Geographical Perspectives on Business Analytics with Applications in Construction Supply and Automotive Colorado State University - Pueblo Justin Holman, PhD Copyright 2014 TerraSeer, Inc. All rights reserved.
Overview Background/Experience/Perspective Branch Location Problem Retail Inventory Assortment Problem Consumer Segmentation Geographical Visualization
Overview Background/Experience/Perspective Branch Location Problem Retail Inventory Assortment Problem Consumer Segmentation Geographical Visualization
Background Education Claremont McKenna College B.A., Philosophy and Mathematics, 1990 University of Oregon Geography, GIS, Spatial Statistics, Cartographic Visualization M.S., 1996, Ph.D., 2004 Northwestern University - Kellogg School of Management Certificate, Designing and Managing the Supply Chain, 1998
Background Professional Experience Dynamix Inc. 3D Simulation Software Development LogicTools, Inc. Supply Chain, Network Optimization, Map UI (Acquired by IBM) US Geological Survey Data Visualization, Spatial Statistics, Environmental Modeling MapInfo Retail Location Research, Applied Statistical Modeling TerraSeer (dba Aftermarket Analytics) Location Analytics, SaaS Development Automotive Aftermarket, Construction Supply Industry
Background: Select Clients
Overview Background/Experience/Perspective Branch Location Problem Retail Inventory Assortment Problem Consumer Segmentation Geographical Visualization
Construction Materials Supply Chain
Raw Materials Manufacturing Assembly Engineering Distribution Construction Materials Supply Chain
Types of Location Problems Raw materials Production/Plant location Assembly/Kitting Distribution Centers (DCs) Cross docks Branch Retail Location Showroom Construction Site Construction Materials Supply Chain
Types of Location Problems Raw materials Production/Plant location Assembly/Kitting Distribution Centers (DCs) Cross docks Branch Retail Location Showroom Construction Site Min Cost Max Revenue Construction Materials Supply Chain
Types of Location Problems Raw materials Production/Plant location Assembly/Kitting Distribution Centers (DCs) Cross docks Branch (Counter + Delivery) Retail Location Showroom Construction Site Min Cost Max Revenue Construction Materials Supply Chain
Overview Background/Experience/Perspective The Branch Location Problem 2 Competing Objectives - Maximize Revenue - Minimize Delivery Costs Retail Inventory Assortment Problem Consumer Segmentation Geographical Visualization
Sugar Land, TX Example
Sugar Land Model Trade Area
Disaggregate Trade Area Forecasting Counter$20,009 Delivery$207,653 Total$227,662 Model predicts anticipated counter sales and delivery sales originating within each trade area ZIP Code
Branch Sales Forecasting Sugar Land$18,500,000 Model calculates a branch sales forecast by summing ZIP forecasts and making adjustments (see below) Adjustments: Beyond Sales (proportion of sales projected to come from beyond the trade area), Branch Size (model assumed an average branch size of 16,000 gsf), Region/DMA (White Cap achieves stronger performance in some markets than others), Contractor Density (markets with exceptionally high contractor counts achieve stronger sales)
Houston Market Results suggest that 3 additional $10M+ branches can be supported in the Houston market
WINNERS LOSERS
Houston Market Results suggest that 3 additional $10M+ branches can be supported in the Houston market
Overview Background/Experience/Perspective The Branch Location Problem 2 Competing Objectives - Maximize Revenue - Minimize Delivery Costs The Retail Inventory Assortment Problem Consumer Segmentation Geographical Visualization
Miami Baseline Demand Met: $29.9 M Delivery Cost: $1.6 M Average Distance: 22 mi
Miami Optimized Demand Met: $32.5 M Delivery Cost: $1.5 M Average Distance: 20 mi
Miami Optimized – Add 1 Demand Met: $32.5 M Delivery Cost: $0.5 M Average Distance: 7 mi
Overview Background/Experience/Perspective Branch Location Problem Retail Inventory Assortment Problem Consumer Segmentation Geographical Visualization
Inventory Assortment Optimization Process
1. Calculate Repair Rates Repair Rate Modeling Process
2. Build Statistical Models Ball Joint Repair Rates by Vehicle Age and Type Repair Rate Modeling Process
3. Create Adjustment Factors Repair Rate Modeling Process - Make, Model, and Regional Adjustments - Analyze Residuals and Calibrate Model 1.Alabama 2.California 3.Georgia 4.Washington 5.Oregon 1.New Hampshire 2.Vermont 3.Maine 4.Massachusetts 5.Rhode Island Lower than ForecastHigher than Forecast Sample Adjustments for Ball Joints
4. Validate Predictive Accuracy Repair Rate Modeling Process Initial model: R 2 = Adjusted Model R 2 = 0.978
Total Demand Forecast Zip Code VIO x Repair Rates
Total Demand > Sales Forecast Create Store Trade Areas (ZIP codes) Aftermarket Adjustment Channel Market Share Part Attributes (good, better, best) Sales Forecast TOTAL DEMAND SKU Level Demand by ZIP
Trade Area Sales Forecast ZIP Code Total Demand x Market Share = ZIP Forecast Store Location Sum of ZIP Forecasts = Store Sales Forecast Key Factors: Market Share, Aftermarket Adjustment, Proximity to Store Location, Competition, Part Attributes
Sales Forecast > Inventory Recommendation Estimated Forecast Error Target Service Level Replenishment Lead Time Order Frequency Holding Cost Optimization Engine Inventory Recommendation SALES FORECAST SKU Level Sales Estimates by Store
Optimize Inventories For Efficiency
Optimal Working Capital Utilization
Recommendations via Web Portal
Select A Store
Recommendations via Web Portal Click To See Inventory
Recommendations via Web Portal
Search Function
Recommendations via Web Portal Multiple Sorts
Recommendations via Web Portal Save To Excel
Recommendations via Web Portal Go Back To Pick Another Store
Inventory Optimization Process
Overview Background/Experience/Perspective Branch Location Problem Retail Inventory Assortment Problem Consumer Segmentation Why Spatial is Special
2013 Q2 Starter Demand By County U.S. Total Units: 2.85M
2013 Q2 Starter Demand By County U.S. Total Units: 2.85M Counties or Aggregate to Fewer Units?
Most use State Borders
But why not use Watersheds
Or Topography
Or Solar Radiation
Or Precipitation
Or Earthquake Zones
Or Population Density
Or Racial Density
Or Sports Affiliation
Or Language Preferences: Pop vs Soda vs Coke
But Since Government Data is Typically Provided by State, Most use State Political Borders
Melanoma Risk
Or Solar Radiation
3. Create Adjustment Factors Repair Rate Modeling Process - Make, Model, and Regional Adjustments - Analyze Residuals and Calibrate Model 1.Alabama 2.California 3.Georgia 4.Washington 5.Oregon 1.New Hampshire 2.Vermont 3.Maine 4.Massachusetts 5.Rhode Island Lower than ForecastHigher than Forecast Sample Adjustments for Ball Joints
Overview Background/Experience/Perspective Branch Location Problem Retail Inventory Assortment Problem Consumer Segmentation Geographical Visualization
Power of Geographical Visualization Maps vs Spreadsheets Pattern Detection Collaboration
Power of Geographical Visualization Pattern Detection
Power of Geographic Visualization Pattern Detection
Would you discover this problem with a spreadsheet?
Power of Geographic Visualization Collaboration
Power of Geographic Visualization Collaboration Wait….wouldnt it be better to plan this war with spreadsheets?
Copyright 2014 TerraSeer, Inc. All rights reserved. blog: justinholman.com
Independent Data: Vehicle Registration Repair Survey Channel Data Store Locations and Attributes Current Inventory By SKU/Store/DC Sales History By SKU/Store/DC Delivery and Service Level Requirements Nice to Have Customer (end-user) Locations Estimated Market Share Demand Forecast Competition Inventory Assortment Data Inputs
1. Model Repair Rates 2. Generate Demand Forecasts 3. Trade Area Sales Forecasts 4. Optimize Inventory 5. Develop Communication Portal 6. Maintain/Refine Models 7. Maintain/Refine/Customize Portal 8. Start Over, Improve, Rinse and Repeat Continuous Improvement Iterative Analytical Approach
Sources of Error in Spatial Analysis Geocoding
Sources of Error in Spatial Analysis Distance Measurement Method of measurement: Straight line distance vs Great Circle distance Scale of Measurement:
Sources of Error in Spatial Analysis Spatial Autocorrelation
Sources of Error in Spatial Analysis Spatial Autocorrelation