Inventory Optimization MILP

The $34 Million Question

Boston Scientific had a problem that sounds familiar to anyone who's managed inventory at scale: too much of the wrong stuff in the wrong places.

Picture this: $34 million worth of medical devices sitting in warehouses. Some locations drowning in excess inventory destined for expiration. Others desperately short, creating backorders and lost sales. Meanwhile, BSC's rigid distribution network only allowed products to flow in one direction—from manufacturing to Tier 1 DCs to Tier 2 DCs—like a highway with no exit ramps.

When Covid hit, demand patterns shifted unpredictably. The imbalances got worse. BSC found itself in the absurd position of scrapping expired inventory in one location while frantically trying to source the same products for another.

This wasn't just an operational inefficiency—it was a strategic blind spot.

The Invisible Network

The solution wasn't obvious because the problem wasn't obvious. BSC's planners could see the imbalances, but they couldn't see the connections.

Traditional supply chain thinking treats each distribution center as an independent node. You plan inventory levels, you hit your targets, you move on. But what if inventory could talk to inventory? What if excess stock in Europe could cover deficits in Asia?

This concept—lateral transshipment—challenges a fundamental assumption in supply chain design: that products must always flow "down" the network hierarchy.

Instead, imagine a network where products can flow horizontally between peer locations. Where excess becomes opportunity rather than waste. Where global imbalances become local optimization chances.

The Math of Rebalancing

Here's where most companies stop: The concept sounds good, but the execution feels impossible. How do you calculate optimal transshipments across thousands of SKUs and dozens of locations? How do you balance transportation costs against inventory holding costs against lost sales costs?

We built a Mixed Integer Linear Programming (MILP) model that treats this as an optimization problem, not a planning problem.

The model considers:

• Transportation costs for lateral movements (estimated using tools like UPS calculators)
• Holding costs for excess inventory (potential scrap value)
• Deficit costs (backorders and lost sales)
• Current inventory positions across all locations
• Demand forecasts and safety stock requirements

The objective: Find the minimum-cost set of transshipments that reduces overall inventory imbalance.

The SKU Selection Challenge

BSC manages over 16,000 finished goods SKUs. Optimizing all of them simultaneously would require computational resources that don't exist and time that won't wait.

We developed two approaches to identify high-impact SKUs:

ABC Classification: Start with the revenue leaders. The top 1,000 SKUs by revenue represent the majority of inventory value and the biggest potential savings.

Potential Balance Opportunity (PBO) Heuristic: A novel approach we created for this project. For each SKU, calculate the minimum value between total excess and total deficit across all DCs. SKUs with high PBO scores offer the greatest rebalancing potential.

The PBO heuristic turned out to be surprisingly powerful—it identified SKUs where small transshipments could create large imbalance reductions.

The Results: When Theory Meets Reality

The MILP optimization delivered results that exceeded our expectations:

ABC Classification Model: 10.2% reduction in total inventory costs. From $34.6M to $31.1M. This involved 1,666 transshipments costing $2.6M, but generated $3.3M in excess reduction and $2.8M in deficit reduction.

PBO Heuristic Model: 25% reduction in total costs. From $87M to $65M. Higher transshipment costs ($7.3M for 3,681 transshipments), but massive savings from reducing excess ($15.2M) and deficit ($13.9M) inventory.

But here's what really mattered: Monte Carlo simulations showed that the rebalanced inventory maintained superior performance even under stochastic demand conditions. The solution was robust, not just optimal.

The Geography of Opportunity

The optimization revealed natural transshipment corridors. The primary flows occurred between major regional hubs: Europe to Asia, North America to Latin America.

This wasn't random—it reflected the reality of global demand patterns and transportation infrastructure. The model found that certain regional imbalances were systematic, not cyclical. Europe consistently generated excess in certain product categories while Asia showed consistent deficits.

These patterns suggested that lateral transshipment wasn't just a reactive tool for fixing imbalances—it could be a proactive strategy for network optimization.

Implementation: From Analysis to Action

The gap between analysis and implementation is where most optimization projects die. Great models that never get deployed are just expensive academic exercises.

We designed a phased approach:

Phase 1: Pilot with ABC-classified SKUs. Leverage BSC's existing internal metrics and processes. Prove the concept with familiar categories.

Phase 2: Scale to PBO-selected SKUs. Introduce the new heuristic after demonstrating success with traditional approaches.

Phase 3: Integrate into periodic inventory review processes. Embed transshipment analysis into existing ERP workflows rather than creating parallel systems.

The key insight: Don't replace existing processes—enhance them. BSC's planners already reviewed inventory levels regularly. We gave them a new tool to optimize those reviews.

The Hidden Costs and Real Constraints

Our models had limitations that matter in practice:

We didn't include intra-DC handling costs—the picking, packing, and receiving expenses associated with transshipments. These costs are real and could affect the optimization.

We didn't model the complexity of global medical device regulations. Some transshipments that look optimal mathematically might be impossible legally.

We assumed point-to-point transportation costs, but real logistics networks offer consolidation opportunities that could improve the economics further.

These limitations don't invalidate the results—they highlight opportunities for refinement and partnership development.

The Compound Effect of Network Intelligence

This project demonstrates something profound about supply chain management: Local optimization often creates global suboptimization, but global visibility enables local intelligence.

BSC's individual DC managers were making rational decisions based on local information. But those decisions, aggregated across the network, created systematic imbalances.

Lateral transshipment isn't just about moving inventory—it's about creating network-level intelligence that enables better local decisions.

When DCs can see global inventory positions and transportation costs, they make different trade-offs. When planners have tools that calculate optimal rebalancing automatically, they focus on strategic rather than tactical decisions.

The Future of Dynamic Networks

The success of this project points toward a fundamental shift in supply chain thinking: From static networks to dynamic networks. From hierarchical flows to networked flows. From local optimization to system-wide optimization.

The technology exists. The modeling techniques are proven. The economic benefits are substantial.

What's needed is organizational commitment to network thinking over functional thinking. To system optimization over local optimization. To short-term complexity in service of long-term simplicity.

The Strategic Lesson

Inventory rebalancing through lateral transshipment delivered 10-25% cost reductions for BSC. But the real value wasn't in the savings—it was in the strategic capability.

BSC now has a methodology for turning imbalances into opportunities. For treating global inventory as a fungible asset rather than a collection of local assets. For responding to demand volatility with flexibility rather than buffer stock.

This capability becomes particularly valuable in uncertain environments. When demand patterns shift—as they did during Covid—companies with dynamic networks adapt faster than companies with static networks.

The question isn't whether lateral transshipment makes sense. The question is whether your supply chain can adapt to the new reality of permanent volatility.

And adaptation, like optimization, starts with seeing the network as a system rather than a collection of parts.