Business Intelligence: Data, Analytics & Decision Support

Core Concepts of Business Intelligence (BI)

Business Intelligence (BI) is the practice of collecting, analyzing, and presenting business data to support better decision-making. It bridges raw data and strategic actions by leveraging technology, statistics, and domain expertise.

1. Data vs. Information vs. Knowledge

mindmap
  root((Business Intelligence))
    Data["Raw Facts (e.g., sales records, customer IDs)"]
    Information["Processed Data (e.g., 'Sales dropped 10% in Q2')"]
    Knowledge["Actionable Insights (e.g., 'Reduce stock of Product X')"]
    Wisdom["Strategic Decisions (e.g., 'Expand to Pokhara')"]

• Data: Unprocessed facts (e.g., customer transactions in eSewa).
• Information: Structured data with context (e.g., "50% of eSewa users are under 30").
• Knowledge: Patterns derived from analysis (e.g., "Young users spend more on mobile recharges").
• Wisdom: Strategic use of knowledge (e.g., "Launch targeted promotions for under-30 users").

2. Components of Business Intelligence

BI systems integrate hardware, software, data, and human processes. Key components:

3. Data Warehousing: The Backbone of BI

A data warehouse is a subject-oriented, integrated, time-variant, and non-volatile repository that stores data from multiple sources for reporting and analysis.

How Data Warehouses Work

flowchart TD
  A["Source Systems\n(e.g., ERP, CRM, IoT)"] -->|"ETL Process"| B["Staging Area\n(Cleaning, Transformation)"]
  B --> C["Data Warehouse\n(Raw Data Storage)"]
  C --> D["Data Marts\n(Department-Specific)\n(e.g., Sales, HR)"]
  D --> E["BI Tools\n(Tableau, Power BI)"]
  E --> F["User Dashboards\n(Executives, Managers)"]

Key Features:

• Subject-Oriented: Organized by business functions (e.g., "Sales," "Finance").
• Integrated: Consolidates data from disparate sources (e.g., Daraz’s website + mobile app data).
• Time-Variant: Stores historical data (e.g., NTC’s call logs from 2010–2024).
• Non-Volatile: Data is never updated; only new data is added.

Worked Example: Nabil Bank’s Loan Default Prediction

Scenario: Nabil Bank wants to predict loan defaults using BI.

1. Data Sources:
   • Customer transaction history (from core banking system).
   • Credit scores (from CIBIL Nepal).
   • Demographic data (age, income, location).
2. ETL Process:
   • Extract data from banking software.
   • Transform: Clean missing values (e.g., "income = NULL" → "income = average").
   • Load into a data mart for credit risk analysis.
3. Data Mining:
   • Use decision trees (a classification algorithm) to identify patterns:
     • If (income < 50K AND loan_amount > 10L) → High default risk.
4. Output:
   • A dashboard flags high-risk applicants, reducing defaults by 20%.

Advantages of Data Warehousing:

• Single Version of Truth: Avoids discrepancies (e.g., Daraz’s inventory vs. sales data).
• Historical Analysis: Tracks trends (e.g., "Khalti transactions spike during Dashain").
• Faster Queries: Optimized for reporting (vs. OLTP systems like MySQL).

Disadvantages:

• High Cost: Requires storage and maintenance.
• Complexity: ETL processes can be time-consuming.

4. OLAP vs. OLTP: The Two Sides of Data Processing

5. Data Mining: Uncovering Hidden Patterns

Data mining uses statistical techniques and machine learning to extract patterns from large datasets.

Common Data Mining Techniques

mindmap
  root((Data Mining Techniques))
    Classification["Categorize data (e.g., 'High/Low Risk')"]
    Clustering["Group similar data (e.g., customer segments)"]
    Association["Find relationships (e.g., 'People who buy X also buy Y')"]
    Prediction["Forecast outcomes (e.g., 'Stock price in 2025')"]
    Text Mining["Analyze unstructured data (e.g., customer reviews)"]

Example in Nepal:

• Pathao’s Ride Demand Prediction:

  • Uses clustering to group high-demand areas (e.g., Thapathali during rush hour).
  • Deploys more drivers in these zones, reducing wait times by 30%.

• Daraz’s Market Basket Analysis:

  • Association Rule Mining: "Customers who buy laptops also buy chargers (80% probability)."
  • Action: Place chargers near laptops in warehouses to boost sales.

6. Decision Support Systems (DSS)

A Decision Support System (DSS) is an interactive system that helps managers make decisions by combining data, models, and user input.

Types of DSS

flowchart TD
  A["Decision Support Systems"] --> B["Model-Driven DSS"]
  A --> C["Data-Driven DSS"]
  A --> D["Document-Driven DSS"]
  A --> E["Knowledge-Driven DSS"]
  B -->|"Example"| F["Nabil Bank’s Loan Approval Model"]
  C -->|"Example"| G["NTC’s Network Traffic Analysis"]
  D -->|"Example"| H["Nepal Police’s Crime Report Database"]
  E -->|"Example"| I["Himalayan Java’s Supply Chain Expert System"]

Worked Example: NTC’s Network Outage Prediction

1. Data Input:
   • Historical outage data (from SCADA systems).
   • Weather data (rainfall, temperature).
   • Maintenance logs.
2. Model:
   • Regression Analysis: Predicts outages based on past patterns.
   • Equation: Outage_Risk = β₀ + β₁(Rainfall) + β₂(Temperature) + ε
3. Output:
   • A dashboard alerts engineers to high-risk areas before failures occur.
   • Result: Reduced outages by 15% in monsoon season.

Advantages of DSS:

• Reduces uncertainty (e.g., banks avoid risky loans).
• Speeds up decisions (e.g., Daraz’s dynamic pricing).
• Improves accuracy (e.g., NTC’s predictive maintenance).

Disadvantages:

• High initial cost (requires expert modeling).
• Over-reliance on data quality (garbage in, garbage out).

7. Business Intelligence Tools

Popular BI tools used in Nepal and globally:

In the Real World

1. eSewa’s Fraud Detection

   • Idea Used: Data Mining (Anomaly Detection)
   • How: eSewa’s BI system flags unusual transactions (e.g., a user suddenly transferring 500K in one go) using machine learning models.
   • Impact: Reduced fraud cases by 40% in 2023.

2. Pathao’s Surge Pricing Algorithm

   • Idea Used: OLAP + Predictive Analytics
   • How: Pathao’s DSS analyzes real-time demand (from GPS data) and adjusts prices dynamically, like Uber.
   • Impact: Maximizes driver earnings during peak hours (e.g., 8–10 PM in Kathmandu).

3. Nabil Bank’s Customer Segmentation

   • Idea Used: Clustering (K-Means Algorithm)
   • How: Nabil Bank groups customers into segments (e.g., "Premium," "Budget," "New") based on spending habits.
   • Impact: Tailored loan offers increase approval rates by 25%.

4. Daraz’s Inventory Optimization

   • Idea Used: Association Rules + Forecasting
   • How: Daraz’s BI system predicts stock needs using past sales and seasonality (e.g., "Diwali = +30% demand for electronics").
   • Impact: Reduces overstocking costs by 18%.

5. NTC’s Customer Churn Prediction

   • Idea Used: Classification (Logistic Regression)
   • How: NTC analyzes call logs, complaints, and payment history to predict which customers might switch providers.
   • Impact: Targeted discounts retain 20% of at-risk customers.

Exam Tip: How to Score Full Marks

1. Define Clearly:

   • Always start with precise definitions (e.g., "A data warehouse is a subject-oriented, integrated, time-variant, and non-volatile repository...").
   • Example Answer Snippet:

     "OLAP differs from OLTP because it focuses on analytical queries (e.g., 'What was the sales trend in 2023?') rather than transactional operations (e.g., 'Process a new order')."

2. Use Diagrams:

   • Draw mindmaps for classifications (e.g., BI components) and flowcharts for processes (e.g., ETL).
   • Label every part (e.g., "Staging Area → Data Warehouse").

3. Relate to Nepal:

   • Examiners love real-world examples. Always tie concepts to:
     • Banks (Nabil, Global IME).
     • E-commerce (Daraz, Sastodeal).
     • Telecom (NTC, Ncell).
     • Fintech (eSewa, Khalti).
   • Example:

     "Like NTC uses OLAP to analyze call drop rates by district, a similar system could help Daraz identify underperforming warehouses."

4. Compare and Contrast:

   • Use tables for OLTP vs. OLAP, or mindmaps for DSS types.
   • Avoid: Generic lists without comparisons.

5. Worked Examples:

   • Must include:
     • A real scenario (e.g., "Nabil Bank’s loan default prediction").
     • Step-by-step logic (e.g., "ETL → Data Mining → Dashboard").
     • Impact (e.g., "Reduced defaults by 20%").

6. Common Pitfalls to Avoid:

   • ❌ Confusing data mining (discovering patterns) with data warehousing (storing data).
   • ❌ Forgetting ETL in data warehouse explanations.
   • ❌ Describing BI tools without linking them to specific Nepalese companies.

7. Exam Questions to Practice:

   • "Explain the role of a data warehouse in decision-making with an example from a Nepalese bank."
   • "How does OLAP differ from OLTP? Give a real-world analogy using NTC and Daraz."
   • "Describe the steps in data mining using Pathao’s ride-demand prediction as a case study."

Final Visual Summary:

classDiagram
  class BI_System {
    +Data Sources
    +ETL Process
    +Data Warehouse
    +OLAP
    +Data Mining
    +DSS
    +Visualization Tools
  }
  class Data_Warehouse {
    -Subject-Oriented
    -Integrated
    -Time-Variant
    -Non-Volatile
  }
  class OLAP {
    +Multi-Dimensional Analysis
    +Drill-Down
    +Slicing
  }
  class DSS {
    +Model-Driven
    +Data-Driven
    +Knowledge-Driven
  }
  BI_System "1" --> "*" Data_Warehouse : contains
  BI_System "1" --> "*" OLAP : uses
  BI_System "1" --> "*" DSS : includes