Friday, September 20, 2013

2013 Predictive Analytics Conference Friday Afternoon

Predictive Analytics in Medicare-Kelly Gent

  • Models in credit card fraud
    • Rule-in FLA with charge in CA
    • Anomaly, 3 tvs in one day
    • predictive model-charges for multiple tvs out of state after a one dollar charge on wednesday
    • social network
    • charges at address known to be used by bad actor
  • Traditional analytics approach
    • run a model, use top tier and run an investigatio
  • Currently running all models mentioned.  Building a good data set.
  • This group is required to publish a fraud prevention report to congress.
  • They stopped prevented or identified 115 million in improper payments which is a 3 to 1 savings.
  • 536 leads for new investigations
  • New info for 511 existing investigations
  • Models are working
  • What is the command center?
    • Center for detection and investigation drivng integrity and innovation
    • Paradigm shift
    • Introduces mission
    • Speeds up actions
  • Old way
    • Have a lead
    • do an investigation
    • Take some action, i.e. remove provider or overpayment
    • Savings
    • LOTS OF people involved
  • New approach
    • Identify better leads faster.
    • Introduced command center to bring all people together in room, turn off blackberries, and solve problems.
  • What's next for the FRAUD PREVENTION SYSTEM
    • Evaluating feasability of expanding analytics in medicaie
      • There are 56 medicaid programs which is a problem.  
    • Activities to analyze feasability
      • focus groups with state medicaid agencies
      • evaluate outcomes of introduing post-payment medicaid data into FPS (e.g. if fraudulent in medicare, will aso be for medicaid
      • Also providing technical assistance
    • Prevention-partnership is designed to share info and best practices to improve det5ection and prevention.  
    • Lots o partners 
    • 11 partners contributed to first information exchange.                                                          
Industry Expert Panel

Daniel porter-pinpointing the persuadables, convicning the right voters to support barak obama
  • Big obama supporter
  • People were writing him off.
  • Pundits ranged from everywhere
  • Nate silver said obama was toast
  • challenge was how to persuade people to vote for obama
  • simulatede electionb ased on different turnout scenarios
  • Under each scenario, obama could not win unless he changed people's mind
  • How persuade president was a better choice than mitt romney?
  • 2 schools of though-election a referendeum
  • important for campaign to make sure it was a choice
  • hope and change was 2008
  • How make sure message doesn't backfire?
  • How determine which voters the campaign hopes to reach?
  • Targeting swing voters nothing new.
  • Targeting independents
  • From campaign manager-measure everything
  • Mandate to bring analytics to every facet of massive operation in just oneo year
  • Models included
    • support
    • turnout
    • generaic national support
    • contactability
    • many others
  • Persuasion challenge
    • not trying to measure who is likely to support obama
    • not trying to measure who is undecided
    • not tryin gto measure who cares about what isssue
    • Trying to measure who is likely to change his or her mind from voting for rmney to voting for obama
  • How did democracts do persuasion before 2012
    • Built support models for all registered voters essentially probability a voter would support a democrat
    • Messages were tested in focus groups with small numbers of voters
    • Small sample size issue.
    • Basing it on what people like, but not on what is persuasion, the true outcome of interest.
    • For persuasion, targeted those who had middle support scores, or people who were independents
  • Prior to 2012, they went after those who had a middle score, i.e. on the fence
  • middle person means they don't have strong partison characteristics
  • Many people not interested in politics
  • Apathetic
  • Low turnout
  • In 2012, they could easily differentiate supporters from non-supporters
  • Persuasion modeling had early promise
  • benefit of reelection is that you know who is the nominee.


  • asdf
  • sadf

No comments:

Post a Comment