Eric Siegel: Analytics is not "weird, difficult or boring"
Eric Siegel, PhD, is founder of Predictive Analytics World and Text Analytics World, as well as Executive Editor of the Predictive Analytics Times, and former professor at Columbia University. His new book on predictive analytics has been very well-received, and we thought JMP Blog readers would enjoy hearing from him. We are grateful that he took the time and appreciate his evangelism of analytics.
Eric: Beyond fully setting forth the concepts for business people of all backgrounds who wish to gain value from predictive analytics, I wanted engage, entertain and enlighten even non-business lay-readers in a "pop science" kind of style, to share with the world at large what is fascinating about this advanced technology and the potential it holds to make the world a better place in so many ways.
Are you surprised at how well your book has done?
Eric: Sure! A part of me never thought "regular folks" out there would really take to this "geek stuff"! :) But perhaps that fear helped drive me to put in extra nighttime hours finessing the copy's flow and filling it out with as many interesting case study examples as possible.
What do you think of people saying it will do for analytics what Freakonomics did for economics?
Eric: That reminds me, I still need to send each of those people the $5 I promised them. But seriously, I get a huge kick out of finding ways to show an "arcane" technology is not weird, difficult or boring -- and is in fact super-relevant to everyone. And so it's very gratifying that some believe I've successfully done so.
You point out several applications of uplift modeling (which is a new capability in JMP Pro 11) across industries including personalized medicine, for which a colleague made an analogous quadrant to the one in your book.
It's interesting to contemplate the complexities since this involves payers, providers and pharmaceutical companies — say a drug is found to only be effective against a certain cancer in people with a specific gene, yet doctors prescribe it for people not knowing if they have the gene or not. The latter group may incur unnecessary side effects (and may have in time gotten well anyway), and insurers could avoid paying for ineffective treatments. Might you write another book with more case studies as we see more applications of this still-emerging technology?
Regarding your healthcare-oriented diagram above, the original marketing version thereof corresponds as follows:
Persuadables = "Treatment cures you." These are the individuals for whom the [marketing -OR- healthcare] treatment would cause a positive outcome.
Sure things = "Get well anyway." Those for whom treatment would make no difference. Happy face. :)
Lost causes = "Incurable." Treatment also would make no difference here. Sad face. :(
Do-not-disturbs (aka "sleeping dogs") = "Treatment harms you." Those for whom the treatment would cause a negative outcome.
As for the healthcare benefits of uplift modeling, we're launching the inaugural Predictive Analytics World for Healthcare (September/October in Boston). Keep your eyes peeled for the forthcoming conference program, which plans to include uplift modeling sessions for which this diagram's conceptual point is in fact the whole point.
The Predictive Analytics World conference series, which you founded, is now offering vertical events in manufacturing and healthcare (in addition to government). Why those two industries?
These two arenas are exploding with movement and interest, so we chose them for our two inaugural events in 2014, in addition to continuing our regular business-facing Predictive Analytics World events in San Francisco, Toronto, Chicago, Boston, London and Berlin, plus PAW Government and Text Analytics World as well.
Anything else you’d like to share?
Eric: Beyond my own writing (book and articles), there's an abundance of quickly growing content in the Predictive Analytics Times, which was just last summer updated to be a full-fledged online portal/news outlet.