Deterministic Logic
  • Watch for questions offering an ML model as an option for tasks that are better solved by standard database queries or simple if/else rules. Always select the non-ML option if the outcome is strictly deterministic.