Demand Planning & Forecasting - Best Practices

Collaborate to improve demand planning; forecasts are never 100% accurate, so gather real-time downstream data from customers and distributors to enhance planning and reduce inventory risk.

Decide the forecast horizon by identifying who uses it and how it adds value in the supply chain. Align horizons—from six months to 12–36 months—with stakeholder needs, not monthly defaults.

Learn how forecast review frequency impacts accuracy, from monthly to daily, using fresh data to tweak plans while balancing time costs and potential chaos with the forecast value kpi.

Learn how to separate base sales from incremental sales during promotions, using simple methods in Excel to forecast the baseline sales and promotional impact in the CPG sector.

Improve demand planning and forecasting by detecting and cleaning outliers through winsorization, replacing extreme values with percentile-based limits (e.g., 90th and fifth percentiles) to enhance forecast accuracy.

Explore cannibalization, or halo effects, in product launches and promotions. Learn to identify cannibalization relationships, forecast its impact from historical data, and coordinate the supply chain to minimize losses.

Explore self cannibalization and cross cannibalization using Krishna buttermilk and Krishna flavored milk examples, showing how promotions shift demand and cause adjacent-month sales dips.

Learn demand segmentation to prioritize forecasting by classifying products into ABCD based on volume and variability, focusing effort from high-volume low-variability A to high-variability B and beyond.

Review the Nike case of 100 million loss from insufficient testing and learn to test forecasting models with diverse splits (80/20, 70/30) and data quantities for better supply chain planning.

use humor to reinforce demand planning and forecasting concepts as the lecture explores memes about procurement, delivery performance, route optimization, and logistics to ease learning.

Compare forecast accuracy metrics such as RMSE, MAE, MAPE, and bias to choose the best KPIs for your dataset, considering outliers, seasonality, and experimentation.

Prioritize SKUs and models with the 80/20 rule, focusing on the top 20% that drive most revenue. Apply advanced models to them and simple methods to non-critical SKUs.

Learn how to split historical data into training and testing sets (commonly 80/20), train an exponential smoothing model, optimize alpha, and evaluate forecast accuracy on unseen data to prevent overfitting.

Explain model initialization for exponential smoothing and show how using future demand creates data leakage that inflates accuracy, then advise removing the initial forecast from accuracy calculations to avoid leakage.

Learn how much historical data to use in demand forecasting, balancing noise reduction with pattern detail. Experiment with granularity from SKU to product family to find the best level.

Identify sources of forecasting error, including process flaws, biased estimates, data quality and cleaning, and model selection, to improve demand planning accuracy.

Apply a framework to choose the best forecasting model by balancing forecast accuracy and bias, rating models, and selecting the top fit for a product family.

Prioritize thousands of SKUs efficiently with an exception review, using error-based and volume based methods to focus on high impact products and reduce manual forecasting effort.

Explore the forecast review process to measure forecast accuracy, compare absolute deviation, percentage-based, and relative-based KPIs, use normalization, and track periodically to guide demand plan adjustments.

Transform demand reviews into a continuous improvement process by using a structured agenda—feedback from SNP, action item reviews, metric evaluations, variance analysis, corrective actions, and consensus on unconstrained demand.

Monitor demand review metrics to improve aggregate demand plan performance, including volume, accuracy, bias, tracking signal, mix accuracy, delivery lead times, accounts receivable days of supply, and revenue performance.

Align demand planning with business strategy by reviewing inputs like market assumptions, history, and forecasts. Engage cross-functional attendees to produce a constrained final plan and identify opportunities and risks.

Discover three tips to improve demand review: prioritize forecast accuracy, run decision-oriented meetings with actionable outcomes, and communicate forecast bias and accuracy to executives to boost revenue and service level.

Traditional stat forecasting models rely on history and methods like moving averages, ARIMA, and linear regression; however their assumptions struggle with changing factors, prompting a shift to machine learning.

Machine learning forecasting models differ from traditional stat forecasting models by integrating internal and external factors, delivering higher short-term accuracy with methods like random forest, xgboost, and prophet.

Compare traditional stat forecasting models (ARIMA, Holt-Winters, linear regression) with machine learning models (random forest, XGBoost, neural networks) to decide the best approach for data volume, quality, and cost.