By Justin Kendig, Principal Product Manager, Supply Chain, C3 AI and Jianchen Yang, Lead Data Scientist, Supply Chain, C3 AI
Two of the most important decisions in inventory management are how much inventory to stock and where to stock it. They are connected, but they are not the same kind of problem, and treating them as one often leads to unnecessary complexity, rigidity, and slow response times. Choosing how much to stock is fundamentally a data problem: forecasting demand, modeling uncertainty, and calculating the right reorder parameters. Choosing where to stock it — that is, how inventory should be allocated across an existing network of plants, distribution centers, and warehouses — is a physical problem, bounded by warehouse space, labor, and the logistics of moving goods.
In practice, planning organizations usually take one of two paths. Some bundle both decisions into a single multi-echelon inventory optimization (MEIO) model that solves quantity and placement across the whole network at once. This provides some genuine coordination, but with compromises: simplified demand and lead-time assumptions, slower runtimes at network scale, one shared planning cadence for two decisions that move at different speeds, and no way to update one answer without rerunning the other. Other approaches optimize each warehouse in isolation, which establishes a fast timeline but duplicates safety stock across sites and gives up the demand-pooling benefits of a network view. The stakes of either path are real: excess working capital, slower responses to disruption, and recommendations planners struggle to explain or own.
Because the two decisions often move at different speeds, belong to different teams, and respond better to different tools, they are best solved as separate but coordinated problems. Splitting them allows quantity models to update safety stock, reorder points, and other inventory targets as often as demand and supply conditions require, while location models run on the longer cadence appropriate for network, capacity, and service decisions.
The separation creates clearly defined objectives, inputs, ownership, and update cadences for each model (making AI recommendations easier to train, govern, and trust), along with faster responses to disruption and a more deliberate split of expertise across planning teams. This blog analyzes the value of separating inventory quantity from inventory location in supply chain planning, and explains how C3 AI Inventory Optimization, our AI-powered supply chain application, was architected around these separatist principles.
Figure 1: By decoupling the question of how much inventory to hold from where to position it, organizations can pursue a fundamentally smarter planning strategy. This separation simplifies day-to-day operations, creates cleaner data inputs for AI and advanced analytics, improves responsiveness and scalability as conditions change, builds a more agile and resilient supply chain, and allows different roles to contribute the expertise most relevant to their domain.
Operational Simplicity: Updating Systems vs. Rearranging Space
One of the strongest reasons to decouple how much from where comes down to a simple principle: the two decisions are nowhere near equally easy to execute. Adjusting how much inventory to stock is largely a digital task. Most companies use material requirements planning (MRP) or enterprise resource planning (ERP) systems that allow for quick updates to reorder points, safety stock levels, and lead times. These changes can be implemented in minutes and tested over time with minimal disruption.
In contrast, deciding where to stock inventory is a complex physical puzzle that involves real-world constraints such as warehouse space, racking configurations, labor availability, and safety regulations. Reallocating inventory locations often requires coordination across departments, physical movement of goods, and sometimes infrastructure changes. This makes location decisions far more rigid and costly to adjust. Separating the two lets businesses remain agile in their stocking levels while making more deliberate, long-term decisions about physical storage.
Quantity Decisions Can Move Faster Than Location Decisions
When demand spikes or supply disruptions hit, companies need to react in hours. If inventory quantity and location are tightly coupled, a change in one area triggers a cascade of adjustments in the other, slowing the response. Decoupling the two lets businesses scale inventory levels independently of physical layout. A sudden jump in demand can be absorbed by adjusting reorder points, with no warehouse reconfiguration required. This flexibility matters in industries with volatile demand or frequent product introductions, such as consumer electronics, fast fashion, and automotive parts.
The two decisions also run on different clocks. How much to stock is a fast-moving decision, best revisited weekly or even daily as demand, supply, and promotions shift. Where to stock is a slower and more strategic decision, typically revisited quarterly, yearly, or when something major changes, such as a new product launch or a new distribution center. Couple them, and the fast decision gets held hostage to the slow one’s review cycle — or the slow one churns every time demand moves. Separate them, and each can run at its natural pace. That independence also makes room for modern distributed fulfillment models such as micro-fulfillment centers or drop-shipping, where inventory sits in many small locations closer to the customer.
Quantity and Location Belong to Different Teams
The quantity versus location question can be divided based on the scope of different roles involved in inventory planning. Demand planners or inventory planners determine how much inventory to stock. Their focus is on forecasting, replenishment strategies, and maintaining service levels. Logistics managers or warehouse operations teams decide where inventory is physically stored. They manage space utilization, labor efficiency, and transportation logistics.
Delineating these responsibilities lets each team focus on its area of expertise, use the most relevant tools designed for their workflows, and make faster, better-informed decisions. It also reduces cross-functional friction and improves accountability. When the two decisions are tangled together, ownership blurs. Inventory planners get pulled into warehouse constraints they don’t control, while logistics teams end up second-guessing stocking math that isn’t theirs to set. Clear boundaries remove that overlap: each team owns the decision it has influence over, so handoffs stay clean and disagreements have an obvious owner, rather than becoming cross-departmental standoffs.
Specialized Models Outperform a Single Model Solving Both Problems
Traditional multi-echelon inventory optimization (MEIO) models, which solve stock levels across multiple stages (echelons) of a supply chain network at once, attempt to answer both the how much and where questions simultaneously. These models often rely on simplifying assumptions about demand variability, lead time distributions, and service levels, which can limit the accuracy of their recommendations.
Figure 2: Real-life supply chain data rarely follows standard distribution patterns. Supply time uncertainty often skews toward late arrivals, order quantity variance can cluster around two (or more) distinct peaks rather than one, and demand fluctuations may appear sporadically with no predictable central tendency, all of which challenge planning models that assume normality.
Separating the decisions of how much versus where allows for more specialized and accurate modeling. C3 AI Inventory Optimization, our AI-powered application designed to right-size inventory and improve service levels, helps inventory planning teams answer both questions. To answer how much, the application employs AI-driven stochastic optimization models that incorporate real network uncertainties like demand, lead time, quantity, other inventory movements (financial, quality, project-specific), and seasonality. The underlying modeling approach in the how much lane is single-echelon stochastic optimization: a simulation-driven process that models how inventory levels evolve over time, then applies optimization to determine the reorder parameters (safety stock, reorder point, min-max levels) for each product or location, balancing cost against service levels under uncertainty.
For the question of where, C3 AI Inventory Optimization uses several AI-based approaches that recommend safety-stock placement while accounting for transportation costs, service level agreements, and facility constraints. These include:
- A guaranteed-service model (GSM) finds the optimal placement of safety stock in a multi-echelon inventory system under bounded demand and guaranteed service time while minimizing cost. The service time can be configured by the user or estimated from demand signals such as sales orders and transfer orders. GSM has been widely studied in academia and recognized in industry, forming the foundation of many advanced supply chain tools by balancing service and cost across multi-echelon and large-scale networks.
- Stochastic multi-echelon inventory optimization with deep neural networks (SMEIO DNN) jointly simulates inventory across multiple SKUs and trains deep neural networks to learn the mapping from demand and lead-time distributions to optimal replenishment policies. Adoption is growing in both commercial platforms and open-source research because the approach is better suited to handle time-series inputs, and it models more sources of uncertainty than stochastic demand alone.
- Joint inventory simulation with particle swarm optimization (MEIO PSO) simulates inventory across the bill of materials or supply chain network: populating demand from downstream to upstream, then running MRP planning, purchasing, producing, shipping, and using gradient-free optimizers to evaluate and refine candidate quantity-buffer solutions. Because the approach takes time-varying inputs and closely simulates the real network, it provides high transparency and explainability. The simulation logic and computation cost can both grow large, but careful validation and distributed compute mitigate both.
These modular approaches let C3 AI Inventory Optimization apply the best-fit tool to each problem rather than forcing a one-size-fits-all model to handle both. Drawing on decades of supply chain experience, we built this decoupled approach into the architecture of the application to maintain separation in the workflows around how much and where. Rather than forcing one model to compromise across both questions, it pairs each decision with the method best suited to it: AI-driven stochastic optimization for the data problem of how much, and specialized placement models for the physical problem of where. That separation is what keeps recommendations accurate as conditions shift, keeps the underlying data clean, and lets planners act on one decision without waiting on the other.
Inventory Optimization in Practice at a Global Welding Equipment Manufacturer
A world leader in welding and cutting equipment, with over 9,000 employees, more than 40 brands, and 20 manufacturing plants across 140-plus countries, carrying roughly $470 million in global inventory, set out to cut excess stock without hurting availability. Its existing “Plan for Every Part” process sorted each SKU-location pair into one of nine replenishment buckets, but it ran on manual data pulls across spreadsheets and multiple systems, so inventory levels were refreshed every six months at best and drifted out of sync across facilities. The company turned to C3 AI Inventory Optimization to unify that data and generate optimal safety-stock and reorder-point levels automatically.
This is the “how much” question treated as a dedicated data problem, exactly the separation this blog argues for. C3 AI Inventory Optimization pulled twelve ERP data sources into a single federated image and ran stochastic optimization (more than 25 million simulations across roughly 2,000 fast-moving SKU-location pairs) to right-size safety stock and reorder points against real demand and supply variability. In a 12-week engagement across eight North American facilities, it identified a 26% inventory-reduction opportunity while improving material availability by 0.5%, with recommendations flowing straight back to the ERP and refreshable as often as daily. Because the quantity decision was optimized on its own fast cadence with no warehouse reconfiguration required, the company could carry less while holding service levels steady. Through the deployment of C3 AI Inventory Optimization across the North American facilities, the company identified a potential inventory reduction value of $10 million.
What This Means for Supply Chain Planners
Treating how much inventory to stock and where to stock it as a single problem to solve feels intuitive — it is, after all, one inventory. But it forces a single model, a single cadence, and a single team to shoulder two fundamentally different jobs. Pulling them apart is what makes each one simpler. The quantity decision can be re-optimized digitally and often, on its own fast cadence, while the location decision gets the slower, more deliberate attention it needs. Each lands with the team that owns it, and each is handled by a model built for its nature: stochastic optimization for the data problem of how much, specialized placement models for the physical problem of where. This is the thinking built into C3 AI Inventory Optimization, and the payoff is a supply chain that responds more quickly, carries less, and holds service levels steady — exactly what the welding equipment manufacturer saw when it decoupled the two. Treat the two as the distinct problems they are, and the whole supply chain gets faster, leaner, and more resilient.
Learn more about C3 AI Inventory Optimization, and contact us to schedule a demo and see how it applies to your supply chain.
Read about C3 AI’s research and advances in AI inventory management.
--
About the Authors
Justin Kendig is Principal Product Manager, Supply Chain at C3 AI and is responsible for product development of supply chain applications. Prior to joining C3 AI, his professional experience included supply chain and procurement in the manufacturing and retail industries. Justin earned a Bachelor of Science in Mechanical Engineering from Michigan Tech University, Master of Science in Mechanical Engineering from Bradley University, and a Master of Science in Predictive Analytics from Northwestern University.
Jianchen Yang is a Lead Data Scientist at C3 AI specializing in developing and deploying machine learning and optimization models for supply chain and inventory management. With a strong background in simulation and predictive modeling, he has contributed to multiple customer deployments of C3 AI Inventory Optimization, ensuring tailored solutions for diverse business needs. Jianchen received his Ph.D. in Biomedical Engineering from the University of Texas at Austin.