QuantumBlack’s Hybrid Approach to Operations Research in the Age of AI

Modern operations research (O.R.) sits at a critical juncture. Traditional optimization methods that have defined the field for decades are now being transformed through integration with artificial intelligence (AI), creating new possibilities – and new expectations – for solving complex business problems. At QuantumBlack, AI by McKinsey, we’re pioneering this integration with our “hybrid intelligence” approach, combining the power of advanced AI with human expertise to revolutionize O.R. applications across industries.

From Formula 1 to Enterprise Operations: Our Data-Driven Heritage

QuantumBlack was born in the data-intensive world of Formula 1 racing, where milliseconds separate victory from defeat and teams live and breathe data. This heritage of extreme performance optimization in complex, constrained environments has also played a significant role in shaping our approach to operations research. As the AI arm of management consultancy McKinsey & Company, we’ve built on this foundation by bringing together the precision of quantitative techniques with the strategic expertise from our consulting functions to help organizations solve their most challenging operational problems.

At QuantumBlack, we bring together the foresight and precision of data and technology with the creativity and understanding of people. The result is hybrid intelligence – a source of competitive advantage that transforms how companies think, operate and disrupt.

The Evolution of O.R. in the AI Era

Traditional operations research has long relied on mathematical programming, simulation and other analytical techniques to optimize complex systems. These methodologies excel at finding optimal solutions within defined constraints but have been limited by computational capacity and often struggle with highly dynamic, uncertain environments.

AI is changing this paradigm in fundamental ways. Machine learning (ML) algorithms can now detect patterns in vast datasets that would be invisible to traditional O.R. methods. They can adapt to changing conditions without explicit reprogramming and incorporate a broader range of inputs, including unstructured data such as text, images and sensor readings.

However, AI alone isn’t the answer. Many systems operate as “black boxes,” making decisions that are difficult to explain or validate. They may also fail to effectively incorporate domain-specific constraints or business objectives. This is where our hybrid approach shines, combining the interpretability and rigor of traditional O.R. with the adaptability and pattern recognition of AI.

OptimusAI: Reimagining Industrial Operations

One of our flagship products, OptimusAI, exemplifies this hybrid approach. OptimusAI is an industry-leading platform developed by QuantumBlack Labs that transforms plant productivity and decision-making in materials, energy and food production by integrating traditional optimization techniques with cutting-edge machine learning.

A recent deployment with a major copper concentrate producer illustrates the power of this approach. The company struggled with complex operations involving thousands of sensors feeding into control systems that had to solve for nonlinear functions and multiple trade-offs. Traditional optimization approaches had reached their limits.

The QuantumBlack team deployed OptimusAI to build an AI model that could use operating data to identify optimization opportunities for boosting copper production while maintaining quality. The system developed a proprietary control-room advisor application that presented straightforward optimization recommendations to operators despite working with thousands of variables behind the scenes.

The results were remarkable: The plant improved throughput by 10-15% and copper recovery by 2-4 percentage points. This represents millions in additional annual revenue with minimal capital investment – a testament to the value of modernizing O.R. with AI.

Merging O.R. and ML: Our Methodological Innovations

There’s a tremendous amount of value and efficiency to be found in systematically bridging the methodological gap between traditional O.R. and modern machine learning. Here are some key aspects of the integrated approach we’ve taken.

Constraint-Aware Machine Learning

Traditional ML algorithms often struggle with the hard constraints common in O.R. problems. We’ve developed specialized architectures that can incorporate operational constraints directly into the learning process. For example, in supply chain optimization, our models can learn efficient inventory policies while respecting warehouse capacity constraints, service level agreements and other business rules.

Reinforcement Learning for Dynamic Optimization

Many operational decisions involve sequential decision-making under uncertainty – a perfect application for reinforcement learning (RL). We’re using RL alongside traditional dynamic programming to solve complex problems in production scheduling and logistics routing. By rewarding algorithms for decisions that optimize long-term objectives rather than immediate gains, we’re achieving solutions that traditional optimization methods would struggle to discover.

These models can adapt to changing conditions in real time, continuously learning from new data to improve their recommendations. This is particularly valuable in volatile environments in which traditional static optimization models can quickly become outdated.

Explainable AI for Operations

One of the biggest challenges with AI-driven decision-making is transparency. Operations managers need to understand and trust the recommendations they receive. To address this need, we’ve prioritized the development of techniques for generating interpretable insights from complex models, making the black box transparent.

For example, in one manufacturing optimization project, we created intuitive visualizations showing how different production parameters contributed to quality outcomes. This helped operators understand not only what to adjust but why those adjustments mattered, which builds trust and facilitates adoption.

Applications in Industry

The true measure of this kind of hybrid approach is the tangible impact it delivers across diverse industries. The following are some examples of how we’re transforming O.R. in practice at QuantumBlack.

Process Manufacturing

In industries such as mining, chemicals and energy, traditional approaches focus on steady-state optimization, but real plants operate in highly dynamic environments with frequent disruptions.

By integrating reinforcement learning and digital twin technology with classical optimization, we’ve been able to create solutions that can respond to changing conditions and recover optimally from disruptions. This approach helps optimize performance in the context of thousands of variables, allowing plants to improve throughput while also becoming more energy efficient – two objectives that are sometimes seen as mutually exclusive.

Healthcare

The healthcare sector presents unique optimization challenges, with human lives at stake and highly complex constraints. For one global pharmaceutical company, we transformed their clinical trial operations using a combination of traditional optimization techniques and machine learning.

We developed models to optimize country footprint and site selection, incorporating predictive elements for enrollment periods and risk factors. The result was a more than 10% increase in clinical trial productivity, accelerating the time to market for critical medications while maintaining rigorous quality standards.

Retail

We can help companies move beyond siloed optimization to integrated supply chain intelligence. For major retailers, we combine traditional inventory optimization with demand forecasting using neural networks and natural language processing to incorporate unstructured data such as social media trends and weather forecasts.

This approach allows for more accurate predictions and optimal stocking decisions, significantly reducing out-of-stock instances while simultaneously optimizing inventory levels and reducing costs.

Building Capabilities for the Future

We recognize that implementing advanced AI/O.R. hybrid solutions often requires organizations to develop new capabilities. This is why our approach includes a specific focus on knowledge transfer and capability building within client and partner organizations. One advantage of being an AI agency within a larger consultancy is that we can assemble multidisciplinary teams that combine various technical disciplines with industry-specific expertise. These kinds of teams are well suited to building capacity, designing training programs and supporting organizational shift with a wide range of clients.

We’ve also developed our own open-source tools to support clients, partners and the wider community in building AI-enhanced O.R. solutions. Kedro, our open-source Python framework for building modular data science pipelines, has been donated to the Linux Foundation to help advance the field more broadly.

As we look ahead, we see O.R. entering a new age of accelerated improvement, powered by the combination of traditional optimization techniques with modern AI capabilities. This hybrid approach will unlock previously impossible solutions to complex operational challenges across industries, and QuantumBlack is committed to leading this transformation. By bringing together the rigor and interpretability of traditional O.R. with the adaptability and pattern recognition of AI, we’re helping organizations navigate complexity and make better decisions in an increasingly dynamic world.

The next frontier will involve even deeper integration of these disciplines, with AI not only solving O.R. problems but also helping to frame them more effectively. As more advanced techniques like reinforcement learning and graph neural networks mature, we expect to see O.R. applications that can autonomously identify optimization opportunities and continuously adapt to changing conditions.

For O.R. professionals, this represents both a challenge and an opportunity. Those who embrace AI as a complement to their traditional toolset will be positioned to deliver unprecedented value in the years ahead. At QuantumBlack, we’re excited to be at the forefront of this evolution, reimagining what’s possible when human expertise, data science and artificial intelligence come together.

Editor's note. The INFORMS Roundtable is the recognized premier destination for top-level leaders in OR/MS practice and delivers unique benefits to member organizations, their representatives and the broader INFORMS community.