Session 26: Capstone Project Presentations

1. Capstone Project Overview & Requirements

The capstone project is the culminating assessment of the Insurtech & Digital Risk Solutions course. It requires you to integrate the tools, techniques, and domain knowledge from all 25 preceding sessions into a comprehensive insurance analytics solution. The project is not just a technical exercise — it is a demonstration of your ability to think like an insurance analytics professional: identify a business problem, apply the right analytical approach, generate actionable insights, and communicate them effectively.

1.1 Project Options — Choose One

1.2 Deliverables

Every project, regardless of option, requires three deliverables:

2. Project Option 1 — Insurance Claims Dashboard

Summary: Build an interactive analytics dashboard that provides comprehensive visibility into claims patterns, settlement efficiency, fraud indicators, and operational performance. The dashboard should enable a claims manager to: identify trends, spot anomalies, track KPIs, and drill down into specific claim segments.

Technical requirements:

• Use the cleaned insurance dataset (insurance_cleaned.csv from Session 05)
• Dashboard tool: Power BI Desktop or Looker Studio
• Minimum: 3 pages, 15+ visuals, 5+ slicers/filters
• Must include: KPI cards with conditional formatting, trend charts with annotations, category comparisons, a detail table with drill-through, and an exception/alert section
• DAX measures: minimum 8 custom measures including time-intelligence (MoM, YoY comparisons)
• Design: follow the principles from Session 23 — "one page, one question," executive summary first, annotations on trends, consistent colour coding

Evaluation emphasis: Data preparation (20%), Visualisation design (25%), KPI selection and analysis (25%), Business interpretation (20%), Presentation professionalism (10%).

3. Project Option 2 — Fraud Detection System

Summary: Build an ML-based fraud detection system that identifies suspicious claims before payment. The system should combine supervised and unsupervised techniques, provide explainable predictions, and include an operational workflow for investigators.

Technical requirements:

• Use the insurance_cleaned.csv dataset with fraud_flag as the target
• Tools: Python (Scikit-learn, XGBoost, Imbalanced-learn) + KNIME workflow (replicating the Python pipeline)
• Minimum 10 engineered fraud indicator features (use Session 15-16 frameworks)
• Train: Random Forest + XGBoost + Isolation Forest ensemble. Compare performance using PR-AUC (not ROC-AUC)
• Optimise decision threshold using cost-based analysis (fp_cost = ₹500, fn_cost = ₹75,000)
• SHAP explainability for at least 3 individual predictions
• KNIME workflow that replicates the scoring pipeline (from data input to scored output)
• Concusion matrix, decile lift chart, and precision-recall curves

Evaluation emphasis: Problem definition (15%), Technical implementation (30%), Insights and recommendations (25%), Visualisation (15%), Innovation and ethics (15%).

4. Project Options 3–8 — Summarised

Option 3: Insurance Chatbot

Scope: Design and test a GenAI-powered customer service chatbot for an insurance company. Write a system prompt with clear scope boundaries, escalation rules, and disclosure requirements. Test with 5 customer scenarios. Evaluate the bot's responses for accuracy, safety, and tone. Provide a critical assessment of where the bot works well and where it fails. Include: the system prompt, the 5 test scenarios with outputs, a limitations analysis, and a deployment recommendation (what use cases to launch, what to hold back).

Tools: ChatGPT, Claude, or any LLM platform. Python optional (for RAG implementation).

Option 4: Customer Risk Scoring Model

Scope: Build a risk scoring model (0–1000) for motor OR health insurance underwriting. Engineer at least 15 features including ratios and interactions (Session 12). Train a Random Forest model, perform hyperparameter tuning, and report feature importance. Map risk scores to premium recommendations (discount/standard/loading/decline). Evaluate calibration (does risk score 600 mean 60% claim probability?). Provide SHAP explanations for 3 individual predictions. Design a phased rollout approach (shadow → assisted → full STP).

Tools: Python (Scikit-learn, Pandas, SHAP).

Option 5: Claims Analytics Using Python

Scope: End-to-end claims data analysis from cleaning to visualisation to predictive modelling. Clean and merge Customer-Policy-Claims data (Session 05). Create 6+ publication-quality visualisations (Session 06 — trend, distribution, correlation, comparison, heatmap, pairplot). Build a simple severity prediction model (linear regression + XGBoost). Perform a claims seasonality analysis using time series decomposition (Session 21). Write a 500-word business report synthesising all findings.

Tools: Python (Pandas, Matplotlib, Seaborn, Scikit-learn, statsmodels).

Option 6: Parametric Insurance Product Design

Scope: Design a parametric insurance product for a climate risk of your choice (rainfall-index for drought, wind speed for cyclone, temperature for heatwave). Use Session 19's framework: define the index, trigger, exit, and payout function. Simulate 20+ years of historical payouts. Calculate pure premium and commercial premium (with 35% loading). Perform sensitivity analysis on the trigger level (50%–80%). Analyse basis risk with ±15% spatial variability. Write a product launch memo covering: target customer, distribution strategy, pricing, basis risk disclosure, and key risks.

Tools: Python (Pandas, NumPy, Matplotlib) + Excel for the risk matrix. Weather data from weather_data.csv.

Option 7: Customer Churn Prediction

Scope: Build a churn prediction model using policy, claims, and customer data (Session 22). Engineer at least 12 churn features. Train XGBoost with scale_pos_weight. Find the optimal decision threshold using cost-based analysis (retention offer cost = ₹200, CLV estimated from premium and retention data). Design a segmented retention strategy with at least 3 different interventions. Calculate the ROI of the retention programme and the break-even save rate. Build a KNIME workflow for weekly automated churn scoring.

Tools: Python (Scikit-learn, XGBoost) + KNIME.

Option 8: Digital Risk Assessment Framework

Scope: Build a comprehensive digital risk assessment framework for a mid-size Indian insurer. Include: (a) a cyber risk matrix (Session 17 — 8 threat types × 5 likelihood/impact levels), (b) a compliance monitoring dashboard (Session 24 — solvency, expense ratio, settlement ratio, rural business, grievance redressal), (c) an AI governance framework (Session 25 — governance structure, risk classification, development standards, monitoring), and (d) a DPDP Act compliance checklist. Integrate all components into a coherent "Digital Risk Report" that the CRO could present to the board.

Tools: Power BI (compliance dashboard), Excel (risk matrix), ChatGPT (governance framework draft).

5. Report & Presentation Guidelines

5.1 Report Structure

The project report (maximum 10 pages excluding appendices) must follow this structure:

5.2 Presentation Format (10+5 Minutes)

Each presentation is strictly 10 minutes followed by 5 minutes of Q&A. The 10-minute presentation should be structured as follows:

PRESENTATION STRUCTURE (10 minutes — 8 slides maximum)

Slide 1: Problem Statement (60 seconds)
  • What problem are you solving? Who cares? Why does it matter?
  • One sentence: "Premium growth is strong but the combined ratio is deteriorating
    because motor TP claims are growing faster than premium adjustments."

Slide 2: Data & Approach (60 seconds)
  • What data did you use? What was your analytical approach?
  • Not a data dictionary slide — a "how I tackled this" slide.
  • "I analysed 10,000 policies with 15 engineered features using XGBoost..."

Slide 3: Key Finding #1 — The Main Insight (90 seconds)
  • The most important finding — lead with this, not with technical details.
  • Chart/graph + one-sentence interpretation + business impact.

Slide 4: Key Finding #2 (90 seconds)
  • The second most important finding. Same format: data → insight → impact.

Slide 5: LIVE DEMO (3 minutes — non-negotiable)
  • Show your solution working. Dashboard with slicers. Code running. KNIME workflow executing.
  • Practice this — demo failures are the #1 cause of marks deduction.

Slide 6: Recommendations (60 seconds)
  • What should the business DO? 3 recommendations maximum. Each with owner + timeline + impact.

Slide 7: Limitations & Next Steps (30 seconds)
  • What are the caveats? What would you do next? Honest about limitations = marks bonus.

Slide 8: Q&A Prep (backup — only if needed)
  • Anticipate questions and have a backup slide.
  • Common questions: "How did you handle missing data?" "What was the biggest assumption?"
    "How would you validate this in production?"

Q&A Period (5 minutes)
  • Listen to the full question before answering.
  • If you don't know, say "I don't know" and explain what you would do to find out.
  • Faculty will ask: "What was the most difficult part of the analysis?"
    "If you had one more month, what would you add?" "How would your approach change
    if the dataset were 10× larger?"

6. Presentation Day — Logistics and Evaluation

6.1 Timetable

SESSION 26 — TYPICAL TIMETABLE (90 minutes for ~6 presentations)

┌─────────────────────────────────────────────────────────────────────┐
│ 09:00  09:05  │ Setup and briefing — Presentation order,           │
│                │ timekeeping rules, evaluation criteria review      │
│ 09:05  09:20  │ Presentation 1 (10+5)                              │
│ 09:20  09:35  │ Presentation 2 (10+5)                              │
│ 09:35  09:50  │ Presentation 3 (10+5)                              │
│ 09:50  10:00  │ Break / Transition                                 │
│ 10:00  10:15  │ Presentation 4 (10+5)                              │
│ 10:15  10:30  │ Presentation 5 (10+5)                              │
│ 10:30  10:45  │ Presentation 6 (10+5)                              │
│ 10:45  10:55  │ Faculty synthesis — Key themes across all projects │
│ 10:55  11:00  │ Course wrap-up, career pathways, final Q&A         │
└─────────────────────────────────────────────────────────────────────┘

6.2 Evaluation Rubric

7. Course Wrap-Up & Career Pathways

You have completed 26 sessions covering the full spectrum of insurance analytics — from insurance fundamentals and risk management (Sessions 1–4) through data analytics and visualisation (Sessions 5–7), InsurTech innovation (Sessions 8–10), artificial intelligence (Sessions 11–14), fraud analytics and cyber risk (Sessions 15–17), climate risk and parametric insurance (Sessions 18–20), advanced analytics and BI (Sessions 21–23), and governance, compliance, and ethics (Sessions 24–25). The capstone project (Session 26) is the culmination — a demonstration of your ability to integrate these skills into a real-world insurance analytics solution.

7.1 What You Have Learned — The Three Pillars

CILO-1 — Insurance Domain Knowledge: You can now analyse the insurance ecosystem, risk management frameworks, InsurTech business models, and the regulatory landscape with confidence. You understand: how insurers make money (the combined ratio, float, underwriting vs. investment income), why risk pooling works (the law of large numbers), and what makes insurance fundamentally different from other financial services (it is a social good, not just a product).

CILO-2 — Analytics Tools: You have hands-on proficiency with six industry-standard tools — Python (Pandas, NumPy, Matplotlib, Seaborn, Scikit-learn, XGBoost, statsmodels), Excel (P&L models, risk matrices, premium calculations), Power BI (DAX measures, 3-page dashboards, drill-through, KPI cards), KNIME (visual fraud and churn workflows), Looker Studio (executive dashboards, Google Sheets integration, scheduled PDF snapshots), and ChatGPT (prompt engineering, policy summarisation, insurance research, governance framework drafting).

CILO-3 — Solution Design: You can design and evaluate data-driven insurance solutions — from risk scoring models and fraud detection systems to parametric insurance products and compliance monitoring dashboards. You can bridge the gap between technical analytics and business decision-making. This is the most valuable skill in the insurance analytics job market.

7.2 Career Pathways

7.3 Continuing Your Learning

• Certifications: AINS (Associate in General Insurance), CPCU (Chartered Property Casualty Underwriter), CFA (Chartered Financial Analyst — relevant for insurance investment), FRM (Financial Risk Manager), and insurance data science certifications from IITs and IIRM.
• Industry bodies: Insurance Institute of India (III), Institute of Actuaries of India (IAI), NASSCOM Centre of Excellence for InsurTech, and the Insurance Information Bureau (IIB) for industry data access.
• Reading: IRDAI Annual Report (published annually — the single best source for industry data and regulatory direction), NASSCOM InsurTech reports, Swiss Re sigma reports (global insurance trends), and the Insurance Journal and Asia Insurance Review for industry news.
• Networking: InsurTech conferences (InsurTech India Summit, Global InsurTech Summit), LinkedIn insurance analytics groups, and the Woxsen School of Business alumni network in insurance and financial services.