Wharton Professors Eric Bradlow and Peter Fader on "The Data Dilemma"

Knowledge at Wharton

Ask about Knowledge at Wharton

Episode Description

Heartland Payment Systems, a credit card processor, may have had up to 100 million records exposed to malicious hackers. Payment processors CheckFree and RBS Worldpay, and employment site Monster.com have all reported data breaches in recent months, as have universities and government agencies. Experts at Wharton say that personal data is increasingly a liability for companies, and suggest that part of the solution may be minimizing the customer information these companies keep. 📂 Marketing

Read the full transcriptHide the full transcript
  1. this podcast is brought to you by knowledge at Wharton for more information please visit knowledge
  2. Horton upenn edu I think there are really two motivations the first one was a lot of times we
  3. spend a lot of time in the field of marketing look at individual level data so building models that take into
  4. account what you might do what I might do etc unfortunately there's a lot of times
  5. that retailers manufacturers people actually making the decisions don't have that level of data so wouldn't it be
  6. wonderful to be able to make statements about who to target with a particular promotion coupon price cut etc based on
  7. aggregate information and so a more general way of thinking about that is what we would call data minimization
  8. meaning even if you had all the individual level data how much would you really benefit by it and to be honest
  9. the first time I heard about it was it was a talk at MIT by another researcher I'm rarely inspired by other people's
  10. research but I had heard someone else give a talk on this and while I don't think their approach was right it
  11. actually made me start to think about how to actually do inference with minimal data I'm a cheapskate at heart
  12. so I like doing a formal academic research like developing elaborate models but I like to make them very very
  13. practical that means two things one it means trying to use easy software trying to use Microsoft Excel instead of fancy
  14. software that you need to write computer code for and second it means trying to use as little data as possible so after
  15. building some elaborate models using very detailed extensive customer level databases I wanted to answer both of
  16. those questions can we reduce the software requirements and if a paper that does that and it's kind of fun and
  17. interesting but can we reduce the data requirements and the answer is shockingly yes instead of having very
  18. detailed transactions about what each customer's doing at each moment in time if you boil it all down to just a simple
  19. histogram how many people bought from us once this year twice three times and you
  20. show me a series of those histograms over time very easy for companies to collect and store no worries about
  21. privacy you can build the same kinds of models with the same degree of accuracy is if you had the original raw data in
  22. the first place companies are afraid to get rid of the data for the same reason that we're
  23. unwilling to clean out our addicts they're thinking that there's going to be some value that at some point this
  24. piece of data knowing this person's demographics are knowing that person's purchase history from 2004 is going to
  25. be of some value and throwing it away means throwing away assets that's a real problem for companies to do that turns
  26. out that of course it's a great cost to them to store the data great risk in terms of information security there's
  27. also very little value for certain kinds of measures aren't valuable at all even
  28. when they're fresh and other kinds of measures really do get dated so it's important for companies to figure out
  29. what they need to keep what they can get rid of and to focus on among the measures that really are timely and
  30. important what to do with them the data by itself isn't that useful unless you really know how to draw insights from it
  31. and that sounds a bit like a trite statement a lot of people saying that data is that you can't turn it into
  32. knowledge and so on and but it's really true and you have to know what kinds of models to build what kind of statistical
  33. assumptions to make what kinds of equations you need to build around it and there's a very nice interplay
  34. between having a certain amount of data and drawing certain kinds of inferences you can really address one of those
  35. issues without the other and I think we've managed to strike a nice balance for the 2 i'm not sure companies today
  36. know how to predict using the data very well so the problem is they're just trying to keep everything in sight it
  37. might turn out that they have the right measures today might turn out that what they want to do in the future changes
  38. and so even if they had a perfectly predictive model they're afraid that general customer behavior will change in
  39. the future therefore the minute they throw something away it's kind of gone forever so I think part of that
  40. mentality is probably ok but I think thing that companies should always way off against is what is the cost of
  41. keeping all of this massive data and you say it's not just the size of the server
  42. that someone has to build it's not just the big staff that someone has to have to manage all of the data it's building
  43. models on massively large data sets is problematic it leads to companies having to hire experts statisticians expert
  44. programmers and it's not that I want to put myself out of a job I think there is
  45. a lot of value for expert statisticians but now not only have be a statistician you have to be a computer scientist to
  46. deal with these massive databases and so they're essentially looking for people that have all of these skills when most
  47. of the data they collect really isn't as useful as they think for more business news and analysis from knowledge
  48. awarding please visit knowledge wharton upenn BDU

Brief Summary

The episode discusses using minimal data for effective marketing strategies and the importance of data minimization in business.

Key points

  • Data Minimization Insights. Exploring how companies can benefit from minimal data while maintaining accuracy.

Related episodes