Mohammad Ghassemi is an independent data science expert with over a decade of experience working with executives who want to leverage data science—from machine learning and data mining to recommendation systems and decision support—to make their businesses more competitive, innovative, and effective.
In our latest Expert Q&A, he talks to us about how to organize data and teams to drive value across the enterprise.
You’ve got a very unique background, with a PhD in Computer Science and experience as a consultant at the Boston Consulting Group.
I was admitted into Cambridge almost 10 years ago to do a Master’s in a business area. When I got there, although I enjoyed the program, I ultimately decided to switch into computer science because I became convinced that, in the future, doing effective work in the business community was going to be very technical in nature.
Most companies know they have a mandate to use data to make smarter decisions. How does your work fit in?
Many of the clients we deal with are in areas where human judgement plays an important role, like healthcare, finance, and law. For these clients, and more generally, there is a trend to use data science to augment, not automate, the workforce—that is, to combine human and artificial intelligence to multiply the effectiveness of an organization’s talent pool and solve problems that neither humans nor machines can solve as effectively alone.
Our group delivers both strategic leadership and technical execution on all aspects of that journey and its related technology innovation.
What are some of the challenges your clients face as they try to focus more on data?
Many of the companies we work with have access to rich data resources and talented teams. But they don’t know how to leverage those resources to drive value back to the organization. How should they organize the data so that their engineers can use it and, more importantly, so that non-technical folks won’t lose the ability to leverage it for ongoing business activities? How should they structure data so that it is easy to expand, cost-effective, and secure?
Structuring teams for success can be a challenge, too.
At many organizations, the data science teams are very new. These folks might be technically talented, but that doesn’t mean they’ve been trained on how to contribute to a software engineering project. They are probably used to having complete control over their projects and using whatever languages they want. They may have collaborated with others, but usually not in the same way you’d expect at a large organization, where there’s periodic code reviews and so on.
Engineers may also fall short in articulating the ROI of their projects. Once a data science tool is built, clients want to know how to translate the performance of that tool back into a figure that is meaningful for their business. For both sides, bridging that communication barrier is essential for success.
What are some of the most effective ways you’ve seen or helped clients address those issues?
In order to use data science to solve a problem in the real world, you’ve got to have intensive, honest conversations with an organization’s stakeholders about what they want to accomplish, whether it’s reducing the number of deaths at a hospital or improving their ability to predict a loan default. The better we understand a client’s objectives and pain points, the more effectively we can develop tools and procedures to help them use data science to solve them.
However, once everyone knows what the tools need to do, we’ve also got to communicate about how that might change moving forward. Inevitably, a company’s goalposts will shift, but if we’ve constructed our data science ecosystem in a thoughtful way from the beginning, the client can continue deriving value from the solution without having to wipe it clean and start from scratch.
What about articulating ROI in terms that are easy to understand?
A lot of the primary takeaways in the data science domain are actually not very complicated. They can be translated into simple ideas—it’s just that the folks who usually do that translation are used to speaking in technically loaded terms that unintentionally create a barrier between the technical and business stakeholders.
One of the things I like to do early in a project is to talk executives through data science concepts and help them understand their relevance to the project we are working on. For instance, it’s almost always useful to discuss how, exactly, the organization wants to assess the performance of the model that we build. Do they simply care about accuracy—that is, how often it’s correct? Or, do they care more about accuracy under a constrained set of circumstances? Is failing under circumstance A more or less important than failing under circumstance B? And can we tie the difference back to dollar values in the organization—or hours saved or lives saved or whatever else that organization cares about?
Tying concepts that, frankly, most executives are already familiar with back to some of the technical terms in data science gives them additional comfort when they’re sitting with the technical team and having those conversations.
How else can companies facilitate better communication between technical and business teams?
I cannot even begin to tell you how many times I’ve seen organizations pay a premium to bring on a group of data scientists, then ignore the ecosystem that’s necessary to make sure they’re successful. Data science is like a river—it has a lot of force, but if you don’t channel it the right way, you won’t be able to spin a turbine and create electricity.
That gets especially problematic when the folks on the data science side don’t have a continuous feedback loop, because then they start to build things that don’t really solve the problems of the business. They might be able to use data sets to make predictions, but they can’t tie those predictions back into metrics that the business cares about. Good data scientists always keep business objectives in mind. On the executive side, of course, you’ve also got to be clear about what you want to see data science accomplish.
Yet not all business leaders understand how to use data science to drive strategic initiatives within an organization.
There’s nothing wrong with that. In fact, I think there’s a wisdom in organizations that say, “We have a lot of data. We don’t exactly know how we can use it to deliver value. We’d like to get somebody in here who’s an expert and who could tell us how.”
As opposed to hiring data scientists and hoping that the conversation happens naturally.
Exactly. Because it might be the case that the organization only needs a team of 10 people, not 100, in order to pull the most value from their data. Or even just a team of two. As with most things in life, doing a little bit of due diligence up front can really help you optimize resources.
About the AuthorMore Content by Leah Hoffmann