We spoke with UC Davis Professors Dr. Bhargava and Dr. Saigal to learn more about prescriptive analytics and how it’s used in business settings. We also discuss their thoughts on the fundamentals of a solid analytics education, the UC Davis approach to teaching analytics, and the top resources for analytics students.
Hemant Bhargava earned his Ph.D. from The Wharton School of the University of Pennsylvania. He is a professor and Jerome and Elsie Suran Chair in Technology Management at UC Davis. His research focuses on many topics, including technology management, management information systems, and the IT industry.
Sanjay Saigal earned his Ph.D. from Rice University. He currently serves as executive director of the UC Davis Master of Science in Business Analytics program. His resume includes tech executive in Silicon Valley, founder of the St. Stephen’s Institute for Management Excellence in New Delhi, India, and professor of Business at Minerva Schools at KGI.
Let’s say that yesterday the pricing engine triggered surge pricing and the CFO wishes to evaluate its impact. Did the surge pricing successfully activate more drivers? Did it suppress ridership? Did it lead to higher profits? Was it a net positive? Backward-looking diagnostic methods that used to answer such questions fall under descriptive analytics.
Following such an audit, the CFO may wonder if making the surge pricing engine more responsive could impact key performance indicators (KPIs) even further. She asks her staff to forecast ridership impact. Techniques for such look-ahead projections belong to predictive analytics.
More robust business intelligence (BI) techniques and systems also fall under descriptive analytics. Finding relationships between classes of data using statistical regression may be thought of as descriptive analytics. You can see that a quantitatively-inclined business analyst is well-positioned to do a lot of such work in a spreadsheet.
Predictive analytics focuses on what is likely to happen. In the generic sense of the word, it’s forecasting. Trend-detection tools such as classical statistics (e.g., time-series analysis) and operations research (e.g., Monte Carlo simulation), to newer methods such as data mining and machine learning (e.g., k-means clustering), are commonly used for predictive analytics. Going back to our car-sharing example, estimating how much the average driver will make over the holiday season is a prediction task. Some of these techniques are usable by the typical business analyst in a spreadsheet, but many require either statistical software (such as R or SAS) or specialized packages.
If you know what happened, and what that implies for the future, your challenge is: what can you do next, say, to improve profitability? Prescriptive analytics is about finding “best possible” actions. This is also called optimization.
The car-sharing company manager may create a profit-maximizing promotion plan for the coming holiday season based on projected demand and driver availability through heuristics (e.g., genetic algorithms) or structured optimization (e.g., integer programming). Sophisticated optimization approaches even account for uncertainty. For example, they can provide anticipated profitability over a range of possible gasoline prices. Though it is possible to do prescriptive analytics on spreadsheets, higher value work usually requires specialized tools (e.g., Gurobi or SAS/OR). It is usually the province of quants – analysts with specialized training.
The US military flew autonomous drones using integer programming more than 15 years ago. But today, autonomous vehicle control systems are primarily based on machine learning. Machine learning methods are used in all three analytics categories we started this chat with: data visualization, prediction and control. So the three categories are somewhat less useful today. I can’t speak about other analytics centers of excellence, but here at the Graduate School of Management at UC Davis, research and teaching activity are not much influenced by them.
HB: Making analytics useful involves change, and so you run into all the common obstacles to change. First, even when all parties are better off, change is hard because of inertia. Many years ago I worked with the US Marine Corps to redesign their system for allocating new recruits to training schools. Allocations were done every week, covering dozens of classes starting that week and hundreds of new recruits, keeping in mind technical requirements, qualifications, etc. The choice of school was the key determinant of the recruit’s future career path. Therefore, to a recruit, it was important to get an allocation that matched their desired direction. However, these schools are a big expense to the Marine Corps, so allocations were made to minimize the number of class seats left unfilled. We designed new technology that managed both “fit” and “fill” and smoothly integrated into existing internal systems. But convincing stakeholders to shift from a “minimize fill” to a “maximize combination of fit and fill” was a substantial challenge. A bigger challenge, though, related to organizational power and politics. Change may not make everyone equally better off, and some might in fact be worse off, and will resist change. The previous system required constant action by external contractors (and payments for these maintenance contracts), who were predominantly retired officers from the same unit in charge of making these school allocations. Hence, there was substantial resistance on this front, because the newly designed system could potentially eliminate a worthwhile career path. Such organizational issues arise in nearly every practical analytics project, and they have to be dealt with smartly.
A third factor to consider is that the analytical models may not be a reasonable enough approximation of reality. As the saying goes, “all models are wrong, but some are useful.” Modeling is an art. When the model artifact captures the core features, it can contribute valuable lessons even though it might have left out several less relevant pieces. Stakeholders have then to be persuaded that the recommendations make sense, or advised how to adjust the recommendations to cover these other factors. But if the model does not capture the essential elements of reality, then the recommendations will not be robust, and the analytics-based technology will not suitably improve upon the human experience and judgment underlying the existing action.
U.S. business schools now offer more than 50 masters programs in business analytics or close variants. These programs combine management and analytics technology in diverse ways: online and traditional classroom, from nine months to two years in duration, research-heavy to market-oriented, directed to traditional industries to tech sector-facing…you name it! What is common to all of them, though, is a focus on creating quantitatively-savvy professionals for a new age of management. Our MSBA program starts in San Francisco this fall.
When I start talking about our program it’s hard to make me stop. So I’ll direct readers to our website and our FAQ instead. But here are three facts to take away. First, the program is taught in San Francisco. As you know, that’s the center of the analytics universe. Our extension, located in the UC Hastings campus, is a $5 ride from analytics leaders like Uber and Twitter. Second, the program is ten months long. With our combination of high-caliber students and best-in-class instructors, that will make it a rather intense ten months! And finally, our students will spend 25 to 50 percent of their total time on the Practicum – an industry-sourced project that begins the first day they set foot on campus, and ends the last time they leave the classroom. These are real projects, with real analytics-derived business value for our industry partners. And of course, for our students, the Practicum is an on-ramp to placement. There is no lack of opportunities to “get their hands dirty with data!”
To answer your second question, we’re looking for students with one or more of four skill-sets. It goes without saying that MSBA students need to be quantitatively sophisticated, but we don’t limit admission to STEM graduates. Some of our brightest students arrive from majors like psychology and journalism. We also look for a certain comfort in computing. If you’re a hacker with a populated Github repository, all the better! Many of our students aspire to shift an ongoing career into a higher gear. From mid-career students we looks for either evidence of business accomplishment or a record of organizational achievement (e.g., in the non-profit sector).
HB: Sanjay mentioned the Analytics Magazine from INFORMS. For those interested in practical applications, I recommend the journal “Interfaces.” For more technical and academic articles, one can look into journals such as Marketing Science, or specific area journals in Statistics and Computer Science. University and departmental sites such as the Living Analytics Research Center at Singapore Management University are also great resources.
Possibly the best resources in metropolitan areas are the diverse analytics, data science and entrepreneurship discussions that can be found via meetup.com. Our program supports several meetups in the Sacramento Capital Region and the San Francisco Bay Area by providing GSM faculty doing analytics-related work, and our great campus facilities in Davis and the Bay Area. That is where you go to learn what is happening today on the frontiers of research and practice.