Predictive
Analytics in Action
San Diego County
• 4,083 sq. miles – 8th largest county by size • 3 million residents – 5th largest by population • 725,000 children
• 13.9% living below poverty level
• Unemployment rate 5.9% (as of 10/14)
Our mission is to enhance our children’s futures by obtaining support for families today
• 73,000 cases
• Over 113,000 children served
• FFY 2013 - $178.7M
• 6% of California’s total caseload
• FPM Performance FFY14 : #1 – 99.8%, #2 – 88.2%,
#3 – 70%, #4 – 69.8%, #5 - $3.85 (FFY2013)
County of San Diego
•
Precision
•Consistency
•Agility
•Speed
•Cost
Predictive Analytics
Child Support Payment Predictor
Scope of Study:
1,033 cases opened 10/2011 through 5/2013
No prior support order
Objective:
To establish a “payer” classification predictive tool that
prognosticates the NCPs who are potentially at risk to fall behind on their child support payments.
Provide staff with a tool to more effectively perform
Child Support Payment Predictor
Data Elements 466 data elements from the case and participants 68 field variables were created as candidates for
exploratory data analysis
29 variables were selected for model building 8 variables remain in the final model
Child Support Payment Predictor
Model data points
NCP’s Employment stability
Incarceration history
Unemployment in previous year
Living status in previous year Default history on first order NCP’s email information
NCP’s employment status at initial order establishment
Child Support Payment Predictor
Payment Predictor Details
Multinomial logistics regression
used
4 payer groups created:
Extremely rare payer (0%-30%)
Rare payer (31% to 50%)
Attempt payer (51 to 80%)
Constant payer (81% or greater)
Distribution of 6 month average payment rate (N=1033)
Child Support Payment Predictor
By payment class
Extremely rare payor (0%-30%)
Rare payor (31% to 50%)
Attempt payor (51 to 80%)
Constant payor (81% or greater)
Total due vs. Total payment for each payor class
Child Support Payment Predictor
Time to order or “waiting time” Shortest - 2months Longest- 19 months
Child Support Payment Predictor
Variable – NCP e-mail address 54% had e-mail addresses Strong impact on payer class
Child Support Payment Predictor
Results: Prediction accuracy of the model 62.5%
Percentage of the caseload more accurately monitored 14%
Able to target/focus on at risk cases before they
What’s Next or in the works?
Business Intelligence – “Smarter Planning” New analytics tools
Greater flexibility and detailed analysis
Review “payment predictor” for areas of further
analysis
Legal Paperless System – customer behavior Performance-based dashboard
Overview
What is Business Intelligence?
Set of theories, methodologies, processes and technologies that
transform raw data into meaningful and useful information.
Why do we need Business Intelligence?
Reaching limit of producing NEW reporting and analytical results. Paradigm shift from reactive reporting (the past) to proactive
reporting (looking into the future).
Who is likely to participate in the process? Who is likely to pay their child support? What will our caseload look like in 5 years?
Smart
er
Planning – A new way to plan
Smarter Strategic Planning
Strategic Planning Tool – predict future case loads, staffing needs, budget and
performance.
Smarter Case/Knowledge Worker
Case Stratification Tool - Based on certain case data predictors, will be able to predict case outcomes with high probability.
“Reduce” Case Loads: Focus on cases that need ‘personal touch’ – automate others.
Change Individual Case and Family Outcomes Proactive Case Management
Targeted Early Intervention
Smart
er
Planning – How do we get there?
4 Phases
Phase I: DCSS Executive Dashboard
Phase II: DCSS Smarter Planning
Phase III: County-wide Data Sharing
Smart
er
Planning – Phase I
Phase I: DCSS Executive Dashboard The Past and Present
Data Visualization of Current Performance Measures using Key Performance Indicators (KPI)*
Summary of Case Load Performance General Health of Department
Central Reporting Portal
*A set of quantifiable measures that a company or industry uses to gauge or compare performance in terms of meeting their strategic and operational goals.
Smart
er
Planning – Phase II
Phase II: DCSS Data Mining and Predictive Analytics Looking at the Future
Predict future case performance, participant and staff behavior
Correlate Staff Demographics to Case Data Smarter Planning Tool
Case Stratification Tool Alerts
Smart
er
Planning – Phase III
Phase III: County-wide Data Sharing Current limits to Business Intelligence based on only Child Support Data (silo effect)
Partner with other Departments including IV-A to Pilot Richer Business Intelligence and Predictive Analytics
Smart
er
Planning – Phase IV
Phase IV: SmarterGoverning
Executive Dashboard, KPIs and Reporting Portal Rollup for PSG
Departments
Smarter, Integrated Knowledge Workers
Probation Officer helping with Child Support Case
Management
Early Case Intervention
Reduce recurring public sector demands
Central Case Management Customer Service Portal Predicting Future Budget, Facilities and Staffing