8/10/09

Forecasting Future Demand of Products

In 1987, Gordon Graham wrote a book, Distribution Inventory Management for the 1990s. In this book, Graham described what he considered to be the best method for forecasting the future demand for both seasonal and non-seasonal products. Let's take a quick look at these formulas:

    Non-Seasonal Products: Calculate demand for the upcoming month by averaging the usage recorded in the past six months.

    Seasonal Products: Calculate demand for the upcoming month by averaging the usage recorded in the upcoming six months, last year, and then applying a "seasonal trend factor" that expresses the anticipated increase or decrease in business experienced over the past year.

These are simple formulas. And at the time Gordon wrote the book, simple formulas were necessary for distributors to successfully manage their inventory:

  • Many buyers could not effectively deal with mathematical formulas or computers. Ten-key calculators were considered "state-of-the-art" technology. In fact, most purchasing decisions at the time were based on "SWAG" (silly, wild-ass guessing). Any formula (including Graham's) introduced to provide consistency in ordering had to be fairly simple and easily replicated on a calculator.

  • Computers did not have the power to perform comprehensive forecasting formulas for thousands of parts within a reasonable period of time. Calculating Graham's simple average for thousands of items stretched the physical capabilities of most computer systems.

The demand forecasts produced by the Graham formulas were generally more accurate than the predictions of the guy with the dull pencil and clipboard out in the warehouse. But there was still a considerable difference between Graham-based predictions and what was actually sold. At the time, these deviations were considered "unavoidable," and there was no way around them.

Now consider how market conditions have changed since 1987:

  • Technology has allowed distributors to expand and increase their market areas. The result: You face more competition than ever. This competition has created more pressure on distributors to consistently have the products their customers want, when and where they want them.

  • Increased competition has also put pressures on profit margins. Distributors have to offer lower prices in order to retain current business and attract new customers.

  • The number of new products introduced to the market continues to increase at a rapid rate.

These conditions present some unique challenges:

  • Decreased margins tend to limit the amount of money a distributor has available to invest in inventory.

  • Distributors must spread the money available to invest in inventory over a greater number of products.

  • Customers are less tolerant if product availability does not meet their expectations.

You're obviously in trouble if you don't have the inventory your customers expect you to have. And if you've bought too much of an item, your money is tied up and can't be invested in the other products that allow you to take advantage of new sales opportunities.

These challenges require the best possible product forecasting. You can no longer accept as "inevitable" great deviations between forecasts and actual sales. Formulas developed just to be "easy to understand" and "better than a guy with a clipboard" have to be replaced with more comprehensive methods.

Products with different patterns of usage, and different replenishment methods require different forecasting formulas. We need more than one formula for non-seasonal products, and one formula for seasonal products. For example, a product whose sales mirror local economic conditions requires a different formula than a product with steady, fairly predictable sales. And just as important, each formula needs to be easy to understand.

During the next several months, we'll look at some of the 29 different forecast demand formulas developed by EIM. We're going to start with a formula for non-seasonal products with fairly consistent usage. These are items that sell regularly and whose volume has increased or decreased less than 20% per month during the last several months.

When forecasting the usage of non-seasonal products with fairly consistent usage, we want to average the usage that was recorded during the past several inventory periods. But we also want to "weight," or place more emphasis on, the most recent month. Why?

  1. There are often trends in a product's usage as it becomes more or less popular over time. For non-seasonal products, demand in the upcoming inventory period will more likely be similar to the usage recorded in the past several inventory periods than what happened six, eight, or twelve months ago.

  2. At the same time, there is usually a certain amount of random variation in a product's usage from one inventory period to another. Notice how the usage of the item in the first example below has fluctuated over the past five months. This "up-and-down" pattern of usage is common for inventory items with moderate-to-high sales. If we were to use just the most recently completed one or two inventory periods in our calculations, the random fluctuations in usage would probably have too great an influence on the forecasted demand. We want to include enough history to ensure that random fluctuations do not have a significant impact on a product's forecast.

Here is a common set of weights to use in calculating demand for a non-seasonal item with moderate-to-high sales:

  • Place a weight of 3.0 on the usage recorded in the most recent period.
  • Place a weight of 2.5 on the usage recorded in the next previous period.
  • Place a weight of 2.0 on the usage recorded in the next previous period.
  • Place a weight of 1.5 on the usage recorded in the next previous period.
  • Place a weight of 1.0 on the usage recorded in the next previous period.

Let's see how the forecast for an item is calculated with the following usage history. Usage is the quantity of a product sold, transferred, used in assemblies or repair orders, or otherwise taken from stock.

MonthTotal UsageNumber of Business
Days in Month
Usage per
Business Day
June148207.4
May133197.0
April126187.0
March110225.0
February104205.2

Note that we've specified the number of business days in each month, and determined the usage per business day. Utilizing usage per business day provides more accurate forecasting than traditional forecasting methods that rely on total monthly usage or usage per calendar day. After all, if a company is closed for several days during a month (remember the Christmas holidays?), considering that month's lower usage equally with the usage recorded in other months tends to underestimate future forecasted demand. For example, in the chart displayed above, total usage recorded in May (133 pieces) is about 5.5% higher than total recorded in April (126 pieces), but the demand per business day is the same.

We will apply the weights of the demand calculating formula to the usage per business day for the five preceding months to determine the forecast demand for July:

MonthWeightUsage per
Business Day
Extension
June3.07.422.2
May2.57.017.5
April2.07.014.0
March1.55.07.5
February1.05.25.2
Total10.066.4

The extension (66.4) is divided by the total weight (10.0) to determine our prediction of the demand per business day for July (6.64 pieces per day). And this demand per day is multiplied by the number of business days in July (21) to predict the demand of 139.4 pieces for the inventory period.

Compare the results of this calculation to the demand predictions provided by other forecast formulas and methods. We think you'll be impressed with the results. Next month we'll look at non-seasonal products with significant increasing or decreasing usage. In the meantime, if you have any specific questions, please let us know.

Some items, like beach umbrellas, are more popular in summer than in winter. On the other hand, portable heaters enjoy much higher sales when the weather gets cold. These are seasonal items. But the weather is not the only factor that determines whether or not an item is seasonal. If a product's usage is controlled by an event (such as Christmas or the start of school) or an annual activity (like yard clean up in the fall), the item is also considered to be seasonal. The usage of a seasonal product rises and falls throughout the year. Look at this seasonal item's usage history:

Jun
1999
May
1999
Apr
1999
Mar
1999
Feb
1999
Jan
1999
Usage
1999
?300150805030

Dec
1998
Nov
1998
Oct
1998
Sep
1998
Aug
1998
Jul
1998
Jun
1998
Usage
1998
50100150300520460400

Usage of the product is very low during the winter months. But in early spring, sales begin a gradual increase and peak during the summer months of June, July and August. If we forecast demand for June 1999 by using the formula for non-seasonal products with consistent usage (described in Part One), we get the following result:

MonthTotal UsageNumber of Business
Days in Month
Usage per
Business Day
May3001915.8
April150188.3
March80223.6
February50202.5
January30221.4

MonthWeightUsage per
Business Day
Extension
May3.015.847.4
April2.58.320.8
March2.03.67.2
February1.52.53.8
January1.01.41.4
Total10.080.6

The extension (80.6) is divided by the total weight (10.0) to determine our prediction of the demand per business day for June of 8.06 pieces. Because June has 20 business days, demand for the inventory period is 161.2 pieces (20 days x 8.06 pieces per day).

Remember that demand is defined as a prediction of the usage of a product during the upcoming inventory period. Is 161 pieces a good forecast of June's usage? Probably not. After all, usage in June 1998 was nearly three times this amount (460 pieces). It is obvious that we need different formulas for calculating the demand for seasonal items.

We've found that one of the best indicators of what demand will be for a seasonal item next month is the usage recorded during the upcoming several months, last year. For example, one formula for forecasting demand for seasonal items considers the usage for the upcoming month and the following month last year, applying the following weights:

  • Place weight of 2.0 on the usage recorded in the month being forecast, last year.

  • Place weight of 1.0 on the usage recorded in the month following the month being forecast, last year.

MonthTotal UsageNumber of Business
Days in Month
Usage per
Business Day
June
1998
4001921.1
July
1998
4601825.6

MonthWeightUsage per
Business Day
Extension
June
1998
2.021.142.2
July
1998
1.025.625.6
Total3.067.8

The extension (67.8) is divided by the total weight (3.0) to determine our prediction of the demand per business day for June of 22.6 pieces. Because June, 1999 has 20 business days, demand for the inventory period is 453 pieces (20 days x 22.6 pieces per day).

But there is a problem with forecasting demand with history that is a year old. Business in the branch where the item is located, or in its particular line of products, may have increased or decreased during the past 12 months. For this reason, a "trend factor" can be applied to the results of the weighted average formula to reflect overall changes in your volume of business.

Many systems will allow you to manually maintain trend factors. Say, for example, you determine that the sales volume in our item's product line increased 20% over the past year. To determine the actual demand forecast for the product, we'd increase the result of the seasonal weighted average formula by 20% to determine the actual demand forecast for June, 1999:

22.6 pieces/day + 20% = 27.1 pieces/day

More advanced systems calculate a suggested trend factor by comparing the total usage in the last three completed months (before the forecast demand calculation) to the total usage in the same three months in the previous year:

Total Usage March,1999 - May, 1999 = 530 pieces

Total Usage March 1998 - May, 1998 = 462 pieces

(530 - 462) ÷ 462 = 14.7%

Business in the past three months was 14.7% greater than the same period last year. This percentage is added to the results of the weighted average formula:

22.6 pieces/day + 14.7% = 25.9 pieces/day

Whether specified manually or calculated automatically by the system, trend factors must be applied whenever seasonal forecast formulas are utilized to compensate for the change in business experienced over the past 12 months.

http://www.effectiveinventory.com/article26.html

No comments: