Distinguishing Forecast and Demand with Examples

Followings are parameters distinguishing them conceptually:

1. Scope
Demand is very specific, being quantity from a single customer also, whereas, the forecast is cumulative and considers more than one customers. Here, in the case of Forecast, customer might be known or unknown party while there will be a specific customer with actual demands. Example: For Bosch India Limited receives separate Demands for Fuel Injectors from Maruti, Hyundai or Mahindra. However, calculated forecast will be cumulative and grouped item model wise and not customer wise. If there are 3 models for Fuel Injectors, A, B and C; demand and forecasts would be represented by following way: Table-1: Forecast (Dec 2011) Model A B C Table-2: Demand (Dec 2011) Maruti Model A B C A B C A B C Units 42 9 15 28 11 18 25 12 30 Units 90 30 60

M&M

Hyundai

2. Inputs
Major Input for any forecast is Historical Data. Historical Data, here, means actual Demands received in the past. Apart from this Historical Data might include: Example: To determine the forecast for the month of August 2011 with moving average method, actual Demand of month of May, June & July would be considered. Table-3: Month April May June July August Demand 160 150 140 140 150 Forecast*

150 143

Note: * Forecast calculated by Moving Average with previous 3 months Demand input

3. Accuracy
Demand being the actual orders from the customers, are very accurate. However, Forecasts are, though systematically estimated but still, assumptions. So accuracy of Forecast is comparatively less than Demand. This less accuracy is measurable and that becomes one of the inputs for upcoming forecast. Mean Absolute Error (MAE) is one of the methods to measure the accuracy of Forecast. Example: There are two sets of monthly forecasts for Diesel Engine. Both are based on same demand data. But both have different Accuracy. Mean Absolute Error (MAE) for Forecast-A is 13 units per month and for Forecast B, it is 8 Units per month. Hence, Accuracy of Forecast B is more than accuracy of Forecast A. Table 4: Month Forecast A Forecast B Demand Jan 150 145 140 10 5 Feb 140 150 160 20 10 Mar 160 165 170 10 5 Apr 170 155 160 10 5 May 180 170 150 30 20 Jun 160 180 170 10 10 Jul 150 170 160 10 10 Aug 140 145 150 10 5 Sep 170 175 180 10 5 Oct 180 170 160 20 10 Nov 150 145 140 10 5 Dec 160 155 150 10 5

- A* - B*

Note: * indicates Absolute Error or Absolute Deviation

4. Risk
Being less accurate compared to Demand, Forecast carries a specific Risk. Demand being confirmed orders from Customer/s, carries almost no risk, as far as deviations are concerned. Amount of risk associated with Forecast would reflect in terms of Excess Inventory in case of overestimated Forecast or would reflect as stock outs i.e. loss of customer service level in case of underestimated Forecast.

5. Tolerance
Forecasts are always wrong This is one of the principles of Forecast. And because of this reason forecasts are always with a Tolerance. There will be upper limit and lower limit for forecast to keep it flexible. Demands are exact quantity required by customer and so don t have such Tolerance.

6. Method or System Dependency

Forecasts are determined by various techniques, broadly grouped at Qualitative Techniques and Quantitative Techniques. And so by using different methods, we may end up getting different forecast even for same SKU for the same time frame. It is important to determine best match technique to determine forecast. This can be achieved during the course of time by evaluating various methods. Example: Here two forecasting methods are used with same Historical actual demand data. There is significant deviation between forecast by Method A and Method B. Forecast-A is moving average method with last 3 months demand data and Forecast B takes present month s actual demand as forecast for next month.

Table- : Month April May June July August September October November December Demand 160 150 140 140 150 160 170 140 150 Forecast A Forecast B 160 150 140 140 150 160 170 140

150 143 143 150 160 157

7. Dependency
Forecasts are dependent on actual demands of previous period, whereas demands are never independent from Forecast of same time frame. Exponential Smoothing Method uses forecasts of previous periods along with Actual demand of previous periods. Hence, Forecasts are dependent not only on Actual Demands but previous Forecasts also.

8. Demand is result of External Factors

Customer s forecasts are major external factor. And Forecasts are results of many factors like Market Trend, Seasonality, Competitors Strategy; Companies own Marketing and Promotional Activities, Government Policies and Global Business Environment.

9. Data Storage
Actual Demands need to be captured and stored systematically to establish Demand Trends and eventually to determine future forecasts. Forecasts need not to be stored to determine future forecasts for few Forecasting Techniques, say; Exponential Smoothing needs only previous month s forecast only. However, to evaluate the effectiveness and accuracy of Forecasting Technique, previous forecast data are required and are stored for that purpose. See: Accuracy of Forecast

10. Effect on Planning Process

Forecasts are inputs for planning at every level. The accuracy of forecasts needs to be increased as we do deeper in planning.

Forecasting
To define with, Forecast is a systematically estimated or calculated assumption of certain quantity of Goods to be sold in market, Whereas, Demand is the actual quantity desired by customer in form of Order.

Product Performance Competitors Market Trend

Globalization

Forecast

Govt. Policy

Customer

Planning

Capacity Planning

Inventory Planning

Logistic Planning

SCM

Right Product At Right Time In Right Quantity At Right Place

Fig-1:

Above diagram shows factors affecting forecasting and importance of Forecasting on Planning activities in Supply Chain Management.

Factors to be considered qualitatively and/or quantitatively for determine Forecast: 1. Customer How much customer will need? This can be either answered in terms of Customer actual demand or/and forecast of customer. There are systematic approaches to estimate what customers would need even before actual order comes. They are called forecasting Techniques. Most important input related to customers is buying pattern. Buying patterns of the Customer can be derived by tabulating or plotting actual demand data on time scale. Example 1: For Cummins Diesel Engine Business Unit, say there are two types of Customers 1. Original Equipment Manufacturers e.g. 2. Dealers and Distributors Month Customer Dealers OEMs Jan 180 140 Feb 120 150 Mar 130 160 Apr 140 140 May 180 170 Jun 120 140 Jul 130 170 Aug 140 150 Sep 180 160 Oct 120 140 Nov 130 140 Dec 140 150

400 350 300 250 200 150 100 50 0 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Customer OEMs Customer Dealers

Here as u can see, For OEM customer there is no pattern which can be noticed but however it s evident that after maximum order the next order is of minimum quantity. For Dealer/Distributor customers, it s evident the there is cyclic trend of demand from them. However, by combining all type of customers buying patterns we can conclude that the peak demand arrives in the month of May and September every year. This is just one set of data and actual data would be huge and complicated. For that various statistical models are being used. They will be discussed in further part of this doc.

Example 2: Another case would be for forecasting demand of Spares/Replacements/Repairs. For any Diesel engine, data about life of each components of engine would be readily available. The source of data would be Design team or After Sales Service Team. From this available data we can derive a formula which would give idea about approx. requirement of Spare parts after certain duration of Operation of Engine. This formula can be as simple as below.

Spares for year (Y4) for Component X = 10% of Sales of Year Y3 + 20% of Sales of Year Y2 + 30% of Sales of Year Y1
Let s assume that the forecasts for total sales for coming months are calculated and fixed. Forecasts for components A, B, C . can be calculated as below.

Month Sales Component A Component B Component C

Jan 1200

Feb 1400

Mar 1500

Apr 1300 790 820 850

May 1400 850 840 830

Jun 1500 850 840 830

Jul 1400 820 840 860

Aug 1300 860 860 860

Sep 1100 860 840 820

Oct 1300 790 760 730

Spares Expected
NA NA NA NA NA NA NA NA NA

Example 3: Companies geographically divide any markets in to zones and then forecast demand zone wise. It increases the forecast accuracy. Such forecasts further can be divided in to distributors territory level and then product level forecasts can be determined.

Overall Market Trend

Year Engines 2011 130000 2012 160000

For Year 2012

Zones Actual Demand of 2011 36400 26000 42900 15600 9100 130000 Forecast for 2012 44800 32000 52800 19200 11200 160000

For South Zone (For Year 2012)

Engine Model B Series ISLe C Series B Gas Total Actual Demand 1716 1287 858 429 4290 Forecast 2112 1584 1056 528 5280

North West South Central East Total

2. Competitors Today s markets are rarely monopolistic for any product or service. There are lots of substitutes for any products and services. Even if the national level customer demand for a product is assumed, Company would never forecast considering only that. Here, competitors performance on sales comes in to consideration. Example: Some market research company would claim demand of 2, 00,000 Passenger cars in India for the year 2012. Company should analyze their forecast considering their market share in Country.

Year Total Commercial Vehicles % Market Share Forecast

2005 362755.00

2010 722199

2015 1081643 45% 486739

25% 90689

35% 252770

3. Market Trend Over the time period, demand of any product has a specific trend. This might be increasing trend or decreasing trend. For example let s refer data of Indian automobile over last decade: YEAR 2010 2009 2008 2007 2006 2005 2004 2003 2002 2001 2000 1999 CARS 28,14,584 21,75,220 18,46,051 17,13,479 14,73,000 12,64,000 11,78,354 9,07,968 7,03,948 6,54,557 5,17,957 5,33,149 % CHANGE 29.39 17.83 7.74 16.33 16.53 7.27 29.78 28.98 7.55 26.37 -2.85 COMMERCIAL VEHICLES 7,22,199 4,66,330 4,86,277 5,40,250 5,46,808 362, 755 3,32,803 2,53,555 1,90,848 1,60,054 2,83,403 2,85,044 % CHANGE 54.86 -4.1 -9.99 -1.2 50.74 9 31.25 32.86 19.24 -43.52 -0.58 TOTAL 35,36,783 26,41,550 23,32,328 22,53,999 20,19,808 16,28,755 15,11,157 11,61,523 894796 814611 801360 818193 % CHANGE 33.89 13.25 3.35 10.39 19.36 7.22 23.13 22.96 8.96 1.62 -2.1

From above data it s evident that demand and production for Automobiles in India is increasing. However, for every year monthly demands will not be steady and cannot be extrapolated on basis of above data. Further to subdivide, demand for automobiles by a particular manufacturer cannot be decided. And at even further level, demand for any model of an Auto Co. cannot be derived with help of above data. But however, the increasing / decreasing trend of demand would be considered while preparing forecasts through statistical model. Market Trend here, also, means fashion or changes in customer preferences due to technology. Example: B/W Display mobiles demand went almost nil because of Technology of Color Display mobile phones. Same way these days, Touch Screen and QWERTY keyboard smart phones are significantly reducing the demand for other types of Mobile phones. Another example would be for apparels. 4. Globalization In today s world, product manufactured in one country, need not to be sold in same country. Healthy international trades have increased multiple sources of demand to be considered to calculate the forecast for any company. At the same time, for the company which is operating and selling products in same country must consider all possibilities of imported products of their competitors. 5. Product Performance Success or failure of any product gives idea about forecast for nearer future. If product is very successful and delighting customer, there will be rise in demand. And if product fails to give value for money or customer satisfaction, the demand would fall rapidly.

6. Government Policy Import Duty on specific Goods, Trade agreements with certain countries, statutory prohibition of certain products, Subsidy or Tax benefits on products etc. are components of Government Policies which actually determines sales of particular product in any market. They have to be considered while preparing the Demand forecasts. However, all of these components need not to be affecting any product demand.

Forecasting Techniques
Broadly speaking, there are two approaches to demand forecasting- one is to obtain information about the likely purchase behavior of the buyer through collecting expert s opinion or by conducting interviews with consumers, the other is to use historical data as a guide through a set of statistical techniques. Both these methods rely on varying degrees of judgment. The first method is usually found suitable for short-term forecasting, the latter for long-term forecasting. There are specific techniques which fall under each of these broad methods. These forecasting techniques are classified into following two categories: 1. Qualitative Forecasting 2. Quantitative Forecasting

Qualitative Forecasting Technique

These methods are used when little or no historical data are available for example in the case of demand forecasts for new products like Revolo technology based automobile performance enhancement Kit . Popular Qualitative Forecasting Techniques are Delphi Method, Market research, Life Cycle Analysis & Judgment methods. These methods rely on expertise and intuition, rather than on statistical analysis of historical data. Such methods are particularly useful when historical data is scarce. The principal, and very significant, advantage of qualitative forecasting techniques is their potential for predicting changes that can occur for future demand. Time series quantitative techniques cannot predict changes in sales or demand patterns. Regression cannot predict changes in the relationships between sales and the predictor variables. Predicting the occurrence and nature of these changes can be accomplished by qualitative analyses based on the knowledge and experience of people internal and/or external to the company. This is valuable by itself or as additional information to be utilized to adjust the quantitative forecasts. There are few difficulties with Qualitative forecasting Techniques over Quantitative Techniques: 1. The ability to forecast accurately can be reduced when forecasters only consider readily available and/or recently perceived information. 2. The ability to forecast accurately can be reduced by the forecasters inability to process large amounts of complex information. 3. Accurate forecasts can be difficult to produce when forecasters are overconfident in their ability to forecast accurately. 4. The ability to forecast accurately may be reduced because of the forecasters tendency to infer relationships or patterns in data when there are no patterns. 5. The ability to forecast accurately can be affected by anchoring; that is, forecasters may be influenced by initial forecasts (e.g., those generated by quantitative methods) when making qualitative forecasts 6. Qualitative forecasting techniques are expensive and time intensive.

1. Sales Force Composite

The sales force composite is a qualitative forecasting method that uses the knowledge and experience of a company s Qualitative Sales Forecasting salespeople, its sales management, and/or channel members to produce sales forecasts. The grass roots approach to a sales force composite accumulates sales forecasts for the regions, products, and/or customers of individual salespeople. The sales management approach seeks sales forecasts from sales executives and is essentially a jury of executive opinion, even though consisting of a narrower range of executives (i.e., only sales executives or only sales and marketing executives). The distributor approach to the sales force composite solicits the sales predictions of independent distributors of a company s products.

2. Delphi Method:
The Delphi method consults a panel of experts in arriving at the forecast & proceeds through a series of rounds. Each expert is asked to make individual predictions based on available data and a final report is compiled with the combined consensus of the experts. The Delphi method is an iterative process used to collect and distill the judgments of experts using a series of questionnaires interspersed with feedback. The questionnaires are designed to focus on problems, opportunities, solutions, or forecasts. Each subsequent questionnaire is developed based on the results of the previous questionnaire. The process stops when the research question is answered, means when consensus is reached, theoretical saturation is achieved, or when sufficient information has been exchanged. The Delphi method has its origins in the American business community, and has since been widely accepted throughout the world in many industry sectors including health care, defense, business, education, information technology, transportation and engineering. There are many approaches to execute Delphi method. One of them is described below:

Round 1: First Questionnaire:

In Round One, the questionnaire defines the problem or issue and asks each expert to list as many responses (ideas, solutions, approaches, etc.) as possible. If this is a mail survey, the response form simply lists the question and provides blank lines for responses. If using email, the message is designed to allow input after hitting the reply email option. An example question might be, what actions could our committee take to improve retail sales in the downtown district? List as many as you can, using just a few words or a phrase. Or what should be our theme for this year s festival? List as many ideas as you can, using just a few words or a phrase. Participants respond anonymously.

Compiling Responses:
The communications manager compiles all the responses and creates the second questionnaire, with space for participants to respond to each idea.

Round 2: Second Questionnaire:

The second questionnaire includes all the responses, and asks participants to evaluate each idea. Participants are asked to clarify or add to ideas, comment on the feasibility of ideas, brainstorm additional strategies to implement ideas, and suggest new ideas. Participants respond anonymously.

Compiling Responses:
Again, the communications manager continues to develop the list of ideas, which now includes comments, additions, clarifications, and strategies. The communications manager develops the third questionnaire, with the additional information provided in Round Two.

Round 3: Third Questionnaire:

The communications manager repeats the process of compiling information, sharing it with participants, and collecting feedback. The third questionnaire may ask respondents to rank ideas in order of importance, in order of timeliness, or other criteria such as my willingness to work on this project.

Compiling Responses:
This is the final round of compiling responses, unless the planning team decides that participants need additional rounds of input and feedback.

Resolution and Report:

By this round, the feasible ideas have been identified, and set in priority order by participants. The communications manager responds to the group with the ideas or strategies, with details of implementation, arranged in priority order. Delphi has the added advantage that it works as an informal, subjective model when the decisions are based on opinion, and can be directly converted to a formal model, when the data is more knowledgebased.

3. Market Research:
Market research helps to estimate the size of market for products or technologies with few selected potential customers. Size of market might be a state or a larger geographic zone of country or even at continent level. There are basic two types of population on which market research can be done. 1. Consumers (End Users) For the case of Cummins, these would be industries using their products or OEMs using their products as component of final products assembly. 2. Distribution Channel Partners These partners might be Distributors, Dealers or Wholesalers. Judgments and ideas of such partners are acquired in form of a Research activity and analyzed to get input for forecast.

4. Life Cycle Analysis:

Products go through a life cycle of introduction, growth, maturity & decline. Based on the experiences of similar products in the past, one can make decision.

5. Informed Judgment:
This forecast is made by an individual or a group based on experience & understanding of the situation. The factors to be considered would be few or all of following: y Past Demand The data available is not being used as input to any statistical model. But it s being referred to make judgments of forthcoming demands. y Lead time of product Lead time here is time market from production. Longer the lead times lesser the forecasting accuracy. And so judgments are made accordingly. y State of the economy In growing economy, industries grow and demand grows. Such growth gives idea about increasing or decreasing future demands. y Planned price discounts If some products are being planned to be marketed at lower price range compared to competitors or customers estimated prices, they are likely to pick up certain increase in the demand.

6. Scenario Writing:
Scenario writing consists of developing a conceptual scenario of the future based on a well defined set of assumptions. Different assumptions will give different scenarios. After several different scenarios have been developed, the decision maker determines which is most likely to occur in the future and makes decisions accordingly.

Quantitative Forecasting Techniques

Extrapolative Method 1. Simple moving average This method averages the last number past years (any time horizons) data of a time series. It is appropriate only for very short or very irregular data sets, where features like trend and seasonality cannot be meaningfully determined, and where the mean changes slowly. To forecast of annual sales of Total Diesel Engines from Year 2010 to Year 2016, 2 yearly and 3 yearly moving averages can be calculated by following way. Moving Average 2 Monthly
NA NA

Month Jan Feb Mar Apr May Jun Jul Aug Sep

Demand 160 150 140 140 150 160 170 140 150

3 Monthly
NA NA NA

155 145 140 145 155 165 155

150 143 143 150 160 157

2. Exponential Smoothing This is another form of moving average method. It involves parameters reflecting the level, trend and seasonality of historical data, usually giving more weight to recent data. This method is widely used because of its simplicity, accuracy and ease of use. This method is useful even when historical data are few or inaccurate. 3. Autoregressive moving average (ARMA) aka Box-Jenkins It is an even more complex class of moving average models, capable of reflecting autocorrelations inherent in data. It can outperform exponential smoothing when the historical data period is long and data are nonvolatile. But it doesn t perform as well when the data are statistically messy.

Explanatory variable methods 1. Regression analysis Fitting a curve to historical data using a formula based on independent variables (explanatory variables) and an error term. Although these methods are relatively simple, and are helpful both in analyzing patterns of historical data and for correlation analysis, they are not generally recommended for forecasting. They have performed poorly in forecasting competitions. 2. Predictive modeling An area of statistical analysis and data mining, that deals with extracting information from data and using it to predict future behavior patterns or other results. A predictive model is made up of a number of predictors, variables that are likely to influence future behavior. 3. Econometric modeling Systems of simultaneous equations to represent economic relationships

Simulation modeling methods Cell-based modeling Modeling of individual homogeneous units (cells) over time, such as age/sex cells in pension forecasting these models are usually deterministic. They are useful to model large systems.

Once these forecasts are ready, they become key inputs for Planning Process: Production Planning Short range Forecasts - 3 6 months Forecasts for Production planning are most accurate than any other forecasts. They are made on SKU level. These forecasts are specific to individual production facilities. Operational level Decisions are made on basis of such Forecasts. Logistic Planning Medium range Forecasts 6 months to 2 years Forecasts for Logistic planning are of medium accuracy level. They are made on Product Family level. This forecast provides inputs for distribution channels related Tactical Level Decision. Decision about warehouse location and load per distribution channels can be taken with these forecasts. Capacity Planning Long range Forecasts 2 to 6 years Forecasts for Capacity planning are least accurate (generally) compared to other forecasts. However they are critical because of Strategic Level Decisions are made on these forecasts. Questions like to expand or to reduce the capacity are answered.

Collaborative Planning, Forecasting & Replenishment (CPFR)

This is a Business practice involving Manufacturer, Supplier, and Distributors and even ends Consumers in Planning, Forecasting and Inventory Management Activities. It creates WIN-WIN situation between every pair of Supply chain Partners. WIN-WIN situation means buffer inventory will be shared by both parties in the pair. Forecasting data base will be common. Everyone s inputs are considered and expected regularly. There are 3 types of CPFRs practiced in present Supply Chain Managements. 1. Basic CPFR: Here only a limited number of business processes integrated between a limited numbers of supply chain partners. Example: Information sharing between Fortuna Engineering, Nashik and M&M Nashik is example of Basic CPFR. 2. Developed CPFR: Here, it will typically involve a greater number of data exchanges between two partners, and may extend to suppliers taking responsibility for replenishment on behalf of their customer. Example: Information Sharing between MUL and FAG Bearings 3. Advanced CPFR Here, it goes beyond data exchanges to synchronise forecasting information systems and coordinate planning and replenishment processes. Information and Communication Technology are keys for this type of CPFR. EDI and RFID are technologies being used here. Example: Wallmart and P&G.