931: Boost Your Profits with Mathematical Optimization, feat. Jerry Yurchisin
🎯 Summary
Podcast Episode Summary: 931: Boost Your Profits with Mathematical Optimization, feat. Jerry Yurchisin
This episode of the Super Data Science Podcast features Jerry Yurchisin, Senior Data Science Strategist at Gurobi, focusing on the power and application of Mathematical Optimization (MO) as a crucial, yet often overlooked, tool for data scientists alongside traditional Machine Learning (ML) and statistical methods.
The main narrative arc established that while ML excels at prediction (forecasting the future), MO excels at prescription (determining the optimal actions to take now). The discussion demystified MO by breaking it down into its core components: defining decisions/levers, establishing constraints/business rules, and defining an objective function (e.g., maximize profit, minimize cost). The output is the precise set of actions that yields the best possible outcome under given limitations.
Jerry emphasized that the complexity of real-world business problems—involving thousands of variables and constraints (like supply chain design or production scheduling)—makes manual or purely predictive solutions inadequate, highlighting the need for algorithmic optimization.
1. Focus Area
The primary focus is Mathematical Optimization (MO), contrasting its prescriptive nature with the predictive nature of Machine Learning and statistics. Specific applications discussed include supply chain network design and resource allocation. A secondary focus was the emerging integration of MO with modern hardware, specifically GPUs, for solving large-scale Linear Programs (LPs).
2. Key Technical Insights
- Prescriptive vs. Predictive: MO provides the optimal decision (“What should I do?”) based on known constraints and objectives, whereas ML/Statistics primarily provide predictions (“What will happen?”).
- GPU Integration Challenge: Traditional MO algorithms rely on “pivoting,” which is poorly parallelizable and inefficient on GPUs. However, new algorithms are emerging that show significant speedups for super large Linear Programs (LPs) on GPU hardware, a frontier Gurobi is actively exploring in partnership with Nvidia.
- Mathematical Twin: MO models serve as a “mathematical twin” of a real-world system, allowing for precise, mathematically guaranteed optimal solutions that are impossible to find via intuition alone, as demonstrated by the Burrito Optimization Game.
3. Business/Investment Angle
- Profitability through Precision: Even small percentage improvements achieved through optimization (e.g., shaving 2% off fuel costs for a massive company) translate into massive financial savings, making MO a direct driver of profitability.
- B2B Dominance: Gurobi is a massive, yet often invisible, B2B enterprise player, trusted by a vast majority of the top US companies to solve their most critical operational problems—a competitive advantage they often keep secret.
- Accessibility for Learning: Despite its enterprise scale, MO is highly accessible for individual learning through free, small-scale licenses and extensive open-source educational resources provided by Gurobi.
4. Notable Companies/People
- Jerry Yurchisin: Senior Data Science Strategist at Gurobi, expert evangelist for MO, and creator of numerous free learning resources.
- Gurobi: The leading commercial provider of mathematical optimization solvers, used by the vast majority of Fortune 100 companies.
- Nvidia: Mentioned due to their recent release of the open-source GPU-focused optimization library, KuOpt, signaling industry movement toward leveraging GPU power for optimization problems.
- Dr. Joel Sokol (Georgia Tech): Collaborator on Gurobi’s Coursera course on optimization.
5. Future Implications
The industry is moving toward leveraging massive computational power (GPUs) to tackle optimization problems that were previously too large or complex, even for high-end CPUs. This suggests a future where prescriptive analytics becomes faster and applicable to even larger, more dynamic enterprise systems. Furthermore, the emphasis on free learning resources indicates a push to embed MO skills within the broader data science and AI community.
6. Target Audience
This episode is most valuable for Data Scientists, AI Engineers, Operations Researchers, and Business Analysts who are looking to expand their toolkit beyond predictive modeling to solve complex decision-making problems and drive measurable business impact.
🏢 Companies Mentioned
đź’¬ Key Insights
"if there are kinds of problems you could be solving in your organization today that you think could only be solved in the future with quantum computing, yeah, that might not be true. Mathematical optimization could be the answer."
"understanding your business case, understanding all the problems that you can approach with this, all the ones you can't, understanding all the stakeholders, their involvement, getting buy-in, getting all the data connected... All that stuff needs to happen regardless of if you're using mathematical optimization or quantum."
"You could revolutionize your supply chain now with mathematical optimization."
"Wouldn't you like to save a ton of money or be super efficient now instead of waiting possibly for the possibility of this happening in the future?"
"The Total Wine folks, they were a team of data scientists, people who did not have a traditional operations research background... Machine learning is not cutting it. What else can we do? Oh, okay, I've heard of mathematical optimization."
"But they had a tool that was optimization in the back, had an LLM interface where the planners can really interact with this and say, 'Okay, well, what if tariffs are this, or what if my supply of this thing was cut in half?'"