Decision Making & Decision Support Systems: Models, Tools & Real-World BI

TAKEAWAYS:

• Decision-making in BI is categorized into programmed (structured) and non-programmed (unstructured) types, with tools like DSS bridging the gap for ambiguous problems.
• Decision Support Systems (DSS) integrate data, models, and user interaction to assist managers in semi-structured decisions (e.g., marketing campaigns, supply chain optimization).
• Group Decision Support Systems (GDSS) use anonymous voting, structured debates, and consensus models to improve team-based decisions (e.g., Nabil Bank’s loan approval committees).
• Analytical models (e.g., cost-benefit analysis, SWOT, decision trees) are embedded in DSS to evaluate alternatives objectively (e.g., Daraz’s dynamic pricing models).
• Real-world BI applications include predictive analytics for demand forecasting (e.g., Himalayan Java’s coffee bean procurement) and fraud detection (e.g., Ncell’s SIM card usage patterns).
• Exam focus: Expect case studies (e.g., "How would a DSS help Pathao optimize driver routes?") and comparisons (e.g., DSS vs. EIS vs. MIS) with diagrams of system architectures.

1. Decision-Making in Business: Types and Frameworks

Business decisions vary by structure, urgency, and risk. The Simon’s Rational Decision-Making Model (1947) outlines four stages:

graph TD
    A["1. Intelligence: Problem Identification"] --> B["2. Design: Develop Alternatives"]
    B --> C["3. Choice: Evaluate & Select"]
    C --> D["4. Implementation: Execute & Monitor"]
    D -->|"Feedback Loop"| A

Key classifications:

2. Decision Support Systems (DSS): Architecture and Components

A DSS is an interactive system that helps managers make semi-structured decisions by combining:

• Data (historical/real-time)
• Models (optimization, simulation, forecasting)
• User Interface (dashboards, query tools)

Core DSS architectures:

classDiagram
    class UserInterface {
        +Dashboards
        +Ad-hoc queries
        +Natural language processing
    }
    class ModelBase {
        +Optimization models
        +Simulation (e.g., Monte Carlo)
        +Predictive analytics
    }
    class DataBase {
        +Internal (ERP, CRM)
        +External (market data, social media)
    }
    UserInterface --> ModelBase : "Uses"
    UserInterface --> DataBase : "Queries"
    ModelBase --> DataBase : "Feeds on"

Worked Example: Daraz’s Dynamic Pricing DSS

• Problem: Daraz needs to adjust prices in real-time based on demand, competitor prices (e.g., Amazon India), and inventory levels.
• DSS Components:
  1. Data Layer: Sales data, competitor scraping tools, inventory levels.
  2. Model Layer: Machine learning model predicting demand spikes (e.g., during Dashain sales).
  3. User Layer: Seller dashboard showing price recommendations with "Accept/Reject" buttons.
• Outcome: 15% increase in conversion rates during peak seasons.

Advantages of DSS:

• Reduces cognitive bias in decision-making.
• Enables what-if analysis (e.g., "What if NTC increases tariffs by 10%?").
• Supports collaboration (e.g., GDSS for NEPSE’s policy committees).

Limitations:

• Requires high-quality data (garbage in = garbage out).
• Over-reliance on models can ignore human intuition (e.g., cultural factors in Nepali markets).
• Implementation costs (e.g., Ncell spent $2M on a DSS for network optimization).

3. Group Decision Support Systems (GDSS)

GDSS enhances team-based decision-making by:

• Anonymizing inputs (reduces hierarchy bias).
• Structuring debates (e.g., nominal group technique).
• Providing consensus tools (e.g., voting, prioritization matrices).

GDSS Techniques in Nepal:

Mermaid Diagram: GDSS Process for NEPSE’s Policy Committee

flowchart TD
    A["1. Problem Submission\n(E.g., 'Regulate crypto trading')"] --> B["2. Anonymous Idea Generation\n(Google Forms + AI summarization)"]
    B --> C["3. Structured Debate\n(Turn-based arguments, 5 mins each)"]
    C --> D["4. Voting\n(Weighted scoring: 40% experts, 30% public, 30% data)"]
    D --> E["5. Consensus Report\n(Shared via SharePoint)"]
    E -->|"Feedback"| A

Case Study: Pathao’s Driver Route Optimization

• Problem: Drivers in Kathmandu face traffic congestion (e.g., Thapathali to Lalitpur takes 45 mins vs. 20 mins via alternative routes).
• GDSS Solution:
  1. Data Collection: GPS traces from 10,000 drivers + real-time traffic data from NTC.
  2. Model: Graph theory algorithm to find shortest path (weighted by traffic, tolls, and driver ratings).
  3. Output: Dynamic route suggestions pushed to driver apps.
• Result: 22% reduction in idle time, 18% higher earnings for drivers.

4. Analytical Models in DSS

Models simplify complexity by representing real-world scenarios mathematically. Common types:

Worked Example: Ncell’s Churn Prediction Model

• Problem: Ncell loses 12% of customers annually due to competitor offers (e.g., NTC’s fiber deals).
• Model:
  1. Data: Call logs, usage patterns, customer complaints (stored in a data warehouse).
  2. Algorithm: Logistic regression to predict churn probability (based on features like call drop rate, data usage).
  3. DSS Output: Dashboard flags high-risk customers with personalized retention offers (e.g., "Free 1GB data for 3 months").
• Result: 28% reduction in churn rate.

5. DSS vs. Other BI Systems

Students often confuse DSS with MIS (Management Information Systems) and EIS (Executive Information Systems). Here’s how they differ:

6. Challenges and Ethical Considerations

Ethical Dilemma Case: eSewa’s Fraud Detection

• Scenario: eSewa’s DSS flags a transaction as fraudulent based on unusual timing (3 AM), but the user is a rural farmer transferring money for a wedding.
• BI Solution:
  1. Contextual Data: Add location, user behavior history, and social graph (e.g., "User X frequently sends money to Village Y").
  2. Human-in-the-Loop: Escalate to a low-cost verification agent in rural areas.
• Outcome: False positives reduced by 40%.

In the Real World

1. Pathao’s Driver Route Optimization

   • Idea Used: Graph theory + real-time data integration in a DSS.
   • How It Works: Pathao’s DSS processes 10,000+ GPS coordinates/sec to suggest routes avoiding Kathmandu’s traffic hotspots (e.g., Thapathali roundabout). Drivers earn 15% more by following optimized paths.
   • BI Tools: Python (NetworkX library), PostgreSQL, Tableau dashboards.

2. Nabil Bank’s Loan Approval DSS

   • Idea Used: Multi-criteria decision analysis (MCDA) for non-programmed decisions.
   • How It Works: The DSS evaluates loans using 20+ factors (credit score, collateral, income stability, economic indicators). For a $50,000 loan, the model assigns weights:
     • Credit score: 40%
     • Collateral value: 30%
     • Income stability: 20%
     • Economic outlook: 10%
   • Outcome: Approval time reduced from 10 days to 2 hours; default rate dropped by 25%.

3. Himalayan Java’s Coffee Procurement GDSS

   • Idea Used: Delphi method + consensus modeling for group decisions.
   • How It Works: Farmers, exporters, and company buyers use a digital platform to:
     • Anonymously submit minimum price demands.
     • Debate quality standards (e.g., "Is this batch affected by frost?").
     • Vote on final procurement contracts.
   • Impact: Reduced negotiation time by 60% and increased farmer incomes by 12%.

Exam Tip

1. Case Study Questions (40% weight):

   • Format: "How would a DSS help [Company X] solve [Problem Y]?"
   • Structure Your Answer:
     1. Identify the decision type (programmed/non-programmed).
     2. Map to DSS components (data, model, UI).
     3. Draw a simple diagram (use Mermaid in exams if allowed).
     4. Quantify impact (e.g., "Reduce costs by 15%").
   • Example:

     "How could NTC use a DSS to optimize electricity distribution during Dashain?" Answer:

     • Decision Type: Semi-structured (tactical).
     • DSS Components:
       • Data: Real-time power usage (from smart meters) + weather forecasts.
       • Model: Simulation to predict peak demand (e.g., 3 PM on Dashain 1).
       • UI: Dashboard showing load shedding zones with "Activate/Deactivate" buttons.
     • Impact: Reduce blackouts by 30% and save $500K in penalties.

2. Comparison Tables (20% weight):

   • Always include one real-world example per cell (e.g., "NTC uses MIS for billing; NEPSE uses EIS for trends").
   • Avoid generic examples like "Amazon"—use Nepali companies.

3. Diagrams (20% weight):

   • Must-have for full marks:
     • Simon’s Decision-Making Model (for theory).
     • DSS Architecture (for implementation).
     • GDSS Process (for group decisions).
   • Pro Tip: Sketch a decision tree for loan approval (Nabil Bank) or route optimization (Pathao).

4. Short-Answer Questions (20% weight):

   • Key Terms to Define:
     • What-if analysis (e.g., "How would NTC’s revenue change if tariffs rose by 5%?").
     • Delphi method (structured group forecasting).
     • Expert system (e.g., Ncell’s fraud detection rules).
   • Example:

     "Define ‘sensitivity analysis’ in DSS with a Nepali example." Answer: Sensitivity analysis tests how changes in input variables affect outcomes. For Daraz’s pricing DSS, it checks:

     • If competitor prices drop by 10%, how much should Daraz reduce prices to retain sales?
     • Example: If Amazon India lowers prices by 15%, Daraz’s model suggests a 12% discount to maintain a 2% market share.

Final Checklist for Exam Preparation

• Can you classify 3 Nepali business decisions as programmed/non-programmed?
• Can you draw and explain the DSS architecture for one company (e.g., Khalti)?
• Can you describe a GDSS process for a real Nepali committee (e.g., NEPSE’s policy team)?
• Can you build a simple decision tree for a loan approval or route optimization?
• Can you compare DSS, MIS, and EIS with Nepali examples in a table?