Identify Practical Use Cases for AI

🤖 Exam Guide: AI Practitioner
Domain 1: Fundamentals of AI and ML
📘Task Statement 1.2

🎯 Objectives

This task is about deciding when AI is useful, when it isn’t, and matching common business problems to the right ML technique and AWS managed AI service.

---

1) Where AI/ML Can Provide Value

AI/ML tends to add the most value when you need to:

1.1 Assist Human Decision-Making:

Examples:

1. underwriting support
2. clinical triage support
3. prioritizing support tickets
4. sales lead scoring

1.2 Scale Decisions That Humans Can’t Make Repeatedly

Examples:

1. moderating content at scale
2. routing millions of requests
3. ranking products for every user

1.3 Automate Pattern Recognition

Examples:

1. detecting defects in manufacturing images
2. identifying fraud patterns
3. extracting entities from documents

1.4 Personalize Experiences

Examples:

1. recommendations
2. targeted offers
3. personalized search ranking

1.5 Improve Predictions and Forecasting

Examples:

1. demand forecasting
2. inventory planning
3. ETA prediction
4. churn probability

1.6 Handle Unstructured Data

Examples:

1. text (emails, chats)
2. audio (call center)
3. images/video (security footage)

AI is strong at probabilistic prediction and pattern discovery, not guaranteed correctness.

---

2) When AI/ML is Not Appropriate

AI may be the wrong choice when:

2.1 A Deterministic Answer Is Required

If you need a guaranteed correct output every time (e.g., tax calculation rules, compliance logic), use rules/logic, not ML.

2.2 There’s No Data or Poor-Quality Data

No representative historical data → models won’t generalize.

2.3 Cost or Complexity Outweighs Benefit

ML introduces ongoing costs:

1. data collection
2. monitoring
3. drift
4. retraining
5. governance.

2.4 The Process is Stable and Well-Defined

Simple rules and workflows can be cheaper, faster, and more explainable.

2.5 High-Risk Decisions Require Strict Explainability

In regulated contexts, a simpler model or non-ML approach might be required for auditability.

2.6 You Can’t Tolerate Model Error

ML outputs are probabilistic, even high accuracy still means some mistakes.

Cost-Benefit Framing

If a rules-based solution can meet requirements with low maintenance, prefer it.
Use ML when the pattern is complex, data-driven, and value justifies ongoing lifecycle management.

---

3) Select The Appropriate ML Technique For A Use Case

Mapping

Use case	Typical technique	Output
Predict house price, demand, revenue	Regression	Number (continuous)
Spam detection, churn yes/no, fraud yes/no	Classification	Category/label
Segment customers, group similar products	Clustering (unsupervised)	Groups/segments
Detect unusual transactions/sensors	Anomaly detection	Outlier score/flag
Forecast sales over time	Time-series forecasting	Future values
Rank products or pages	Recommendation / ranking	Ordered list

Quick Decision Guide

1. - “How much / how many?” → Regression
2. - “Which category?” → Classification
3. - “Find groups with no labels” → Clustering
4. - “Find weird/rare events” → Anomaly detection
5. - “Predict the future from time” → Forecasting

---

4) Examples of Real-World AI Applications

4.1 Computer vision

1. Defect detection
2. face blur
3. safety monitoring
4. OCR + form extraction

4.2 NLP

1. Sentiment analysis
2. entity extraction
3. ticket classification
4. summarization
5. search

4.3 Speech Recognition

1. Transcribing calls
2. meeting notes
3. call analytics

4.4 Recommendation systems

1. “Customers also bought”
2. content feeds
3. personalized search ranking

4.5 Fraud detection

Flag suspicious payments or account behavior

4.6 Forecasting

1. Predict demand
2. staffing needs
3. inventory
4. traffic
5. energy usage

---

5) Capabilities of AWS managed AI/ML Services

These AWS services let you use AI without building everything from scratch.

Core Services

5.1 Amazon SageMaker AI

Managed platform to build, train, and deploy ML models
Amazon Sagemaker AI includes tools for:

1. data prep
2. training jobs
3. endpoints for inference
4. monitoring

5.2 Amazon Transcribe

Speech-to-text service that convert audio to text

Amazon Transcribe Common Use Cases:

1. call center transcription
2. meeting transcription
3. voice analytics pipelines

5.3 Amazon Translate

Machine translation between languages

Amazon Translate Common Use Cases:

1. multilingual support content
2. translation of chats/tickets

5.4 Amazon Comprehend

NLP for text analysis
Amazon Comprehend Common Features:

1. sentiment
2. key phrases
3. entities
4. language detection
5. topic modeling/classification

5.5 Amazon Lex

Build chatbots and voice bots
Amazon Lex understands user intent and manages conversation flows (often paired with Transcribe/Polly in voice scenarios)

5.6 Amazon Polly

Text-to-speech service that converts text into lifelike speech
Amazon Polly Common Use Cases:

1. voice assistants,
2. reading content aloud,
3. IVR (Interactive Voice Response) systems

Mapping

Need	Service
Train/deploy your own ML model	SageMaker
Convert audio → text	Transcribe
Convert text between languages	Translate
Extract meaning from text (sentiment/entities)	Comprehend
Build a chatbot / conversational interface	Lex
Convert text → spoken audio	Polly

---

💡 Quick Questions

1. Predict whether a transaction is fraud (yes/no): _?
2. Group customers by behavior with no labels: _?
3. Convert a recorded call into text: _?
4. Create a customer support chatbot: _?
5. Predict next month’s demand: _?

Additional Resources

1. Machine Learning (ML) and Artificial Intelligence (AI)
2. AWS AI: Innovation that drives results faster
3. Choosing an AWS machine learning service
4. Choosing an AWS generative AI service
5. Amazon Bedrock or Amazon SageMaker AI?

✅ Answers to Quick Questions

1. Predict whether a transaction is fraud (yes/no): Classification
2. Group customers by behavior with no labels: Clustering (unsupervised learning)
3. Convert a recorded call into text: Amazon Transcribe
4. Create a customer support chatbot: Amazon Lex
5. Predict next month’s demand: Time-series forecasting