Sales Forecasting Key Points: I. Why is Sales Forecasting Important? II. Estimating Market Potential III. Sales Forecasting IV. Sales Forecasting Example I. Why is Sales Forecasting Important? Sales Forecast = an estimate of future sales. Companies need to forecast their sales to make good decisions on staffing, purchasing, retail store locations, and a host of other marketing decisions. The sales forecast is the basic tool of managerial accounting and is used to compute expected cost of goods sold, planned output levels, and expected profits for a given time period. The sales forecast is probably the most crucial and difficult step in the strategic planning process because of the large number of factors that can influence actual sales volume. Your goal in sales forecasting is to be as accurate as possible. This is very difficult to do because so many things can influence sales. Just a handful of things that can influence sales are price, product performance, promotion, competitor actions, interest rates and even the weather. For example, when it is warmer people drink more soft drinks compared to when it is colder. A hot summer often yields high soft drink sales. There are two basic types of sales forecasting: qualitative and quantitative. Qualitative sales forecasts use human judgement. Quantitative sales forecasts use math. The sales manager does not care if the forecasting technique is qualitative or quantitative. What the manager really cares about is if the forecast is accurate. According to an analysis of a large number of marketing job advertisements on Monster.com, forecasting is an important job skill for 27% of entry level marketing jobs, 33% of marketing jobs requiring 2-4 years of experience, and 55% of all upper management marketing jobs. The higher up you get in an organization, the more likely the job will require sales forecasting skills (Schlee and Harich 2010). Sales managers are often required to produce sales forecasts. Couple of Things to Remember About Sales Forecasting: 1] Sales Forecasts are almost always wrong. It is very difficult to predict the future. But they must be done. 2] It is easier to make an accurate sales forecast for the near term. Predicting sales tomorrow or next week is much easier than predicting what sales will be 180 days from now. The farther in the future you are making your sales forecast, the more inaccurate it will be. Long run sales forecasts for strategic planning purposes is especially difficult. All sales forecasting techniques suffer from this problem. Discuss and illustrate. 3] Making accurate sales forecasts is much harder for new products compared to existing brands. 4] Forecasts are no substitute for knowledge of actual demand. This seems obvious but has some important ramifications. If a customer has placed an order for 500 units deliverable next month, the 500 units needed next month is the actual number which should be produced for that customer. We would use that number for the customer instead of a sales forecast. 5] Sales forecasts are usually done for each product and sales territory of a firm. Sales forecasts are very useful in setting up equitable sales territories and sales quotas. 6] Many companies use multiple forecasting methods. Forecasters see how accurate a particular technique is compared to actual sales. About the only thing we have to go on is how accurate a particular sales forecasting technique has worked in the past. If we have historic sales data quantitative sales forecasts can be computed and compared to actual sales results. One can also assess how accurately qualitative sales forecasts have worked over time. Business have sales records on all of their products. They generally also keep records on past qualitative sales predictions. Both the qualitative and quantitative estimates can be measured for accuracy using mean absolute percentage error (MAPE). MAPE is calculated by taking the average of the sum of the absolute value of all past forecasting errors: MAPE = sum(absolute value (sales forecast - actual sales)/ actual sales)) number of forecast & actual pairs Example: Suppose your sales results and sales forecasts for the last 5 years were: Year Sales Sales Forecast Absolute Value ((forecast - actual)/actual) 2000 $100,000 $95,000 .050 2001 $135,000 $140,000 .037 2002 $122,000 $140,000 .148 2003 $144,000 $135,000 .063 2004 $156,000 $155,000 .006 2005 $165,000 MAPE = .050 + .037 + .148 + .063 + .006 = .0608 5 Another means of assessing sales forecast accuracy is computing the root mean squared error (RMSE). MAPE is easier to interpret than RMSE. For that reason I’m not covering how RMSE is calculated. When we look at past sales data we look for trend, seasonality cycle and outliers. This is often done by examining the sales data graphically. Trend = the overall direction of sales over time. The trend may be increasing, decreasing or constant. Seasonality = sales in a calendar year react in a predictable fashion. For example, candy sales tend to increase for Valentines Day, Halloween and Easter. In Auburn, hotel bookings greatly increase during home football games and graduation. Cycle = variation in sales caused by the business cycle. Overall sales for many products increase when more people are employed and when household income rises. During recessions overall sales fall for many products. Outliers = A highly unusual level of sales that is not expected to repeat. For example, the summer school with the highest enrollment in College of Business history was the last summer before the GPA requirement to take business school classes went from a 2.0 to a 2.2. II. Estimating Market Potential A. Introduction: Market potential is the estimate of maximum demand in a time period for an industry (not a company or brand). Market potential is based on the number of potential users and their purchase rate. Company sales potential - is the maximum amount a firm can sell under optimum conditions. Estimation of market potential is strongly linked to the economic resources of a given area and the expected performance of the national and local economy. It is also tied to the overall economic picture facing the industry. Sales forecasts are predictions of the actual sales volume that is expected in a future time period. Sales forecasts are almost always less than company potential. B. Estimating Market Potential - A forecast of company potential is computed prior to the sales forecast. Overall company potential is usually broken down by geographic area or customer group in order to more effectively allocate resources to the distribution, selling, and promotional efforts. It is also the first step in the sales forecast. All estimates of potential are based on two components, the number of possible uses of the product and the maximum expected purchase rate. 1. Sales and Marketing Management's Buying Power Index is published in July and gives an composite index of consumer demand for specific metropolitan areas, cities, counties and states for the U.S. SMM buying power index gives a percentage of buying income for each area. The firm then obtains national sales figures for the product of interest from internal, trade, or government sources and determines the expected sales potential for a given area. Example: Say Columbia, S.C. has .01% of national retail sales, .09% of national buying income and .02% of the U.S. population. Columbia's buying power index is 5I + 3R + 2P/10 =.052. Then if expected national retail sales is $2,000,000,000,000 [2 trillion bucks], Columbia would have .052% x 2 trillion = 1,040,000,000 or 1.04 billion dollars or .052% of the U.S. buying potential. The Buying Power Index has had a good track record for accuracy, but it is often difficult to get relevant sales data for certain products. 2. NAICS Method (North American Industry Classification Method) gets industrial market potentials gained from the U.S. Census of Manufacturers. The US Census of Manufacturers is done once every 5 years - which can make the data stale if you are late in the survey cycle. It gives data like number of companies, number of employees, value of materials and shipments, etc... Briefly discuss the data problems that the NAICS code data suffers from. When using this method, first identify all NAICS codes that make use of the product or service. Next, the firm must select an appropriate data base for estimating the amount of product that will be used by each SIC code (example: number of machines for number of employees or number of machines by x dollars in sales for each firm in the industry). 3. Trade Press Sources - discuss. 4. Chain Ratio Method - determining company potentials for individual products by applying a chain of ratios or usage rates to an aggregate measure of demand. A firm could start with the % usage rate by industry or the final consumer. Then estimate replacement rates and the percentage of expected market share. Essentially, it is backing into a company potential by making a series of estimates about the general market and then the company’s potential. The chain ratio method will often make use of scrappage rates of machinery or appliances. This does not work for new items where customers are still being introduced to the product. II. Sales Forecasting Sales forecasts can be based on past sales (historic data), indicators of future sales activity, or expert opinion of future sales. When no historic data is available (new territory or new product), or when changing market conditions cause the historic data to be biased or unusable, then qualitative techniques must be used. A. Qualitative Sales Forecasting Methods - can be used when historic data is or is not available. The accuracy of past qualitative sales forecasts can be assessed using mean average percentage error (MAPE) just like quantitative forecasts. Discuss general advantages and disadvantages of qualitative sales forecasting methods. Sales Force Composite, Jury of Executive Opinion and Surveys of Customer Buying Intentions are all very popular sales forecasting methods in industry. All of these methods are only as good as the people making the estimates. Thus, they can be very accurate, or inaccurate. However, they can directly input knowledge about future events while quantitative sales forecasting methods relies only on historical data. 1. Sales Force Composite is easily and cheap to obtain. Salespeople are asked to predict sales for a future time period by customer, territory and/or product. Very popular with industrial marketers. Forecasts prepared by salespeople may be biased because the projections are often used to set sales quotas. Salespeople will hesitate to predict more sales volume than they can deliver. 2. Jury of Executive Opinion - is predictions of sales by product and territory by a group of experienced managers. It is generally faster to obtain than the sales force composite (explain). Since estimates are based on experience, it is difficult to teach new managers how to do this accurately. 3. Surveys of Customer Buying Intentions - ask customers what they intend to buy in the future. These can be time consuming and expensive. They are most successful in business markets where there are limited numbers of customers with specific needs and when those firms tend to follow through on their purchase intentions. B. Quantitative Sales Forecasting Methods - with the exception of leading indicators, all use historic data. A data set is included after the leading indicators section and examples of all of the calculation methods are applied to this data set. 1. Leading Indicators are computed by the government. It gives a good and free estimate of expected future national business activity. Often is a good indicator of future overall economic trends, but for many products is only a part of the influence of goods sold. However, if sales are heavily influenced by basic changes in the economy, leading indicators can be very useful. The leading indicators are not giving direct sales forecasts and are often used as inputs into time series models, sales force composites or jury of executive opinions. Leading indicators have the additional advantage in that they are often sensitive to turning points in a series (changes in direction of the economy). The idea is to find a general time series that is closely related to company sales, yet is available several months in advance. Examples are GNP and new housing starts. 2. Seasonal Adjustments adjusts the past sales data for seasonal effects. Many products have sales that vary by seasonal effects. Seasonal adjustments gives managers a tool to adjust sales forecasts to reflects these seasonal changes. Seasonal adjustments are widely used in business and lower forecasting error. The seasonal adjustment can be either quarterly or monthly. Annual forecasts cannot be seasonally adjusted. Often, data is seasonally adjusted and then applied as an input into another quantitative forecasting method. 3. Naive Forecasting - using the previous periods performance as a forecast for the current period. 4. Moving Averages expected future sales is a function of a simple fraction of past sales. The crucial decision is the number of periods to include in the averages which determines the speed that the forecast reacts to recent sales patterns. Two and three period moving averages are the most commonly used. St+1=[(St)+(St-1)+(St-2)+...+(St-n)]/N 5. Exponential Smoothing uses a decreasing weighted average of past sales to forecast future sales. The weights are a fraction between 0 and 1 and the selection of the fraction determines how quickly the forecast reacts to changes in recent sales. Unlike the other methods discussed above, the forecaster has to choose the alpha level that will be used. Under most circumstances, the alpha will be greater than .5 to put more weight on the most recent actual sales results. Generally, the alpha level will be chosen by assessing how accurate different alphas were in the past using historical data. With computers, it is easy to determine the accuracy of different alpha levels. Unlike the other methods, exponential smoothing calculations use two different formulas. You use formula 1 to start the exponential series. You use formula two for the other data points in the series. Formula 1: Alpha is between 0 and 1. St+1 = aSt + (1-a)St-1. Formula 2: St+1 = a(St) + (1-a)last period forecast. Go to the Crystal Pure Water Company data. For an alpha = .9, the forecast for 1998 is .9(3523) + .1(2843) = 3,455 units. Forecast for 1999 is: .9(4507) + .1(3455) = 4401.8. Note, for the actual forecast we would round to whole units. Seasonal Adjustments, Moving Averages, Exponential Smoothing, and the qualitative methods have an advantage over linear and multiple regression models in that the data is readily available. Seasonal Adjustments, Moving Averages, Exponential Smoothing models all assume that there will be no external shocks that will impact future sales. Although they may provide accurate forecasts most of the time, they are unable to adjust immediately to a drastically different situation. Discuss and illustrate. All of the mechanical methods suffer from this to an extent and for that reason qualitative methods can be very important. 6. Linear & Multiple Regression Econometric Models give basic linear model, discuss the importance of being able to accurately gather forecasts of independent variables easier than that for dependent variables. 7. Trend Projections - are usually done either a] graphically or b] through time series regression. The forecaster estimates the trend from past data and adds that figure to current sales level to obtain the sales forecast. E. Picking a technique depends on 1) time factors, 2) cost factors, 3) computer availability, 4) mathematical expertise, 5) data requirements and most importantly - 6) accuracy. Discuss. Many firms generate multiple forecasts using different methods and then decide on a final forecast number. The accuracy of forecasting methods used in the past can be assessed using Mean Absolute Percentage Error (MAPE). MAPE can be used to assess the accuracy of both qualitative and quantitative methods (explain). F. Summary - Because the sales forecast is crucial to the planning process and each method of sales forecasting has different strengths and weaknesses, most firms use more than one sales forecasting method. You should understand why firms estimate market potential and forecast sales, and be able to interpret and apply the various techniques discussed. Schlee, Regina Pefanis and Katrin R. Harich (2010), “Knowledge and Skill Requirements for Marketing Jobs in the 21st Century,” Journal of Marketing Education, 32 (3), 341-352. IV. Example Problem In the package and on my web site is the annual sales data from 1996 to 2011 for the Crystal Pure Bottled Water Company. You need to determine which sales forecasting method is most accurate and to give your sales forecast for 2012. You should check the following methods for accuracy: naive method, moving averages, exponential smoothing and the sales force composite. Your solution should include the following: 1] Your recommended sales forecast for 2012 and the justification for choosing that forecast. 2] A summary table of the methods you attempted, the forecast for 2012, and the accuracy of each method. 3] An appendix should be attached that contains your calculations. Solution (file is with the lecture and on my website). My recommendation is a forecast of 50,000 units for 2012. This recommendation is based on the sales force composite estimate which was the most accurate method over the past 15 years. I assessed the following 5 forecasting methods for their accuracy in the past: Naive method, 2 period moving averages, 3 period moving averages, exponential smoothing and the sales force composite. Summary Table Method 2012 Forecast Naive Method 47,325 2 Period Moving Average 43,373 3 Period Moving Average 40,279 Exponential Smoothing 46,475 (Alpha = .9) Sales Force Composite 50,000 MAPE 0.1681 0.2959 0.2939 0.1814 0.0544 Note that the Sales Force Composite forecasts have to be given, they are not calculated from the historical data.