A Quant, Physicist, & Chemist Walk Into HR

A Quant, Physicist, & Chemist Walk Into HR

HR is not the first industry to see some cross pollination oddities. In 1991 some strangeness was taking place in Sante Fe, New Mexico of all places where a group of punk physicists wearing "Eat the rich" t-shirts were applying black box trading techniques to the stock market. Their story was later written in the book the Predictors:

Since the success of the prediction company other hedge funds and wall street quant shops have hired a range of disciplines outside of quantitative finance including biology, meteorology, and in my case biomedical/chemical engineering.

Cross Industry Pollination Creates Mutations:
There is a lot of value hiring outside the core discipline because it adds a layer of diversity that can lead to creative breakthroughs (mutations). For example, traditional stock models had focused most of their attention around Markowitz optimization and cleverness around the covariance matrix. New cross pollination brought neural networks, genetic algorithms, sentiment trading algorithms, and deep-learning to stock.

How Does One Go From Wall Street To HR?
During the summer of 2013 I was conducting my personal market adjustment research and my friend Jake Reni told me about a hot startup in South Jordan, Utah named HireVue. As soon as he said it was an HR company I instantly said "No thanks". I had a strong stereotype of HR, coming from Intel/Micron I had always classified HR as the department for problems:

HR: This is where the adults who haven't figured out adulthood go to be disciplined.

Later when I learned HireVue was closing a round from Sequoia capital I took a closer look at the company and saw an incredible opportunity to cross pollinate. This opportunity exists in every HR organization, an industry ripe for disruption.

HR Has Some Of The Most Difficult Data Problems:
I have worked on some hard problems ranging from the hedge fund work to semiconductor, the problems encountered in HR are by far the most challenging. The data is typically unstructured, sparse, and messy. Employee satisfaction surveys are open ended, rating systems are sparsely populated, and subjective metrics and decisions are scattered throughout. If all of that wasn't hard enough add adverse impact possibilities everywhere. HR needs cross pollination to handle these problems and to leverage the latest available data tools being used in other industries.

Resume modeling:
Take resume modeling for instance. This is one of my favorite ice breaking questions for HR analysts when I ask them:

How would you go about creating models to stack rank resumes based on performance?

The traditional HR analyst's brain enters blue-screen-of-death right now, they have no idea. The reason they don't know is because this is a very unstructured dataset. You have content, layout, and file type as potential inputs. You also have sparseness where plenty of expected features (i.e. GPA) can be missing. This is a fun problem on so many levels. A data scientist will see they must use a third party parsing service to either build a structured categorical/numerical Excel style dataset, explode the entire thing into raw text, or convert the published format to a raster image for downsampling and unraveling. These actions, which may sounds like gibberish to some, are steps taken to convert this raw data form into machine ready inputs. Unstructured>structuring.

Interview Modeling:
Do you ever make hiring decisions based solely on a single resume or assessment? No, of course not. You typically do an interview. So why not model the interview? If heads weren't spinning before around the resume modeling approach, now consider how you would handle a 10 minute video interview as a model input? This suddenly becomes a hard problem, even for the quants and data scientists of the world.

HireVue will be presenting with Twitter, LinkedIn, Facebook, and Google around their text prediction capabilities used for part of their interview model in San Fransisco on April 29-30th here. Should be a great event for those interested in leveraging text processing capabilities.

Call To HR:
If you are a nuclear physicist, chemist, mechanical engineer, data scientist, meteorologist consider HR. A new hunting ground for some of the most difficult problems.

Keywords:
HR Analyst, big data, unstructured data, structured data, random forest, deep-learning, predictive analytics, sentiment analysis, resume modeling.

///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
Here are my most popular posts:
This Is Why Your Data Scientist Sucks (10,249)
4 Reasons To Work For A Startup Instead (3,731)
Death Of The Data Scientist (2,859)
How To Find The Smartest Data Scientist! (2,659)
How To Land Your First REAL Data Science Job (2,505)
This Is Why Your Plots Suck
Pinch Time Not Pennies

Abhilash Bodanapu

Global Head of People Analytics Solutions@Capgemini

8y

Beautiful thoughts Benjamin. Thanks ! So far I've thought off Techies coming into HR more aggressively with analytics picking up. But your thoughts truly encourage us to build a Design Thinking based HR and this kind diversity in HR function is surely going to help us to be future ready.

E.K. TORKORNOO, M.Sc. (Econs), CCP

Consultant, Contractor & Change Agent: Board RemCo, Governance, People / HR / Talent, Transformation, Total Rewards (Compensation, Benefits, Pfce Mgt., Recognition, Wellbeing, EX, etc.), OE, OD, Leadership

8y

Very interesting. Some things lend themselves to analytics. Others don't; at least not so easily. On the somewhat facetious side of things here is one: Can you predict the future level of performance of internal or external candidates for a promotion based on their facial expressions? Other body language consistent with cultural norms? Some old body of work tells me that many techies (including professionals in engineering, science, and math) would flunk because they tend to have body language that is usually outlier in nature, relative to 'norms' suggested by some writers for SOME cultures (e.g., look you in the eye, give a firm handshake, not fiddle or look elsewhere consistently when given tough questions, not develop a tic somewhere in the neck, not stutter, stumble, mumble, ramble on in technicalities nobody understands, etc., etc., ad nauseam). What if the techie was from one of those cultures where you avoid looking people in authority in the face. Do we develop some algorithm for all these cultural variations? Can we predict levels of sustained employee engagement (or other desired attitudes and behaviors) when given: 1. a flat across-the-board amount of pay raise? 2. a flat across-the-board percentage of base pay as a raise? 3. a massive group discount on land for building their homes? 4. a massive discount on cable or satellite TV programming? The answer to all these real-life problems? It depends. Modelling may help. but the imponderables are too many. And such is the nature of (wo)man. Enter the need for judgement. Numbers and analytics help. But they are just aids to decision-making. Not absolutes. In the realm of Humans, people can be difficult to figure out. And just when you think you got it, they change, again.

Like
Reply
Satish Salivati

Talent Assessment and Analytics Specialist ♦ #Blockchain Practitioner

8y

Nice one Benjamin. Food for thought for a lot of Recruitment/HR folks.

Like
Reply
Joe Pimbley

Principal of Maxwell Consulting, LLC

8y

I love this post because it raises genuinely new ideas (modeling a resume or interview, for example). Also absolutely true that cross-discipline thinking and action is excellent. But now let me show my contrary streak. (Good quants, physicists and chemists, in my opinion, are reflexively contrary.) I fear that the STEM-type people we're lauding here are simply applying their favorite tools of modeling and data analysis. Just because you have a tool you really like does not make that tool the best for the problem. In this case, there are evident alternatives to a souped-up (possibly Rube Goldberg) prediction algorithm based on resumes, interviews, blood samples or whatever. Instead of prediction, throw the candidate into a real work situation (a day, week, month, maybe longer as a contractor) and measure what happens. OR, for another data analysis-based concept, identify your firm's star performers in various roles and reverse engineer THEM with lots of measurements and analysis. Then hire candidates that fit the most dominant positive parameters. The secret benefit of this idea is that it requires the firm first to define how to measure value of existing employees. That's an UNSOLVED PROBLEM, in my view.

Riccardo Bua, MBA

Cybersecurity - Technology - Customer Experience Executive - I design secure solution for Agile - Digital transformations (Critical Infrastructure, Governance, Risk, Crisis Management and Enterprise architecture)

8y

Nicely written :-)

To view or add a comment, sign in

Insights from the community

Explore topics