Three of the top five “most promising” jobs in the United States this year are high-tech data positions. But not every role on an analytics team requires a math or science degree. “In fact, in the role of a translator, having a computer background can go against you,” says Prudencio Pedrosa Grandes, a partner based in Madrid who holds degrees in economics and business.
He should know—he was the first analytics translator at McKinsey and helped to define the role more than 3 years ago. “I don’t have advanced programming skills. But I think logically and understand how to make analytics create tangible value in day-to-day business—that’s the crux of what the translator does.” In late 2013, Pruden worked on an early analytics project, helping a bank reduce its number of nonperforming loans—and he was hooked.
Jump ahead to 2015, and data technologies such as machine learning were gaining traction. As demand for analytics work grew, the need for the role of a translator to lead projects was clear. An early goal was to train a group of some 200 translators, including delineating the role, establishing a learning program and defining a career path.
This year, the translator program…with Carnegie Mellon University, will have graduated more than 1,500 translators in the Americas, Asia, and Europe.
Jordan Levine, a senior analytics learning manager trained in math and engineering, had joined McKinsey in 2011 after a stint in the Marine Corps. He, along with Pruden and Aaron Horowitz, a junior analyst at the time, quickly assembled 30 participants for the first learning program—“We got about 50 percent of the curriculum right”—and they have been refining the training ever since. This year, the translator program, which includes classes conducted in partnership with Carnegie Mellon University, will have graduated more than 1,500 translators in the Americas, Asia, and Europe.
“Translators come in with a wide range of experience,” observes Jordan. “We teach them a hybrid of skills essential to leading projects with analytics components. This means articulating a problem statement, identifying and collecting data, applying a modeling technique, leveraging the model to influence decision making, and capturing the value from better decisions.”
In the classes, students can present use cases they are currently working on. “We’ve seen projects to optimize railroad use, improve the quality of corrections facilities, and even to find the most effective way for power companies to trim trees to reduce outages,” says Carla Vale, who started her career in physics research and now runs the learning program.
So what do translators actually do? They have the traditional responsibilities of an engagement manager—leading a team, managing the project, and serving as the key liaison with the client. And they apply their domain expertise in an industry or function, using a deep understanding of the drivers, processes, and metrics of a business to better leverage the data. And, of course, they have a fluency in analytics: wrangling large data sets, understanding the ins and outs of various algorithms and models, managing “quant” colleagues such as data scientists and developers, and knowing how to turn data insights into concrete measurable actions.
“Analytics projects are different,” explains Carla. “In other types of work, we talk about the 80/20 rule [the belief that roughly 80% of the effects come from 20% of the causes], but not in modeling. There is no quick and easy answer—it takes time and painstaking effort to manage multiple, immense data sets. There are false starts, dead ends, and a lot of ambiguity. We have to find new tips and tricks so that translators can help their clients know what to expect.”
We developed a model with an almost perfect GINI score…but it was a complete failure…
Once the translator and client define which business problem to focus on, they have to identify the data that are the most valuable: “It is like opening a closet in complete disorder,” says Pruden. “We meet with different groups—business intelligence, IT, and sales, for example—to ask about the types of data they have. Everything is in different formats, measures different things, and is sometimes incomplete. What can create the greatest value? Is it usable? Is it relevant? How do we put everything together to build the model?”
The translator then needs to know enough about the nuances of various models to ensure that the team solves the client’s problem. Pruden shares an early experience with a class. “We developed a model with an almost perfect GINI score. Data were accurate, current, and clean. It was beautiful,” he pauses, “but a complete failure because the model couldn’t be updated in real time.”
Translators then help the client integrate the analytics model and data results into their ongoing processes. One sign of success, Carla points out, “Last year we extended our translator academies to several clients. They have to know how to use the model themselves after the project is done and our team has gone.”
This year, it’s estimated that some 25 percent of our projects will include analytics. Does this mean a promising future for this new role? “For now, yes,” says Pruden. “But the role of translator is transitory—in 5 years, we should all be translators.”