Predicting Shrink and
Allocating Resources
Strategically
Office Depot, Inc.
Aashish
Amin
Topics
•
Predictive Modeling
• What, why, and how?•
Identifying Attribute Impacts and Importance
•
Implementation of Change
• Opportunity Stores • Target Stores • Threshold Stores • Store Visits•
Impact
Predictive Modeling
•
What is it?
• Statistical process using patterns as well as current and historical data to predict future results
•
Why use it?
• More science by incorporating more data
•
How to use it?
• Identify the proper software to handle the data and analytics • Applied Predictive Technologies (APT)
How did we use Predictive Modeling
Find more efficient methods to reduce shrink in stores with the
most opportunity for improvement
Shrink Reduction Phases
•
Phase 1: Build an accurate model to predict shrink
•
Phase 2: What can we learn from good performing stores to help
understand under-performing stores?
Example
• How do you tell stores apart?
• Historical Shrink for 2 Random Retail Locations
• Should these stores be treated differently?
• How do I treat them differently?
• And still drive shrink reductions?
2012 2011
Store A ($85,000) ($83,000)
Attributes to Model
Store level attributes are Controllable or Uncontrollable
Controllable
Controllable
Uncontrollable
Uncontrollable
Shrink
Audit Scores
Population
Distance to Highway
Shrink Reduction Phase 1
1. Determine attributes that are highly correlated to inventory loss • Identified over 800 attributes per location
• Loss Prevention metrics
• Real Estate demographic data • Human Resources metrics • Store Traits
• Operational metrics
• Ideal model will have 10-30 attributes based on statistical impact.
1. Combine key attributes into regression model to predict inventory loss by location
Level of Accuracy with Historical Data
•
Model predictions are highly accurate across all quartiles
•
Model Results:
• Model correctly predicts year-over-year shrink change by location 90% accurately
• Looking at actual shrink % predictions, model was 86% accurate (+/- 0.10% shrink change)
Opportunities for Success
•
How many locations underperformed versus their predicted
inventory loss %?
• Less than half the stores
•
If these locations achieved their predicted results, how would the
Company’s overall inventory loss results change?
• Impact could be worth Millions
Example
•
Are the stores still the same?
•
Should they be treated differently?
•
Why are they different?
2013 Savings
Predicted 2012 2011 Opportunity
Store A ($90,000) ($85,000) ($83,000) ($5,000)
Shrink Reduction Phase 2
What can we learn from good performing stores to help
understand under performing stores?
Winner’s Profile
1
Determine What to Analyze Annual shrink % for all stores2
Identify Uncontrollable Correlates of Performance Which attributes outside of a store’s control (i.e. trade area demographics) affect performance?3
Model Uncontrollable Attributes Quantify the impact of different uncontrollable characteristics4
Apply the Model to the Entire NetworkIdentify Uncontrollable Attributes
•
•
Higher Cap Index (Aggravated Assault) in locations with worse
than predicted inventory loss results
• Proximity to larger households increased with stores that
performed worse than predicted
Attribute Impact on Actual Shrink Versus Prediction
•
Historical shrink % is highly indicative of future shrink results
•
Audit scores are great indicators of operational compliance
•
Overall operational accuracy is higher in stores with better performance•
Less-experienced managers have lower performing stores
•
Experience in the position•
Tenure with the companyFocus on Underperforming Stores
• Increase Audit presence where it’s most needed
• Partner with Regional Vice Presidents and District Managers to improve
performance and compliance
• Revised Audits for more focused reviews
• Training
• Site visits notes
• Recap each visitLeveraging Data – Opportunity Store Program
•
Identify stores based on their predictive modeling shrink
savings potential.
•
Objective: Respond to potential shrink trends and proactively
impact shrink.
Recognizing Shrink – Target Store Program
•
Identify stores by their actual inventory results and shrink.
•
Opportunity and Target Stores share enhanced Loss
Prevention services.
Opportunity Stores
:
identified based on leading indicators
(predictive modeling of shrink exposure)
Target Stores
:
identified based on lagging indicators
Field Staff Adjustments
•
Store Visit Schedules
•
Program stores have increased frequency•
Action plan review and follow-up•
Remote Management Skills
•
Change in service protocol for stores•
Emphasis on communication•
Investigations
•
Data mining•
Telephone InterviewsOpportunity / Target Store Visits
•
Monthly Loss Prevention Audit
•
Research Primary Focus Areas
• Review shrink performance and adjustment accuracy
• Follow-up on prior visit / audit exceptions
•
Training
• Document store staff training each visit
•
Communication
• Review trending with store staff
Program Impact
• Quantifying Results
• Weekly status reports: Red / Yellow / Green progress indicators
• Program membership review cycles
• Accountability for performance
• Training and Staff Development
• Increased presence and training opportunities in program stores
• Focus provided for operational skills to impact shrink
• Catalyst for Loss Prevention Manager skill development
• Increasing Scope of Influence
Review of Objectives
• Leverage Predictive Modeling Data
• Recognize areas of potential shrink
• Design programs to focus assets for best impact
• Identify Areas of Historic Exposure
• Supplement Predictive Modeling with market experience
• Maximize Impact of Assets
• Increase frequency of visits to emphasize training & behavioral change
• Reduce frequency of Loss Prevention visits in top performing locations