Demand planning in the seed industry: It’s a lot like fashion

Demand planning in the seed industry has some unlikely similarities to demand planning in the fashion industry. As in the fashion world, the seed market demands constant innovation to keep pace with the competition. It takes a constantly improved variety of products to satisfy customer demands.

Here are a few of the seed industry’s consistent challenges:

• Seasonal, intermittent sales
• Erratic demand patterns and low predictability
• Siloed short- and long-term forecasts
• High proliferation of crop SKUs with shorter life cycles
• A relative shortage of big-volume drivers
• High sensitivity to external factors such as weather and macroeconomic conditions

The seed industry’s demand planning process involves two time horizons: the long-term and the short-term. The long-term (60 months) drives production, planning and budgets. The short-term (18 months) deals with seed allocation, capacity assessment, preprocessing operations and monthly purchase orders.

Demand classification

The process of demand classification analyzes the primary forecast variable: sales history. It’s important to understand sales behavior, including trend, seasonal, intermittent or lumpy. This information aids in finding the best forecast algorithms. For example, using Croston’s algorithm can help you forecast intermittent demand for effective inventory management.

In practice, the standard coefficient of variation (COV), standard deviation/mean and volume classify demand into high/low X variability/volume. However, COV doesn’t work for highly seasonal data. Statistical COV through decomposition facilitates a more accurate identification of variability. It classifies a sales data time series into a trend, seasonal and error component.

This approach increases intersections for an advanced statistical forecasting technique.

Increased SKU count via standard COV of a low-variable category “X.”

Increased SKU count via statistical COV of a low-variable category “X.”

Forecasting levels

There’s also a growing need for different crop-based levels of forecasting. Traditionally, a single-level product-customer-time combination is fixed across the product line and the customer.

Using different levels to forecast by crop/crop type can be an excellent way to determine demand with greater accuracy. For example, a biennial crop requires a different time level than a perennial crop.

You can achieve this through exploratory data analysis (EDA). Different item/product/time hierarchies help you identify the level(s) with the best forecast drivers. Among these are appropriate lifecycle classification and accurately classifying trend, seasonality and intermittency.

With the correct demand classification and an appropriate forecast level, accuracy improves by 20–30%.

Accuracy of a forecast generated at traditional method-selected levels.

Accuracy of a forecast generated at analytical method-selected levels.

New product introduction (NPI) and phase-in, phase-out

With consistent customer demand for newer and better seed varieties, NPI is a standard part of the process, as is the resulting SKU proliferation. NPI management is critical because it demands a higher degree of effort and increased costs from a mature demand planning process.

Phase-in and phase-out in the seed industry is different than the same process in other industries. It’s a primarily manual, siloed process. You can make phase-in and phase-out decisions easier. Use a systematic approach to NPI forecasting with a similar product as a baseline (based on characteristics such as seed grade or treatment method) and a machine learning-based algorithm to account for the seasonality curve.

These developments have led to the design of highly sophisticated data-archiving procedures. With them, you can pick an appropriate “similar product” seed based on its characteristics to create a detailed forecast by sales, statistics and consensus.

You can systematically incorporate these approaches into NPI projections with no manual intervention. It’ll help you project an NPI launch curve with the option to ramp up percentages.

Leveraging the capabilities of artificial intelligence (AI)/machine learning (ML)

The need to incorporate next-generation AI/ML algorithms is growing. Many factors, such as weather and other climate characteristics, macroeconomic conditions and consumer behavior, impact seed demand. Applying ML to these factors can help you quickly respond to changes in external drivers and proactively adjust their forecasts.

For example, a crop’s variety price index can positively impact demand when accompanied by another variable, like precipitation. This sort of pattern recognition and demand sensing capability can result in 5–10% increases in forecast accuracy and is only possible when you take advantage of this technology.

— By Diwakar Prasad