Supply Chain Knowledge Base

Demand Forecasting

A forecast estimates future demand so the business can plan capacity, materials and inventory. Forecast true demand (orders), not just shipments, which understate demand during stockouts.

Components of a demand series

• Level, trend, seasonality and noise (random variation).

Common methods

• Naive and seasonal naive: benchmarks every model should beat.
• Moving average: mean of the last n periods.
• Simple exponential smoothing:

F_t+1 = α × A_t + (1 − α) × F_t, with 0 < α < 1

• Holt-Winters: exponential smoothing with trend and seasonality.
• Regression / machine learning: adds drivers such as price, promotions or economic indicators.

Accuracy and bias metrics

Metric	Formula	Notes
MAE	average of abs(A − F)	In units; easy to interpret
MAPE	average of abs(A − F) / A	Undefined when A = 0; unstable for low volumes
WAPE	Σ abs(A − F) / Σ A	Volume-weighted; good for portfolios
Bias	average of (F − A)	Positive = over-forecast; negative = under-forecast

Worked example: actuals 100, 120, 80; forecasts 110, 110, 100. MAE = 13.3 units; WAPE = 40 / 300 = 13.3%; MAPE = 14.4%; bias = +6.7 units.

Good practice

• Plot the data before modeling; flag outliers and one-time events.
• Forecast at the level where decisions are made; aggregate forecasts tend to be more accurate.
• Track forecast value added: does each manual adjustment improve accuracy?

Back to contents →

Inventory Optimization

Inventory policy balances service against the cost of holding stock. The formulas below are standard textbook models; they assume roughly normal demand and independent demand and lead time.

Key formulas

Reorder point: ROP = d̄ × L + SS

Safety stock (demand variability): SS = z × σ_d × √L

Safety stock (demand and lead-time variability): SS = z × √(L × σ_d² + d̄² × σ_L²)

Economic order quantity: EOQ = √(2DS / H)

Where d̄ = average daily demand; σ_d = standard deviation of daily demand; L = lead time (days); σ_L = standard deviation of lead time (days); z = service factor; D = annual demand (units); S = cost per order; H = annual holding cost per unit.

Service factor (z) for common cycle service levels

Cycle service level	z
90%	1.282
95%	1.645
97.5%	1.960
98%	2.054
99%	2.326

Worked example A: demand variability only

d̄ = 40 units/day, σ_d = 12, L = 9 days, 95% service (z = 1.645).

• SS = 1.645 × 12 × √9 = 59.2 → 60 units (rounded up)
• ROP = 40 × 9 + 60 = 420 units

Worked example B: add lead-time variability

Same item with σ_L = 2 days.

• √(9 × 12² + 40² × 2²) = √(1,296 + 6,400) = √7,696 ≈ 87.7
• SS = 1.645 × 87.7 ≈ 144.3 → 145 units
• ROP = 360 + 145 = 505 units

Worked example C: EOQ

D = 12,000 units/year, S = $75 per order, H = $4 per unit per year.

• EOQ = √(2 × 12,000 × 75 / 4) = √450,000 ≈ 671 units

Practical notes

• Keep time units consistent (daily demand with lead time in days).
• Cycle service level is not the same as fill rate.
• Intermittent or lumpy demand often violates the normal assumption.
• Segment items (ABC/XYZ) and set service targets by segment.

Read more: Safety Stock and Reorder Points: A Practical Guide.

Back to contents →

Supply Chain KPIs

KPI	Definition / formula
OTIF	Share of orders delivered on time and in full ÷ total orders
Fill rate	Units (or lines) shipped from stock on first attempt ÷ units (or lines) ordered
Perfect order rate	Orders on time, complete, damage-free and correctly documented ÷ total orders
Inventory turns	Cost of goods sold ÷ average inventory value
Days of inventory on hand (DIO)	Average inventory value ÷ COGS × 365
Cash-to-cash cycle	DIO + DSO (days sales outstanding) − DPO (days payables outstanding)
Forecast accuracy	Often 1 − WAPE, by family and horizon
Forecast bias	Average of (forecast − actual)
Supplier on-time delivery	Receipts on or before promised date ÷ total receipts
OEE	Availability × Performance × Quality

Examples: COGS of $6,000,000 and average inventory of $1,500,000 give 4.0 turns and about 91 days on hand (365 ÷ 4). OEE with 90% availability, 95% performance and 98% quality = 0.90 × 0.95 × 0.98 ≈ 83.8%.

Define each KPI once, write the definition down, and keep it stable so trends mean something.

Back to contents →

from math import sqrt
from statistics import NormalDist

def safety_stock(service_level, sd_demand, lead_time, avg_demand=0.0, sd_lead_time=0.0):
    z = NormalDist().inv_cdf(service_level)
    return z * sqrt(lead_time * sd_demand**2 + (avg_demand * sd_lead_time)**2)

ss = safety_stock(0.95, 12, 9)        # about 59.2
rop = 40 * 9 + ss                     # about 419.2

import pandas as pd

df = pd.DataFrame({"actual": [100, 120, 80], "forecast": [110, 110, 100]})
err = df["forecast"] - df["actual"]
mae = err.abs().mean()                          # 13.33
wape = err.abs().sum() / df["actual"].sum()     # 0.1333
bias = err.mean()                               # 6.67

demand = pd.Series([100, 120, 90, 110])
level = demand.ewm(alpha=0.3, adjust=False).mean()
next_forecast = level.iloc[-1]                  # 103.84

WITH cogs AS (
  SELECT item_id, SUM(cogs_amount) AS annual_cogs
  FROM sales
  WHERE sale_date >= DATE '2025-01-01' AND sale_date < DATE '2026-01-01'
  GROUP BY item_id
),
inv AS (
  SELECT item_id, AVG(on_hand_value) AS avg_inventory_value
  FROM inventory_snapshots
  WHERE snapshot_date >= DATE '2025-01-01' AND snapshot_date < DATE '2026-01-01'
  GROUP BY item_id
)
SELECT c.item_id,
       c.annual_cogs / NULLIF(i.avg_inventory_value, 0) AS inventory_turns,
       365.0 * i.avg_inventory_value / NULLIF(c.annual_cogs, 0) AS days_on_hand
FROM cogs c
JOIN inv i ON i.item_id = c.item_id
ORDER BY inventory_turns;

WITH usage AS (
  SELECT item_id, SUM(quantity * unit_cost) AS annual_value
  FROM consumption
  WHERE txn_date >= DATE '2025-01-01' AND txn_date < DATE '2026-01-01'
  GROUP BY item_id
),
ranked AS (
  SELECT item_id, annual_value,
         SUM(annual_value) OVER (ORDER BY annual_value DESC
                                 ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW)
           / SUM(annual_value) OVER () AS cum_share
  FROM usage
)
SELECT item_id, annual_value,
       CASE WHEN cum_share <= 0.80 THEN 'A'
            WHEN cum_share <= 0.95 THEN 'B'
            ELSE 'C' END AS abc_class
FROM ranked
ORDER BY annual_value DESC;

SELECT supplier_id,
       AVG(CASE WHEN received_date <= promised_date THEN 1.0 ELSE 0.0 END) AS on_time_rate,
       COUNT(*) AS receipts
FROM purchase_order_receipts
GROUP BY supplier_id
ORDER BY on_time_rate;

Glossary (A–Z)

38 supply chain, planning, trade and AI terms, A–Z.

• ABC analysis: Ranking items by annual usage value (or another measure) into A, B and C classes so attention and service targets match importance.
• Backorder: Customer demand that could not be filled from stock and will be shipped later.
• Bill of materials (BOM): The structured list of components, sub-assemblies and quantities needed to make a product.
• Bullwhip effect: The amplification of order variability as demand signals move upstream from customers to suppliers.
• Capacity planning: Matching the resources (labor, equipment, suppliers) needed to meet the demand plan.
• Country of origin: The country where a good was produced or substantially transformed, as determined under applicable customs rules.
• Cycle service level: The probability of not stocking out during a replenishment cycle; the basis of the z-factor safety stock formula.
• Days of inventory on hand (DIO): Average inventory value divided by cost of goods sold, times 365.
• Demand planning: The process of forecasting demand and agreeing an unconstrained demand plan.
• Economic order quantity (EOQ): The order quantity that minimizes total ordering and holding cost: √(2DS/H).
• Fill rate: The share of demand (units or lines) shipped from stock on the first attempt.
• Forecast bias: A consistent tendency to over- or under-forecast, measured as the average of forecast minus actual.
• Harmonized Tariff Schedule (HTS): The U.S. classification system, published by the USITC, used to determine duty rates for imported goods.
• Incoterms®: ICC rules defining buyer and seller responsibilities for delivery, risk and costs in trade contracts.
• Inventory turns: Cost of goods sold divided by average inventory value.
• Just-in-time (JIT): A production approach that aims to receive and make items only as needed, minimizing inventory.
• Kanban: A visual pull signal (card, bin or electronic) that authorizes replenishment.
• Lead time: The time between placing an order (or starting a process) and the item being available for use.
• Master production schedule (MPS): The item-level plan of what to produce, in what quantity and when.
• Material requirements planning (MRP): A system that calculates component requirements and timing from the MPS, BOM and inventory.
• Mean absolute percentage error (MAPE): The average of absolute errors divided by actuals; undefined when actuals are zero.
• Nearshoring: Moving production or sourcing to a nearby country.
• On time in full (OTIF): The share of orders delivered on the promised date and in the full quantity.
• Overall equipment effectiveness (OEE): Availability × Performance × Quality for a machine or line.
• Prompt engineering: Structuring instructions, context and examples so an AI model produces useful, accurate output.
• Quality at source: Building inspection and responsibility for quality into each process step rather than end-of-line checks.
• Reorder point (ROP): The inventory position that triggers a replenishment order: d̄ × L + SS.
• Reshoring: Returning production or sourcing to the home country.
• Safety stock: Extra inventory held to buffer demand and supply variability during lead time.
• Sales and operations planning (S&OP): A monthly cross-functional process that produces one agreed plan for demand, supply and finance.
• Stock-keeping unit (SKU): A unique identifier for a distinct item, often by item and location.
• Tariff: A tax or duty imposed on imported (or occasionally exported) goods.
• Total cost of ownership (TCO): The full cost of a sourcing option, including price, freight, duty, inventory, quality and risk.
• Uncertainty (demand/supply): Variability that cannot be forecast; quantified by standard deviation and buffered by safety stock or capacity.
• Vendor-managed inventory (VMI): An arrangement where the supplier monitors and replenishes the customer’s inventory.
• Weighted absolute percentage error (WAPE): Sum of absolute errors divided by sum of actuals; a volume-weighted accuracy metric.
• XYZ analysis: Classifying items by demand variability (X = stable, Z = highly variable), often combined with ABC.
• z-score (service factor): The number of standard deviations of protection that corresponds to a target cycle service level.

Back to contents →

Further Reading

Authoritative sources only: professional bodies, universities, official government sites and primary documentation.

Professional bodies and research

• ASCM: Association for Supply Chain Management: CPIM and CSCP certifications, standards and the SCOR model.
• Institute for Supply Management (ISM): Procurement and supply management certification and research.
• ISM PMI® Reports: Monthly manufacturing and services indexes.
• MIT Center for Transportation & Logistics: Supply chain research and education.

Forecasting and data

• Forecasting: Principles and Practice (3rd ed.): Free online textbook by Hyndman and Athanasopoulos.
• statsmodels: Time Series Analysis: Python exponential smoothing, ARIMA and more.
• pandas documentation: Python data analysis library.
• Python statistics module: Includes NormalDist for service-level z values.
• Census M3: Manufacturers’ Shipments, Inventories, and Orders: Monthly U.S. manufacturing data.

Trade, tariffs and sourcing (official sources)

• Harmonized Tariff Schedule (USITC): Official U.S. tariff classification and rates.
• CBP: Basic Importing and Exporting: Import process fundamentals.
• CBP Rulings: Classification and origin rulings.
• USTR: Section 301 Investigations: Official Section 301 actions and notices.
• International Trade Administration: Market and trade resources from the U.S. Department of Commerce.
• Bureau of Industry and Security: U.S. export controls.
• U.S. Census Bureau: International Trade: Official import and export statistics.
• ICC Incoterms® rules: Delivery terms used in trade contracts.
• Reshoring Initiative TCO Estimator: Free total-cost-of-ownership tool for sourcing decisions.

AI governance and data security

• NIST AI Risk Management Framework: Voluntary framework for managing AI risks.
• NIST Privacy Framework: Managing privacy risk in systems and data.

Back to contents →

Maintained by Peter Klausz, MBA, founder of Statadventure, LLC and former Pratt & Whitney F100 supply chain manager. I’m open to contracting (consulting and contract engagements) and new full-time roles in supply chain, data/BI, operations, and AI/automation, especially with defense, aerospace, and Boston-area employers. Get in touch.