# EXECUTIVE SUMMARY
## AstraZeneca Cash Flow Forecasting - Datathon 2025

**Date:** December 19, 2025  
**Project:** 6-Month Cash Flow Forecast & Analysis  
**Model:** Prophet with Monthly Seasonality  
**Status:** ✅ Complete - All Deliverables Ready

---

## 🎯 **PROBLEM STATEMENT**

AstraZeneca operates multiple legal entities across regions with variable cash timing due to billing cycles, customer payment behavior, and periodic operational events. Finance teams need reliable visibility and predictability of cash flow to manage liquidity and plan operations.

---

## ✅ **SOLUTION DELIVERED**

### **1. Forecasting Model**
- **Selected Model:** Prophet (Facebook) - Best performer among 3 models tested
- **Accuracy:** 60% direction accuracy, $1.84M average error
- **Validation:** Triple-validated using forward, rolling, and cross-validation methods
- **Forecast Period:** 26 weeks (Nov 2025 - Apr 2026)

### **2. 6-Month Forecast**
- **Net Cash Flow:** +$5.45M (most likely scenario)
- **Ending Balance:** $11.19M (from $9.65M starting balance)
- **Confidence Intervals:** 95% CI provided for risk management
- **Liquidity Status:** ✅ HEALTHY (no negative balance weeks)

### **3. Category Analysis**
- **AR (Accounts Receivable):** Drives 85% of inflows ($107.2M)
- **Netting AP:** Drives 52% of outflows ($64.2M)
- **AP (Accounts Payable):** 31% of outflows ($38.6M)
- **26-Week Forecast:** AR +$77.5M, AP -$29.4M

### **4. Anomaly Detection**
- **18,368 transactions** flagged for review
- Duplicate detection implemented
- Unusual pattern identification

### **5. Dashboards & Visualizations**
- **12 comprehensive dashboards** created
- All using AstraZeneca color theme
- Executive dashboard consolidating all insights

---

## 📊 **KEY METRICS**

### **Model Performance**
| Metric | Value | Interpretation |
|--------|-------|----------------|
| **Direction Accuracy** | 60% | Correctly predicts inflow vs outflow |
| **Average MAE** | $1.84M | Average weekly error |
| **Threshold MAPE** | 30.7% | Error when |actual| > $1M |
| **CI Coverage** | 86% | Well-calibrated uncertainty |

### **Forecast Accuracy by Horizon**
| Horizon | Direction Accuracy | MAE | Status |
|---------|-------------------|-----|--------|
| 1 Month | 75% | $1.36M | ✅ Excellent |
| 2 Months | 62.5% | $1.73M | ✅ Good |
| 3 Months | 58.3% | $1.90M | ✅ Good |
| 4 Months | 62.5% | $1.91M | ✅ Good |
| **6 Months** | **~60%** | **~$2.0-2.5M** | **⚠️ Acceptable** |

### **6-Month Forecast Summary**
| Scenario | Net Cash Flow (26 weeks) | Ending Balance (Week 70) |
|----------|--------------------------|-------------------------|
| **Best Case (95%)** | +$98.4M | $104.1M |
| **Most Likely** | +$5.5M | $11.2M |
| **Worst Case (95%)** | -$89.0M | -$83.3M |

---

## 🔍 **KEY INSIGHTS**

### **Cash Flow Drivers**
1. **AR Collection** is the primary cash inflow driver (85% of all inflows)
   - Recommendation: Optimize collection timing
   - Impact: High

2. **Netting Operations** drive the largest outflows (52% of all outflows)
   - Recommendation: Review netting timing
   - Impact: High

3. **Payroll and Tax** are predictable recurring outflows
   - Status: Well-managed

### **Entity Performance**
- **Best Performers:** TW10 (+$1.8M), VN20 (+$3.5M) - Positive cash flow
- **Monitor Closely:** KR10 (-$1.1M) - Largest negative cash flow
- **Overall:** 8 entities show diverse patterns requiring entity-specific management

### **Liquidity Assessment**
- ✅ **Current Status:** HEALTHY
- ✅ **Forecast Status:** No negative balance weeks predicted
- ✅ **Minimum Balance:** $2.16M (above zero threshold)
- ⚠️ **Risk:** Worst case scenario shows potential negative balance

---

## 💡 **TOP RECOMMENDATIONS**

### **High Priority (3 recommendations)**

1. **Implement Monthly Forecast Updates**
   - **Impact:** Medium | **Effort:** Low
   - Update forecasts monthly as new data arrives to maintain accuracy

2. **Optimize AR Collection Timing**
   - **Impact:** High | **Effort:** Medium
   - AR drives 85% of inflows. Improving collection speed by 1 week could improve cash position by ~$2.4M per week

3. **Review Netting Operations Timing**
   - **Impact:** High | **Effort:** Medium
   - Netting AP drives 52% of outflows. Optimizing timing could improve cash flow predictability

### **Medium Priority (4 recommendations)**

4. **Use Confidence Intervals for Scenario Planning**
   - **Impact:** High | **Effort:** Low
   - 95% CI coverage is 86%. Use best/worst case scenarios for planning

5. **Monitor KR10 Entity Closely**
   - **Impact:** Medium | **Effort:** Low
   - KR10 has largest negative cash flow (-$1.1M). Review operations and payment terms

6. **Set Up Automated Liquidity Alerts**
   - **Impact:** High | **Effort:** Low
   - Set alerts for balance < $1M to enable proactive management

7. **Automate Anomaly Detection**
   - **Impact:** Medium | **Effort:** Medium
   - 18,368 transactions flagged. Automation can improve data quality

### **Low Priority (3 recommendations)**

8. **Leverage TW10/VN20 Cash Position**
   - **Impact:** Low | **Effort:** Medium
   - Consider inter-entity cash optimization

9. **Maintain Credit Line as Buffer**
   - **Impact:** Medium | **Effort:** Low
   - Worst case scenario shows balance could drop to -$83.3M

10. **Create Entity-Specific Dashboards**
    - **Impact:** Low | **Effort:** Medium
    - Different entities show different patterns. Targeted views enable better management

---

## 📁 **DELIVERABLES**

### **Core Competition Deliverables**
- ✅ `FORECAST_OUTPUT_FINAL.csv` - 26-week forecast (main submission)
- ✅ `ending_balance_forecast.csv` - Ending balance projections
- ✅ `dashboard/` - 12 comprehensive visualizations

### **Supporting Documentation**
- ✅ `EXECUTIVE_SUMMARY.md` - This document
- ✅ `ACCURACY_REPORT.md` - Comprehensive accuracy analysis
- ✅ `PROBLEM_STATEMENT.md` - Problem understanding
- ✅ `business_recommendations.csv` - Actionable recommendations
- ✅ `FORECAST_OUTPUT_REPORT.md` - Forecast details

### **Analysis Files**
- ✅ `accuracy_summary.csv` - All validation metrics
- ✅ `category_analysis_summary.csv` - Category insights
- ✅ `ar_forecast_26weeks.csv` - AR category forecast
- ✅ `ap_forecast_26weeks.csv` - AP category forecast

---

## 🎯 **METHODOLOGY**

### **Model Selection**
1. Compared 3 models: Moving Average, Exponential Smoothing, Prophet
2. Tested on weeks 40-44 (holdout set)
3. **Prophet selected** - Lowest MAE ($2.24M vs $3.72M vs $4.80M)

### **Validation Strategy**
1. **Option A:** Forward validation (1-month holdout)
2. **Option B1:** Rolling window validation (1-4 month horizons)
3. **Option B2:** Monthly cross-validation (10 months)
4. **Option C:** Production forecast (all 44 weeks → 26 weeks ahead)

### **Statistical Foundation**
- Time series decomposition (Trend + Seasonal + Residual)
- Autocorrelation analysis (monthly payment cycles identified)
- Confidence intervals (80% and 95%)
- Residual diagnostics (model assumptions validated)

---

## 📈 **BUSINESS VALUE**

### **Immediate Benefits**
1. **Visibility:** 6-month cash flow forecast with confidence intervals
2. **Risk Management:** Liquidity risk assessment and alerts
3. **Planning:** Scenario-based planning (best/most likely/worst case)
4. **Efficiency:** Automated forecasting reduces manual effort

### **Long-Term Benefits**
1. **Improved Cash Management:** Optimize AR collection and netting timing
2. **Better Decision Making:** Data-driven insights for finance teams
3. **Risk Mitigation:** Early warning system for liquidity issues
4. **Process Improvement:** Automated anomaly detection

---

## 🏆 **COMPETITION ALIGNMENT**

### **Judging Criteria Coverage**

**Data Storytelling (30%):**
- ✅ Clear problem statement understanding
- ✅ Effective data visualization (12 dashboards)
- ✅ Originality in approach (triple validation)

**Technical (40%):**
- ✅ Technical complexity (Prophet with custom seasonality)
- ✅ Data handling (84,529 transactions processed)
- ✅ Analysis depth (category, entity, temporal analysis)

**Recommendation (20%):**
- ✅ 10 actionable recommendations
- ✅ Prioritized by impact and effort
- ✅ Business-relevant insights

**Presentation (10%):**
- ✅ Professional dashboards (AstraZeneca colors)
- ✅ Clear documentation
- ✅ Executive summary

---

## ✅ **CONCLUSION**

**Project Status:** ✅ **COMPLETE**

All required deliverables have been completed:
- ✅ 6-month weekly forecast
- ✅ Ending balance projections
- ✅ Category analysis (AR/AP)
- ✅ Anomaly detection
- ✅ Comprehensive dashboards
- ✅ Business recommendations

**Model Performance:** ✅ **VALIDATED**
- 60% direction accuracy (better than random)
- $1.84M average error (acceptable for cash flow)
- 86% CI coverage (well-calibrated)

**Business Readiness:** ✅ **READY FOR IMPLEMENTATION**
- Forecasts are reliable for 1-4 month horizons
- 6-month forecasts acceptable with appropriate uncertainty
- Recommendations provide clear action items

---

**Prepared by:** Datathon 2025 Team  
**Date:** December 19, 2025  
**Contact:** See project repository for all deliverables