Andy Pulkstenis is a program director of analytics for State Farm Insurance in Bloomington, Illinois. In this role, he leads a team of advanced analytics professionals who provide statistical analysis and predictive modeling support for the enterprise across a variety of business units.
So you’ve made the decision to pursue a career in statistics, but what now? Generally speaking, there are four primary employment avenues:
-
Academia
Government
Biostatistics
Business
Multiple articles could be written about each of those areas, but I want to speak specifically to those who are contemplating an applied statistics career in a business setting.
Why Should I Consider Business?
Security: Whether you call it business analytics, data mining, predictive modeling, or applied statistics (at State Farm, we call it “advanced analytics”), growth in this area has been explosive over the past 10 years, and the outlook appears even more outstanding.
Various companies (and even entire industries) are realizing the importance of proper analytics and the value tools like predictive modeling, experimental design, and regression-based business analysis add to a firm. When they make a successful movie starring Brad Pitt about baseball teams using statistical analysis to direct player acquisition strategies, it’s safe to declare that analytics has gone mainstream. The emergence of Big Data concepts continues to force the issue, requiring old school, Excel-based MBA approaches to be replaced with more powerful statistically based methods.
What gets you the job?
– Master’s degree or above (statistics)
– Internships/applied experience
– Communication of technical concepts, especially the ability to map a business problem into the statistical world and map the stat solution back into business terminology
– Problemsolving
– Technical ability (know “why” vs. just knowing “how”)
– Logistic regression with Big Data
– GLM/regression theory
– Breadth and depth beyond logistic and GLM
– Useful “niche” skills like DOE, time series, survival analysis, or multivariate methods
Deal breakers?
– Poor communication
– Technical inability
– Being wrong, but convinced you are right
– Lack of problemsolving ability
– Faking it (know what’s on your résumé)
As demonstrated by “Moneyball,” better analysis leads to better strategic insight, which in turn leads to better decisions and a direct positive impact on the metrics that matter to an organization. The statistician’s toolbox comes loaded with methodology and approaches well suited to the problems encountered in a business setting, and quite frankly the demand continues to dramatically outpace supply. It was an issue when I entered the work force nearly 20 years ago, and it’s even worse today. If you are a well-trained statistician, that’s what we call a “buyer’s market.” At State Farm, we’re always on the lookout for quality graduate-level statisticians.
Variety: People often comment that you get to play in everybody else’s sandbox as a statistician, and that is true in business settings as well. Throughout my career I’ve worked on problems involving credit cards, loans, marketing, the Internet, greeting cards, grocery stores, airlines, tire manufacturers, auto manufacturers, theme parks, chemical manufacturers, and pharmaceutical companies. In my current role at State Farm, my analytic project partners have included actuaries, our banking business, marketing, our Internet group, insurance agents, claims, IT….It’s tough to get bored when you are always exposed to new parts of the business.
Variety (Part II): So the business areas are varied, but what about the methodology? Many people have the mistaken idea that applied statistics in a business setting is basically straightforward logistic regression, linear regression, confidence intervals, rinse and repeat, but that couldn’t be further from the truth. Sure, those techniques are widely used and applicable to a broad number of business problems, but I’ve also been involved with projects using advanced time series methodology, cluster analysis, PCA, nonlinear regression, experimental design, machine learning, multivariate outlier detection, decision trees, and survival analysis over the past year. One thing I love about business analytics is how often and deeply we reach into the corners of the stat toolkit.
Challenge: The complexity of the problems we face in business is primarily driven by the following:
1. The size and inherent bias of internal observational data sources
2. Regulatory or organizational constraints around any solution we’d like to implement
3. An environment in which many conditions theoretically required for clean methodology are violated in one way or another
The combination of all three often leads to challenging situations in which we are trying to find the most helpful yet “least wrong” approach to the problem versus finding the theoretically correct answer. And because business is competitive, there is not usually a ton of published research to draw from. It may require pulling analysis ideas from biostatistics to apply to a small sample problem in auto research, or applying manufacturing quality control ideas to some non-manufacturing business process. Recently, my team had to find a way to apply predictive modeling concepts in a useful way to a situation in which we didn’t have a single instance of a target variable! It can be quite challenging, but incredibly rewarding, when successfully solving some previously “unsolvable” problem that doesn’t quite fit into anything we learned in our books.
Academia, government, and biostatistics tend to get more promotion at the university level, but don’t neglect this growing and rewarding field of “advanced analytics” in a business setting. The security, variety, and challenge are off the charts, and the breadth of techniques you’ll encounter present a great environment for continuing to develop technically throughout your career while you grow as a professional.
I’m just about to enter a master’s program in statistics as a 48-year old with a Ph.D. in the social sciences. In order to plan my degree program, I’m interested in knowing to what extent programming skills are useful in addition to those skills listed above. Does it make sense to take computer science courses, or to focus on the stats while in grad school?
Hi Steve,
Your question is an interesting one. Five+ years ago I would have said “focus on the statistics” but the rapidly evolving computing environment, the recent innovations in parallel processing and high performance statistical computing, and the rise of “Big Data” as a hot topic have led to many hybrid analytic positions where someone needs to be conversant in statistical concepts but also proficient in Python, Hadoop, or some other such programming specialty.
I suspect that going forward there will continue to be many satisfying analytic roles for pure statisticians, pure comp scientists, and for people who are a blend of both. So the easy answer is focus on what you want to focus on, but maybe dabble in some of the other side too. It will benefit the statistician to at least gain familiarity in some of the newer computing environments – I seem to see Python and Hadoop a lot on resumes for people with comp sci backgrounds and also in analytic job postings. It will also benefit the computer scientist who wants to work with analytics to become familiar with some of the most commonly used tools, such as logistic and linear regression, or maybe machine learning is a better fit if you are more comp sci focused. Perhaps talk to your professors and the ones in the comp sci department to help pick the right subtopics for you and your goals.
I tend to be more purely statistics-focused, but that may be more the result of my background than anything else. I also prefer to cover all bases by having some team members that are more comp sci focused, or have advanced degrees in pure mathematics, for example, in addition to traditional statisticians. I think in the end analytics is becoming so wide spread that we will continue to see expanding opportunities for both specialties, and new opportunities for those that want to walk in both worlds.
Thanks for your reply, Andy. It’s very helpful.
Steve
Thank you Andy sharing all your insights!
I’m a 27 yr-old with an undergrad degree in Economics who is enrolled to take real analysis and statistics courses this year in preparation for advanced study.
My question to you is does an analytics degree is better or worse suited to the private sector vs a pure masters in applied statistics?
(like those analytics degrees listed here: http://www.informationweek.com/big-data/slideshows/big-data-analytics/big-data-analytics-masters-degrees-20/240145673?pgno=11)
In your opinion, does choosing either of these degrees over the other have a significant effect on the career arc of the student? Are they viewed differently by employers in the private sector? Thank you!
Hi Ravi,
Apologies for the late reply – I don’t get a notification when someone posts a question. I think it really depends on the nature of the analytics degree, and the specific material covered.
I recently went on a recruiting trip to a top school with both a stats department and a very well recognized “analytics” program, but we only visited the stats department. That was intentional because we have found that the graduates we have seen from that analytics program didn’t understand the fundamentals of statistics or modeling, and that’s so important to what we do it was a deal breaker. They knew how to push the right buttons and in what order, but the conversation broke down when we asked them why they push those buttons, and what happens if they deviate from that script?
Contrast that with the University of Illinois, which also has both, the analytics degree being noted by an “Analytics Concentration.” At the U of I this degree is essentially a full MS in stats with a couple of additional applied SAS courses, so we hire lots of those candidates.
My OPINION: the analytics programs in general don’t provide the statistical knowledge I look for in the teams I lead, so I seek the traditional MS and PhD statisticians (and econ, and econometrics, etc). But these anlytics programs (the better ones) don’t have any trouble placing graduates so there’s clearly demand for that mix of expertise as well. Your best bet is to decide what you want to focus on for your career, and hit the program that best prepares you for that. Do you want to be an expert practitioner of statistics who learns the business stuff on the fly or a more well-rounded analyst who has broad knowledge but isn’t particularly deep in any area? My vote would be traditional stats at a top 30 school because the business stuff is easy to pick up if you can do grad-level statistics(US News & World Report has some rankings that are helpful), but I’m admittedly biased by my background, my experience with graduates from each type of program, and my own analytic needs when hiring. It has nothing to do with the insurance industry specifically, though. My hiring stat focus was the same when I interviewed in my consulting days and also at Capital One. I’ll stop here and see if you have any follow-ups.
Predictive modeling for insurance companies is growing exponentially. There is huge upside for statisticians looking to jump into the big data arena.