“It can’t be bargained with. It can’t be reasoned with. It doesn’t feel pity, or remorse, or fear. And it absolutely will not stop, ever, until you are dead.” from the Terminator (1984)
Analytics is the hottest, absolutely-must-have corporate buzzword these days. Out with the fuzzy and in with hard data. Leaders and Managers have been known to repeat Deming’s quip (and wrongly get credit for it at times) “In God we trust; all others must bring data” when faced with business proposals that are based on ‘intuition and gut-feel.’ Finance and Operations have been using advanced modeling techniques for years, fine tuning the art of showing that big revenue spike (or cost saving) just around the corner. (Hasn’t happened in the last 10 years, but next two quarters are going to be huge!) And now it’s the turn of HR. People Analytics is the next big frontier and machines are being primed to crunch numbers on human attributes and ‘go where no machine has gone before.’
Correct Data is good, Intelligent Analytics is even better. I have had enough experience with gut-feel and arbitrary extrapolation driven disasters in my life to have a deep respect for both Data and Analytics (with the qualifiers firmly in place). I also know that modeling human behavior (or worse projecting it into the future) is something best left to Ethan Hunt.
‘Let’s collect all possible behaviors about employees, get a bunch of statisticians and lock them in a room till they come up with a formula to predict who is good and who is not,’ sounds like a good idea. Only problem – seems people don’t like it. And not just any people – top notch engineers at Google who live and breathe data and analytics.
“Not only must Justice be done; it must also be seen to be done.”
Promotions are a big deal at any organization. It represents an acknowledgement of the company’s belief that you have done an excellent job in your current role and so are ready to take up further or different responsibilities. Nominating the wrong person for a role can be one of the most disengaging acts in an organization. The person(s) who lose out in the race often choose to leave the company and walk right next door to the competitors HQ.
Google has a rather elaborate process involving self-nominations, committees and appeal process for promotions of engineers. As you can imagine this process is costly, time consuming and can be tedious at times. With the rather noble intention of saving a bit of effort for everyone, the People Analytics team decided to explore the possibility of getting an algorithm, which they can use instead.
They did come up with one. And statistically it was awesome!
No possibility of bias, absolute transparency, much less effort, very accurate (based on fitment with past data). One might expect everyone (especially engineers) to love the ultimate solution to getting promotions right. All the disengagement rising from favoritism, bias, perception et.al. out of the window in one masterstroke.
Guess what happened.
The Engineers hated it!
“They didn’t want to hide behind a black box, they wanted to own the decisions they made, and they didn’t want to use a model.”
At the end of all the research, the ultimate takeaway for Google was that ‘people need to make people decisions’ Analytics serves an important role in providing the decision makers with data points and insights, but it can never replace them. It is highly unlikely that we will ever reach (or accept) a situation where algorithms and black boxes are seen as taking decisions (even if you put a human face on the screen). [See Prasad Setty talk about it in the video at the end of this post]
Speaking of algorithms and disasters: Remember the Black-Scholes Equation and 19 October 1987 (Black Monday)? And for those with shorter memories there is the Gaussian Copula function and the 2008 meltdown. And those are failures when modeling movement of financial instruments (and therefore indirectly just one aspect of human behavior which ultimately drives price of those instruments).
It has taken us decades finally realize the tyranny of the “Bell Curve” in performance evaluation though most organizations still are stuck to using what is essentially a convenient misuse of statistical formulation.
Hopefully business leaders will appreciate the pitfalls in giving into the lure of expecting everything to be boiled down to an algorithm. (Elon Musk is rumored to have referred to AI as “summoning the demon”) Even if I don’t quite share Elon’s assessment of the scenario of doom (yet), in my opinion hoping to click a button to decide people’s career path is bit of science fiction, wishful thinking and lazy management all rolled into one.
But if you are a math wiz, don’t care about what old geezers like me have to say about “free will” and social cognition, then your mission, should you choose to accept it …
(Unfortunately this blog post will not self-destruct in 5 seconds)
Acknowledgements and References for this post:
Image courtesy of FreeDigitalPhotos.net
Google came up with a formula for deciding who gets promoted—here’s what happened, Analyze This, QZ India, Max Nisen, November 20, 2014
Recipe for Disaster: The Formula That Killed Wall Street, Tech Biz, Wired Magazine, Felix Salmon, February 23, 2009
The mathematical equation that caused the banks to crash, Mathematics, The Observer, Ian Stewart, February 12, 2012.
Will the machines take over? Why Elon Musk thinks so, Science, The Christian Science Monitor, Anne Steele, October 27, 2014