A few years ago a large insurance company decided to form a new data science group. The senior managers of the company were determined to find the best leader possible. Many were inclined to hire a prominent Ph.D. in finance at a major research university as their head of data science. They identified a candidate with great credentials and brought him in for interviews. He had an excellent publication record and strong analytical skills, but he had little experience in managing people, little focus on most aspects of the industry, and didn’t seem interested in day-to-day management activities. The company quickly surfaced another candidate-an MBA who had run an analytics group at another company. He was judged less likely to create brilliant new algorithms or publish his research, but was an experienced analytical manager who could both hire and inspire people. The company made him an offer, and he accepted the job.
We think this company made the right move. While we are supportive of companies’ efforts to hire quantitative Ph.D.’s to practice data science, we believe that most firms are better off hiring people with other types of training and general management skills to manage analytics and data science groups. Why? Because there are a series of traits that make for effective managers of such groups, and most Ph.D.’s don’t tend to have them. At least one “Chief Data Intelligence Officer” has suggested that Ph.D. degrees are behind the failure of some analytics leaders. We’ve seen no data on the topic, but he may have a point.
We describe ten of those traits below, and the reasons why they are unlikely to be found in the average doctoral degree holder. The list of traits may be useful for anyone seeking to hire a leader of analytics or data science functions-whether they are considering Ph.D.’s or not.
1. RECRUITMENT / RETENTION / PEOPLE DEVELOPMENT
The competition for analytics and data science resources is fierce, qualified resource supply is low, and an analytics leader needs to be good at attracting, hiring and retaining top talent if analytics initiatives are going to deliver the desired target business and economic value. This is a time-consuming process that requires a vision and the ability to sell people on the vision, to get the right people on the bus, and then to keep them engaged, happy and productive. Hiring for “soft skills” like communications, work ethic, attitude and cultural fit, is as important as heavy duty technical skills. Ph.D.’s haven’t generally had much business hiring experience, and if they hired other academics, they may have attempted to find more people similar to themselves.
2. GENERATING DEMAND
Doug had a boss in the software industry who used to say, “Nothing happens until somebody sells something.” A primary role of the analytics/data science leader is to sell quantitatively-oriented projects to the business people who need them. It’s critical to identify projects that a) are aligned with the company’s strategy and performance targets, b) can garner a high level of executive support, and c) can secure the necessary funding and resources to execute the projects and deliver the desired business value. Ph.D.’s may have experience selling research projects to the National Science Foundation, but most have little experience selling analytical solutions to businesspeople.
3. UNDERSTAND THE BUSINESS DOMAIN IN QUESTION
Gordon Bethune, the legendary former CEO of Continental Airlines, used to say that “If you are going to run a watch company, you better make sure you know how the (expletive omitted) watch works.” In short, understanding the business domain is critical before adding analytics or data science to it. There are often “business rules”, e.g., contractual, legal, or regulatory implications, that constrain the decision or limit the full measure of benefit that can be achieved. As Doug has learned at Southwest, doing analytics successfully for an airline requires more than a modest understanding of how an airline operates. And within an airline, applying analytics to revenue management or network planning is very different than using them in network operations control or ground operations, or the more traditional business functions such as HR, finance or marketing. Most Ph.D.’s, of course, do not come into businesses with a high level of understanding of business processes and functions. Some even come with little desire to learn about them, which we believe is a fatal flaw.
4. RELATIONSHIP BUILDING
Trust arrives on foot and leaves on horseback. People don’t care how much you know until they know how much you care. Trite clichés? Perhaps, but many analytical leaders tell us that trust is critical to success. Analytics and data science often expose huge opportunities for performance improvement, which inevitably can make it look like someone wasn’t doing their job, or is less than completely competent. Depending on the organization, it can take years of painstaking effort and a lot of coffee and lunches, to build trusting relationships. A business leader has to have a big problem that they really need your help with before they will risk their budget, career, and political capital on any project, let alone one involving a heavy dose of math they don’t fully understand. People with Ph.D.’s can inspire trust, but it may take longer than for people whose backgrounds and attitudes are more similar to their internal business customers.
5. CHANGE MANAGEMENT
Analytics and data science can often drive enormous changes in business processes, organizations, and jobs. While analytics leaders may not need to be the “change management guru” in their companies, they need to be very sensitive to the shock waves that analytical results can have on an organization-the human impact as well as the financial, operational, and economic implications. There will inevitably be changes in data requirements, systems, processes, organizational structures, and decision-making (inserting the “model” in the human decision-making loop). Most Ph.D.’s have rarely had to oversee such change unless they were unfortunate enough to spend time as a dean or other senior university administrator.
6. ANALYTICS/DATA SCIENCE MODEL DEVELOPMENT AND DEPLOYMENT
Analytical and data science models and systems are often built using a project management methodology approach, most commonly agile or scaled agile (SAFe). Scope, timing, resources, and quality are the four primary dimensions of projects, and scope (creep) often is the most difficult to manage. Experienced project managers will know how to wrestle with these factors intuitively; Ph.D.’s are likely to have little exposure to them.
7. COMMUNICATION SKILLS
Even the greatest analytics or data science project using the most sophisticated techniques and delivering substantial business value and ROI is completely useless if you cannot convey its value and impact. The communications must be aimed at both business (at all levels from the Board of Directors and executives down to rank-and-file individual contributors, and everyone in between) and technical (Ph.D.’s in physics to B.S. in computer science grads) audiences alike. Leaders need to know how best to communicate complex concepts to each audience type. Ph.D.’s may be skilled at communicating technical or mathematical ideas, particularly if they have some experience in teaching. But they may not know the language of business.
8. PLANNING / BUDGETING / ADMINISTRATION / P&L MANAGEMENT
Analytical and data science leaders may be called upon to run their teams like a business. Like any successful business enterprise, the value the group adds must offset its costs. Experience running either a consulting group or a product-oriented business is great training for this skill set. Most Ph.D.’s are unlikely to have such a background.
9. PRACTITIONER EXPERIENCE AND EXPERTISE
Analytical and data science leaders, like many managers of knowledge workers, must be player/coaches, capable of at least occasionally working as a practitioner. Such an ability requires an in-depth knowledge of the analytical disciplines, i.e., methods, lexicon-jargon, KPIs/metrics, tools, and technologies. The best leaders of quant functions will have worked at least 3-5 years as a “hands on” analytics practitioner, i.e., one who builds models and analyzes data for a living as a primary job function. Understanding the math underlying models, the data underlying the business domains, the technology underlying systems, and most importantly the process and methodology of developing, testing, verifying and validating sophisticated models that solve complex business problems requires hands on experience to make judgment calls. Some Ph.D.’s will have such experience, but most probably don’t.
10. INFORMATION TECHNOLOGY (IT) EXPERIENCE AND EXPERTISE
Analytics and data science practice inherently involves technology. Data, software, servers, and transactional systems all come into play-not just a laptop running R or Python. Models may make the enterprise smarter, but models embedded in production systems and business processes make the enterprise more efficient. Whether the analytics/data science team builds the systems in which to embed your models or someone else will (perhaps the IT function), analytical leaders need to understand how to make models perform at scale, i.e., high availability, high reliability, robust/fault tolerant. Such IT issues are rarely found in university research labs.
Of course, there are exceptions to any set of generalizations, and you may well find a Ph.D. who has all or most of the capabilities we describe. Chris Lofgren, for example, has a Ph.D. in Industrial and Systems Engineering, and was able to lead an analytics and engineering group at Schneider National, the trucking and logistics firm. His managerial aptitude must have been high, since he also has been CEO of the company since 2002.
It’s certainly true that sheer intellectual horsepower, which many Ph.D.’s have in substantial amounts, is always a useful trait. It’s also a good bet that a Ph.D. in a scientific or quantitative discipline will have high levels of statistical and mathematical aptitude. However, we suspect that pure IQ is less important than many of the other factors we have mentioned, and statistical and mathematical abilities can often be found in people with degrees other than Doctor of Philosophy.