I graduated from the University of Pennsylvania, majoring in cognitive science and computer science. I have always been interested in the intersection between humans and technology and how the two can best work together.
I spent more than five years working for a US Department of Defense contractor, supporting organizations, such as the Army, Navy, and NASA. During my time there, I also earned my master’s in computer science at Tufts University, and I became eager to work with private sector clients. McKinsey had then recently acquired QuantumBlack, an advanced analytics firm, and was building a team in Boston, not far from me.
I joined the firm as a senior data scientist three and a half years ago, and I have since assumed the roles of junior principal and principal data scientist.
My role at McKinsey
My team works with clients who are undergoing analytics transformations. They're trying to revolutionize their business by introducing advanced analytics and machine learning (ML) models to automate their daily business operations. More specifically, I focus on solving problems that require sales forecasting and the optimization of marketing strategies to maximize value. We deliver custom solutions, leveraging open-source assets and other available technology, while building the ML models ourselves.
As for my role, I bring clients through the lifecycle of an analytics use case. We start with ideation where we pinpoint high-impact opportunities to introduce ML. Then I guide clients, such as North American retailers and global healthcare companies, through the conception of a minimum viable product (MVP). Finally, we scale the model for transition to the client’s production environment.
About 30% of my work is in code and development, while the other 70% is managing a team of data scientists and machine learning engineers. We take an agile development approach, working in weekly or bi-weekly sprints, and it’s my job to problem solve the analytics approach, manage quality assurance and risks, and ensure we’re hitting the right targets.
Improving the effectiveness of AI and ML with MLOps
Through our engagements, we discovered many pain points that stall organizations’ models from ever reaching production or having long-lasting impact once made. For example, once ML models are in production, failures regularly occur, and issue detection and resolution are slow. Often models are maintained through laborious manual processes, and even well-staffed teams miss issues. In cases where organizations don’t have a dedicated team to monitor models, performance can degrade, posing risks to the business and eventually leading to retirement of the models. We found that organizations were investing in analytics, but they were not getting a long runway of return on their models.
MLOps solves these challenges by helping to ensure models continue to sustain ML-driven impact cost-effectively and at scale. By institutionalizing MLOps tools and best practices, we automate ML model lifecycle steps, including model retraining, versioning, validation, deployment, and monitoring.
However, another big component of what we do focuses on capability building by providing clients with the skills and tools in-house to monitor and manage ML models and follow proper workflows when alerts are triggered. For our clients, the benefit is the assurance that their business-critical models continue to work over a longer period of time.
Read more about the business impact MLOps can have for organizations.
More about Alison
Alison is a leader within McKinsey’s Data Science community. She is committed to fostering a sense of belonging for data scientists and mentoring data scientists to grow their analytics and consulting toolkits. She lives with her husband in Boston, where she was born and raised, and is an avid Dunkin’ Donuts drinker and New England sports fan.