Inventory Optimization in Python — Core Concepts

Inventory optimization sits at the heart of supply chain management. The goal is to minimize the total cost of holding inventory while maintaining a target service level — the percentage of customer orders fulfilled immediately from stock.

The cost tradeoff

Every inventory decision balances three costs:

• Holding cost — warehouse space, insurance, spoilage, and tied-up capital. Typically 20-30 percent of product value per year.
• Ordering cost — placing a purchase order, receiving goods, quality inspection. Each order has a fixed overhead regardless of quantity.
• Stockout cost — lost sales, expedited shipping to recover, and damaged customer trust. Often the hardest to quantify but the most expensive.

The optimal order quantity minimizes the sum of these three.

Economic Order Quantity (EOQ)

The classic EOQ formula finds the order size that minimizes holding plus ordering cost:

EOQ = √(2DS / H)

Where D is annual demand, S is cost per order, and H is annual holding cost per unit. Python makes this trivial:

import math

def eoq(demand, order_cost, holding_cost):
    return math.sqrt(2 * demand * order_cost / holding_cost)

EOQ assumes constant demand and instant delivery — unrealistic but useful as a starting point. Real systems layer variability on top.

Safety stock and reorder points

Demand fluctuates. Lead times vary. Safety stock is the buffer that absorbs these surprises. The reorder point tells you when to place a new order:

Reorder Point = (Average Daily Demand × Lead Time) + Safety Stock

Safety stock depends on how much variability you expect and how rarely you want to run out. A 95 percent service level means you accept stockouts 5 percent of the time. Higher targets require exponentially more safety stock.

ABC analysis

Not all products deserve the same attention. ABC analysis groups items by revenue contribution:

• A items (top 20% of SKUs, ~80% of revenue) — tight control, frequent review, low safety stock tolerance.
• B items (next 30%, ~15% of revenue) — moderate oversight.
• C items (bottom 50%, ~5% of revenue) — simple rules, higher safety stock acceptable because the cost of attention exceeds the cost of extra inventory.

Python’s pandas makes ABC classification straightforward: sort by annual revenue, compute cumulative percentage, and assign categories.

Demand patterns matter

Inventory models assume a demand distribution. Three common patterns:

• Steady demand — bottled water in an office. Normal distribution works well.
• Seasonal demand — sunscreen peaks in summer. Requires time-series decomposition before setting reorder points.
• Intermittent demand — spare parts for rare machines. Croston’s method or bootstrapping outperform normal-distribution assumptions.

Choosing the wrong demand model leads to either chronic overstocking or frequent stockouts.

Python tooling

• pandas — data wrangling, ABC classification, demand aggregation.
• scipy.stats — distribution fitting for demand variability.
• statsmodels — time-series forecasting to feed demand estimates into inventory formulas.
• PuLP or scipy.optimize — for multi-SKU optimization where budget constraints link products together.

A common misconception

Many teams treat inventory optimization as a one-time calculation. In practice, demand shifts, suppliers change lead times, and costs fluctuate. The system needs to re-run at least weekly, feeding updated sales data into the models and adjusting order quantities.

The one thing to remember: Inventory optimization balances holding, ordering, and stockout costs using demand forecasts and safety stock buffers — and Python automates the math that makes this feasible across thousands of products.