The most difficult issue is product mix, seconded only by optional product features. In the retail bedding market, an example of product mix is size (e.g., sheet size); and an example of product features is packaging (e.g., one or two sheets). Other markets have other alternatives for product mix and product features. But, without accurate quantities, they all present a large stumbling block to manufacturing. Manufacturing needs to know how many of each to buy material for, and how many of each to build, and when to build them. The sales revenue forecast requires manufacturing to guess on mix and feature quantities; and no one can run a business on guessing!
Companies who do the best forecasting use a combination of both the top-down sales revenue forecast and the bottom-up product line-item forecast; and they use a well defined business process to marry the two together into a final company forecast.
The Bottom-Up Product Forecast
The bottom-up product forecast is a prediction of future product demand from the customers. Many times a company will use past performance to predict future demand for each product line-item; however, past performance is only one element of a good forecast. In addition, the business analyst must take into account any changes that will occur due to competition, due to changes in the sales channels, due to company promotional events, and most importantly due to the movement of the product through its natural life cycle from cradle to grave.
Consequently, a great product forecasting tool must take into account many key influencing factors.
The Seven Characteristics of a Great Product Forecasting Tool
Some companies sell through a one-step sales channel; some companies sell through a two-step sales channel; and some companies sell through a combination of both types of sales channels. A one-step sales channel is one in which the company sells directly to the consumers of the product; and a two-step channel is one in which the company sells to another company which sells to the consumers of the product. An example of a two-step channel is a wholesaler who sells to multiple retailers, or a store chain which sells through many retail outlets. Sometimes, however, a two-step channel acts like a one-step channel. This occurs when the channel will not provide point-of-sale information about their sales.
When the point-of-sale information is available from a two-step channel, the forecast has the potential of being more accurate, since it starts with actual sales to the users of the products. A forecast in a two-step channel is performed on rate-of-sale information, which is defined as the point-of-sale quantity divided by the number of sales outlets; while a forecast in a one-step channel is performed on the quantity of products sold by the company, called sales-out data. Almost all product forecasting tools use the sales-out data only; they do not use rate-of-sale data at all.
A great product forecasting tool will be able to use either type of product sales data to calculate its forecast, and will even be able to use both types in the same forecast.
The heart of any good product forecast is the actual sales data for the product. As we have seen earlier, this actual sales data differs depending upon the sales channel. A forecast for a two-step channel uses point-of-sales quantities and store counts; while a forecast for a one-step channel uses the quantity of products sold by the company. In both cases, it would be impossible to predict future demand without having past sales upon which to base the forecast. For example, if my sales averaged 100 units per week for the last 10 weeks, my best first guess at a future demand forecast would be 100 units per week, assuming all other influencing factors stay the same.
A great product forecasting tool must be able to use past product performance as a basis for calculating future product demand, assuming past performance is available.
Every product progresses through a similar product life cycle from cradle to grave, although the times during each stage may differ. A product’s sales characteristics will be different during the different stages of its life: beginning, middle, and end.
The most commonly understood forecasting methodologies are used in the middle of the product’s life. At this stage there is actual sales data and a number of well understood analytical methods to use; such as average, rolling average, least squares (linear, polynomial, exponential, logarithmic, and power), and exponential smoothing (single and double).
At the end of a product’s life it is important to determine how to predict the reduction in sales up to the point if its end-of-life, at which time there will be no more sales. Here the business analyst needs to be able to force the reduction in sales in an attempt to reduce the final inventory down to zero.
The beginning of a product’s life is the most difficult time to accurately predict future demand. This is when the business analyst needs to make their best guess. There are no formal forecasting methodologies to help at this stage of the product’s life. However, what can be done is to offer a number of ways to model the beginning stage based on other similar products; and these methods must model the way different retail chains make their initial purchases of new products.
A great product forecasting tool must provide a number of different forecasting methodologies to be used at each of the three stages of a product’s life.
It is important to understand historical trends of various product categories, and to then be able to use this trend information in predicting future demand. The most commonly used forecast trends are seasonality and advertising.
Seasonality takes into account the seasonal trends of a product category which appear to occur every year. A typical example of seasonality is sales have been 20% higher during the first two weeks of December and 10% lower during the month of August.
Advertising takes into account the effect of advertising on weekly sales. For example, if in the past we typically got a 10% jump in sales the week after a certain kind of promotion, and a 20% jump the next week, and then a 5% decrease the week after that; then the next time we run a similar advertising promotion we would expect a similar effect on the weekly sales.
Another historical trend that might appear in a two-step channel is product adoption by stores. For example, in the beginning of a new product’s life a chain may introduce it into a limited number stores to start, and if the product does well they will then introduce it into all their stores; and at the end of a products useful life a chain may begin reducing the number of stores each week.
A great product forecasting tool must be able to take into account at least these typical historical trends in predicting future demand.
There are times when the actual sales data will contain anomalies. A typical example is missing data for one of the weeks; or maybe we just don’t believe that one of the week’s data is accurate. It is important, in order to get good forecasting results, to be able to modify the data for these types of identified singularities; and it is important to be able to do this regardless of the channel. In addition, there will also be times when the forecast results will need to be modified before it is finalized.
A great product forecasting tool must allow the business analyst to be able to modify data anomalies in order to get the most accurate future demand possible, given all the information we have.
There are times when the business analyst will want to look at the results from many different forecast methodologies, at the same time, in order to be able to get the most accurate prediction of future demand possible. We mentioned a number of well understood analytical methods above; such as average, rolling average, least squares (linear, polynomial, exponential, logarithmic, and power), and exponential smoothing (single and double). In addition to these, the analyst will need to be able to combine two or more typical analytical methods with mathematical expressions and functions; such as plus, minus, times, divide, power, sum, min, max, and average.
It is also important to allow the business analyst to select the resulting method to use, or to allow the product forecasting tool to select the best resulting method; where “best” can be defined as the minimum mean squared error (MSE) or the minimum mean absolute error (MAE).
A great product forecasting tool must be able to allow the analyst to specify a multiplicity of forecasting methodologies for each product line-item, and then tell the tool how to make the best selection, automatically or manually.
A great product forecasting tool will allow the business analyst to visualize the forecast results as weekly quantities and as a graph of the quantities. Here is an example:
In addition, a great product forecasting tool will allow the analyst to compare past forecasts with the latest actual sales numbers so that the worst offenders can be fine tuned in future forecasts. Here is and example of this actual vs. forecast analysis:
Product Forecaster™
Ken Gilbert & Associates’ Product Forecaster™ is a great product forecasting tool; and it incorporates all seven characteristics described above.
Our strategy was not to automate all aspects of the product forecasting process, but to provide a comprehensive set of forecasting capabilities which allow the business analyst to interact with the Product Forecaster™. This way the analyst can more accurately predict future demand based on a combination of past performance and the analyst’s knowledge of key influencing factors. A high-level view of the process is:
A functional view of the Product Forecaster™ is:
Our key differentiators are:
■ Best-of-breed Product Forecasting
● Forecasting based on:
- sales-out (one-step) and/or point-of-sale (two-step) data collected from the channel
- where the product is in the product life cycle
- multiple scenarios of “un-trended” data
● The ability to see potential issues in the channel before they occur
● Low risk implementation
■ Realistic Modeling
● Mathematical Modeling
- of the specific requirements of a one-step or a two step channel
- of the specific phases of the product life cycle
● Statistically precise forecasting methodologies
● Visually oriented real-time interaction with the data
■ Flexible Operation
● Easy data migration to and from the corporate accounting system
● Forecast for a one-step or a two-step channel, or both
● Standard internal API for adding additional forecasting methodologies
■ Cutting Edge Technology
● Standard statistical forecast algorithms, built-in functions, and arithmetic expressions
● Microsoft development tools and run-time environment
- Built in the .NET Framework
- Using C#.NET programming
- Using ADO.NET for database access
● Migration to any standard database technology
__________
Product Forecaster™ is a trademark of Ken Gilbert & Associates.
Copyright © 2004 Ken Gilbert & Associates. All rights reserved.
