🧠 Business Intelligence, Analytics, AI & ML
🧠 Business Intelligence (BI)
Think of a cricket team. Raw data = every ball bowled. BI = the scoreboard, stats, and match reports that help the coach make decisions.
BI is the process of collecting raw business data, transforming it, and presenting it as meaningful information to help managers and executives make better decisions.
Steps in BI
- Data Collection — gather data from sources (databases, CRMs, ERPs)
- Data Storage — store it in a Data Warehouse or Data Mart
- Data Processing — clean and transform via ETL & ELT|ETL
- Data Analysis — run queries, reports, dashboards
- Data Visualization — present via tools like Power BI, Tableau
- Decision Making — business acts on the insights
BI answers "What happened?" — it's backward-looking (historical data).
📊 Analytics
Analytics goes beyond BI — it not only describes what happened but also predicts what will happen.
| Type | Question Answered | Example |
|---|---|---|
| Descriptive | What happened? | Sales dropped 20% last month |
| Diagnostic | Why did it happen? | Discount campaign ended |
| Predictive | What will happen? | Sales will rise 15% next quarter |
| Prescriptive | What should we do? | Run a targeted offer campaign |
BI ≈ Descriptive + Diagnostic. Analytics pushes into Predictive + Prescriptive.
🤖 Artificial Intelligence (AI)
AI is the ability of machines to simulate human intelligence — reasoning, learning, problem-solving.
A human learns from experience. AI does the same — but with data instead of life events.
AI is the umbrella term. Everything below (ML, Deep Learning) lives inside AI.
AI
└── Machine Learning (ML)
└── Deep Learning (DL)
🧬 Machine Learning (ML)
ML is a subset of AI where machines learn patterns from data without being explicitly programmed.
You don't teach a child every dog breed by name. You show them 1000 pictures of dogs. They learn the pattern. That's ML.
Types of ML
1️⃣ Supervised Learning
The model is trained on labelled data (input + correct output).
- Classification — output is a category
Input: Email text → Output: Spam / Not Spam Input: Tumor scan → Output: Malignant / Benign - Regression — output is a number
Input: House size, location → Output: Price (₹) Input: Hours studied → Output: Exam score
Labelled = you already know the answer. You're teaching the model using past correct examples.
2️⃣ Unsupervised Learning
The model is trained on unlabelled data — it finds hidden patterns on its own.
- Clustering — group similar things together
Customer data → Group into: Budget buyers, Premium buyers, Occasional buyers - Association — find rules between items
Supermarket data → "People who buy bread also buy butter" (Market Basket Analysis)
No right answer given. The model discovers structure by itself.
3️⃣ Semi-supervised Learning
Mix of both — small labelled data + large unlabelled data.
Example: Only 500 labelled medical images + 50,000 unlabelled images.
Model uses both to learn better than using only 500 labelled images.
Used when labelling data is expensive or time-consuming (e.g., medical imaging, legal documents).
4️⃣ Reinforcement Learning (RL)
The model learns by trial and error — gets rewards for good actions, penalties for bad ones.
Example: AlphaGo (chess-like game AI)
Self-driving car (reward = safe driving, penalty = crash)
Robot learning to walk
Like training a dog. Good behaviour → treat. Bad behaviour → no treat. Over time it learns what to do.
🔬 Deep Learning (DL)
Deep Learning is a subset of ML that uses neural networks with many layers to learn from massive amounts of data.
ML can recognize a cat from structured features (fur, ears, tail). Deep Learning looks at raw pixels and figures it out itself — like a human brain's visual cortex.
Where Deep Learning shines:
| Use Case | Example |
|---|---|
| Image recognition | Face unlock on your phone |
| Natural Language Processing | ChatGPT, Google Translate |
| Speech recognition | Alexa, Siri |
| Video analysis | YouTube recommendations |
Deep Learning needs massive data and high compute (GPUs). It's overkill for small datasets — use regular ML instead.
🗺️ Full Picture
AI ──────────────────────────────────────────
│
├── Machine Learning
│ ├── Supervised → Classification, Regression
│ ├── Unsupervised → Clustering, Association
│ ├── Semi-supervised
│ └── Reinforcement Learning
│
└── Deep Learning (subset of ML, needs big data + GPU)
🧪 Practice Drill
// Try answering these:
// Q1. A bank wants to predict whether a loan applicant will default or not. Which type of ML is this?
// Q2. An e-commerce platform groups customers based on purchase behavior without any predefined labels. Which type of ML is this?
// Q3. A game AI learns to win by playing millions of matches and getting a score for each win. Which type of ML is this?
// Q4. What's the difference between BI and Analytics in one line each?
// Q5. Arrange these from broadest to most specific: `Deep Learning`, `AI`, `Machine Learning`
💡 Click for Solutions
A1. Supervised Learning — Classification (output is a category: Default / Not Default)
A2. Unsupervised Learning — Clustering (no labels, grouping by pattern)
A3. Reinforcement Learning (reward = win score, learns by trial and error)
A4.
- BI → "What happened?" (historical, backward-looking)
- Analytics → "What will happen / what should we do?" (forward-looking, predictive)
A5. AI → Machine Learning → Deep Learning
← Previous Topic | Next Topic → Next Topic