Assembling the Crystal Ball: Using Demand Signal Repository to - PowerPoint PPT Presentation

Assembling the Crystal Ball: Using Demand Signal Repository to Forecast Demand Authors: Ahmed Rashad & Santiago Spraggon Advisor: Shardul Phadnis Sponsor: Niagara Bottling LLC. MIT SCM ResearchFest May 22-23, 2013

Agenda • Overview • Methodology • Conclusion May 22-23, 2013 MIT SCM ResearchFest 2

Agenda • Overview • Methodology • Conclusion May 22-23, 2013 MIT SCM ResearchFest 3

What is Demand Signal Repository (DSR)? • Demand forecasting technique • Using external Signals • Aggregated in a single Repository External Signals Repository (Database) May 22-23, 2013 MIT SCM ResearchFest 4

When to use DSR? 1. What are we forecasting? 2. What data is available? 3. What stage in the product lifecycle? 4. Is the investment worth it? May 22-23, 2013 MIT SCM ResearchFest 5

When to use DSR? • Depends on what are we forecasting Seasonality Demand Trend Unexplained Base Demand Time Special Events + Trends and Special Events Trends and Patterns Patterns Time Qualitative DSR Series May 22-23, 2013 MIT SCM ResearchFest 6

When to use DSR? • Depends what data is available Little or No Sufficient History Sufficient History History + External Data Time Qualitative DSR Series May 22-23, 2013 MIT SCM ResearchFest 7

When to use DSR? • Depends on stage in the product lifecycle Decline Introduction Growth Maturity Qualitative Qualitative Time-Series Sales Causal Causal Time May 22-23, 2013 MIT SCM ResearchFest 8

When to use DSR? • Depends on the investment Total System Cost Cost of Cost of Inaccuracy Forecasting Costs Target Area Forecast Accuracy May 22-23, 2013 MIT SCM ResearchFest 9

How can we develop a Demand Signal Repository (DSR) to better predict demand? May 22-23, 2013 MIT SCM ResearchFest 10

Agenda • Overview • Methodology • Conclusion May 22-23, 2013 MIT SCM ResearchFest 11

Method Used • Planning • Literature Review Initiation • Interviews • Requirements Data • Collection • Validation Management • Initial Models Modeling • Analysis • New Models May 22-23, 2013 MIT SCM ResearchFest 12

Modeling Dependent Variables Independent Variables Liters per Customer, in a State, per day or week Product Customer Geography Time Growth Seasonality All Niagara Annual All cases Wholesale Price All Niagara Merchandizing Region Quarterly All liters Retail Price Natural Disasters Top 12 State Monthly Weekly Cycles Category City Weekly Buying Patterns Temperature Top 3 3-Digit Zip SKU Daily code Food Stamps • 240+ Models • 60%+ Customer – State - Daily • 85%+ Customer – State - Weekly May 22-23, 2013 MIT SCM ResearchFest 13

Agenda • Overview • Methodology • Conclusion May 22-23, 2013 MIT SCM ResearchFest 14

Key Findings • Most Significant : • Ordering patterns & POS quantity • Seasonality • POS revenue (proxy for price) • Least Significant : • Temperature • POS quantity and revenue May 22-23, 2013 MIT SCM ResearchFest 15

Challenges and Caveats • Accuracy vs. Practicality • Recording Data • Retailer Policies • How much Technology? May 22-23, 2013 MIT SCM ResearchFest 16

Conclusion • DSR could Significantly increase forecast accuracy (60%-85%) • Accurate models are good, Simple models are better (>5 Factors) • Perceptions can be misleading (Temperature) May 22-23, 2013 MIT SCM ResearchFest 17

Assembling the Crystal Ball: Using Demand Signal Repository to Forecast Demand Authors: Ahmed Rashad & Santiago Spraggon Advisor: Shardul Phadnis Sponsor: Niagara Bottling LLC. MIT SCM ResearchFest May 22-23, 2013