The Technology Behind Modeling at Scale

Modeling supply chains at scale is a major technological challenge - one that surpasses traditional optimization methods. As supply chain complexity continues to rise, so does the difficulty of finding the best solution. Did you know that a redesign selecting the 45 best locations from 50 options, results in 2.1 million possible combinations to sift through. What happens when we expand that to more like 90 locations out of 100? That skyrockets the combinations to 17 trillion! At this scale, conventional approaches simply can't keep up.

Early network design tools were constrained by desktop computing. Cloud-based solutions improved accessibility, but most still rely on traditional modeling techniques. Lyric is breaking this mold, enabling enterprises to move from incremental improvements to full-scale network reimagination. Our platform is built to handle massive datasets, integrate multiple optimization layers, and orchestrate end-to-end decision-making—turning supply chain modeling into a strategic, AI-driven advantage.

"From desktop models to AI-driven intelligence—supply chain modeling has evolved, but many solutions still rely on conventional techniques."

In this blog, we’ll explore how network modeling technology has evolved, the challenges of modeling at scale, and how Lyric’s approach is redefining what’s possible.

Understanding the Fundamentals

The core of supply chain modeling is the "network model" - an abstraction capturing all network structure nuances: suppliers, plants, production lines, warehouses, customers, and associated costs (production, warehousing, transportation). These models excel at abstracting network complexity and identifying the right network configuration for the business.

Technically speaking, network models are linear programs (LPs) or mixed-integer programs (MIPs). Understanding this mathematical foundation helps reveal the inner workings of the technology itself. At a basic level, LPs optimize business goals like cost reduction while respecting constraints around capacity, demand, and transport limitations. MIPs add complexity by introducing whole number decisions - like facility count or binary choices around opening locations.

The Challenge of MIP Problems

What makes MIPs particularly difficult is that they are classified as NP-hard problems. Simply put, this means that as the problem grows—say, by adding more facilities or decision variables—the number of possible solutions grows exponentially. This makes it incredibly challenging to find the best solution, even with advanced algorithms.

"Network models help businesses design efficient supply chains by mapping suppliers, plants, warehouses, and transportation costs. The challenge is making them adaptable and scalable."

Remember, if you're deciding how to optimize 90 locations out of 100, there are more than 17 trillion possible combinations to consider. This exponential growth in complexity is what makes MIPs one of the hardest algorithm categories to solve. It's also why having the right solver is so important. In the early days, solvers like CPLEX were used, but today, Gurobi is the industry standard, known for its speed and ability to handle the enormous complexity of modern supply chains, especially in solving NP-hard MIP problems.

The Role of Software in Network Optimization

Network modeling software performs three key functions:

1. Constructs complex mathematical models

2. Interfaces with solvers to process solution spaces

3. Presents results through intuitive visualizations and reports

The software also enables scenario analysis through "what-if" modeling capabilities, helping teams evaluate multiple potential network configurations.

"Network models help design efficient supply chains by mapping suppliers, plants, warehouses, and transportation costs. The challenge is making them adaptable and scalable."

Beyond Network Optimization

While network optimization is critical, it's far from the only area requiring modelers’ attention. A comprehensive solution must address several additional optimization challenges as well:

• Transportation Optimization: Ensuring efficient final mile delivery requires specialized optimization. Direct Store Delivery, for example, focuses on getting products to retail locations on time at minimal cost. When the network is reorganized, routes must be redrawn - in fact, most savings typically come from redesigning transportation routes.

• Inventory Strategy Optimization: Managing inventory levels is a balancing act between minimizing stockouts and reducing holding costs. This involves optimizing replenishment strategies, safety stock levels, and service levels across the supply chain.

• Policy Simulation: Beyond mathematical models, it's essential to simulate the impact of different policies, including changes in transportation routes, sourcing strategies, and inventory policies. This allows companies to understand decision impacts before implementation.

Each area requires deep expertise and specialized algorithms. A holistic supply chain optimization strategy must consider all these interconnected pieces.

The Critical Data Management Challenge

Running a large-scale supply chain modeling program isn't just about optimization algorithms and technology; it's equally important to have a robust system to manage and organize the necessary data.

To execute an effective supply chain model, companies need to build and maintain a comprehensive data catalog. This catalog should include essential supply chain data points such as:

• Production/Sourcing Capacity Data: Information about the maximum output a facility/supplier can produce/supply

• Production/Sourcing Capability Data: Detailed specifications about what products can be produced at which locations

• Warehouse Cost and Capacity Data: Data on how much inventory a warehouse can hold and the associated operational costs

• Transportation Cost and Capacity Data: Cost data for different transportation methods and the available capacity for moving goods

Some of this data can be pulled from enterprise systems like ERP, WMS, TMS, etc., but a significant portion often resides in spreadsheets, planning systems, or tribal knowledge held by key individuals within the organization.

The Risks of Poor Data Management

Not having this critical data organized—or lacking a mechanism for continuous validation and updating—can lead to serious challenges in supply chain modeling. These challenges include:

• Slow Execution: Without a structured data catalog, teams spend too much time searching for, cleaning, and verifying data. This delays model execution and decision-making

• Inaccurate Results: Poor or outdated data leads to flawed models, which can result in suboptimal decisions

• Unsuccessful Programs: Supply chain modeling programs often fail to achieve their goals due to inadequate data management, causing friction and inefficiencies throughout the optimization process

To avoid these issues, companies need to establish a centralized, validated, and continuously updated data catalog. This streamlines the modeling process and ensures that the data being used is reliable, current, and comprehensive, enabling more accurate and faster decision-making.

The Lyric Advantage

Over the past three years, the Lyric team has quietly and purposefully built what we believe is the solution to the challenges the industry faces in executing a successful supply chain modeling discipline at scale. The heart of this innovation is a completely new architectural layer between the model and the application, which we call the 'sequence.'

The Sequence Layer: Game-Changing Modeling Tech

Lyric Studio’s sequence layer is designed to give modeling teams the ability to create comprehensive, algorithmic workflows that cover the entire journey—from raw data to data transformations and all the way through to model-ready data. Through this sequence, modelers can daisy chain models (in a no-code environment) to build complex algorithmic workflows tailored to the organization's unique needs. Moreover, this algorithmic workflow can be wrapped in a custom-built application, through a streamlined point and click environment, to meet the specific requirements of various stakeholders throughout the organization.

"Modeling at scale requires more than optimization—it needs a structured approach to data processing, workflow automation, and cross-team collaboration."

Key Advantages of Lyric's Upgraded Architecture:

1. Scalable Data Processing for Massive Datasets: Lyric's architecture is built to handle billions of rows of data, providing immense flexibility for modeling teams. This capability allows teams to start with raw datasets and adjust the granularity of the models as needed. Whether the goal is to build highly detailed models or aggregate data for a broader view, Lyric can handle it all. Moreover, this scalable data processing power enables modelers to train large prediction models for model input data such as production run rates, warehouse processing rates, and more.

2. Collaborative Data Products for Validation: The sequence layer enables cross-team collaboration by creating tailored data products that validate the inputs going into the model. For example, adjacent teams can verify metrics like truck volume utilization or warehouse inventory turns, ensuring that the data feeding into the models is accurate and up-to-date at any point in time.

3. Daisy Chain Models for Complex Use Cases: Perhaps most importantly, Lyric's architecture allows users to daisy chain multiple models together seamlessly. Whether it's a series of network models or a combination of network and transportation models, Lyric's system enables this level of complexity without any extensive coding or technical expertise.

4. Distributed Framework with Specialized Compute Capabilities: Built on a distributed framework, Lyric empowers modelers to assign the appropriate hardware to different parts of the compute process. For example, the system may leverage Nvidia GPUs for accelerated data processing, use CPUs for optimization model runs, and/or tap into Nvidia GPU-accelerated cuOpt for transportation optimization. This flexibility ensures that the models are running in the most efficient environment for the task at hand.

5. Tailored Applications for Stakeholders: The point and click environment allows you to build custom applications that cater to different stakeholders. Whether it's a supply chain manager, a logistics operator, or a senior executive, each can interact with the data and models through an application designed specifically for their needs. This flexibility ensures that the insights from complex models are easily accessible and actionable across the organization.

Looking to the Future

The convergence of next-generation modeling platforms and advanced computing is opening new frontiers in supply chain modeling. In 2025, several key innovations are set to transform the industry:

• Advanced Modeling Capabilities: Lyric's sequence layer and no-code environment will continue to democratize access to complex supply chain optimization

• Accelerated Compute: The combination of GPU-accelerated frameworks and specialized solvers promises to dramatically reduce solution times

• Integrated Solutions: The seamless integration of network design, transportation optimization, and inventory management will enable truly holistic supply chain transformation

As supply chains become more complex, traditional modeling methods have really struggled to keep pace. The scale of today’s challenges requires more than just incremental improvements—it demands a new way of thinking. Lyric’s approach makes it easier to handle massive datasets, connect different optimization layers, and build flexible workflows without any coding. By reimagining how supply chains are modeled, we’re making it possible to tackle larger problems, test more scenarios, and make smarter decisions faster.