Demand Supply Gap in Business

An Introduction to Demand Forecasting

Demand Forecasting is a fancy topic not just because it gives you psychic powers to know the future demand but also because it is an important part of the supply chain process since all the subsequent phases of supply chain are dependent on it. A key part of supply chain planning involves demand planning and the associated demand forecasting process. Forecasting is one of the three components of an organization – Demand Planning, Demand Forecasting and Demand Management process. Demand Planning helps us to find what we should do to shape and create demand for our product (Production, Packaging, Pricing, Planning, etc.). Forecasting helps us to find the upcoming demand and confirm whether a plan is in place to deliver. Demand management helps us to prepare for and act on the demand when it materializes (Sales and Operation Planning).

Demand Supply Gap in Business

There are two types of forecasts, a top-down forecast and a bottom-up forecast. A top-down forecast essentially estimates total sales demand and then divides those sales value level by level until the Stock Keeping Units (SKU) is reached. A bottom-up forecast starts with the forecast at the SKU level and then aggregates those demand level by level to reach forecast at a company-level.

There are three levels within the demand forecasting process with its own time horizon and purpose. Strategic forecast is used for capacity planning, investment strategies and is done in yearly buckets. Tactical forecast is used for sales plan, inventory planning, etc. and is generally done in weekly, monthly or quarterly buckets. Operations forecast are used for production, transportation and inventory replenishments and is done in hourly or day buckets.

Delving deep into the methods of demand forecasting

Let’s dive deeper into the process of demand forecasting to know how the actual process of demand forecasting is carried out. A key concept in forecasting is that the forecasting must be done on an end-item product/finished goods level.

The Forecasting method or process can be divided into two types based on the approach taken for it, i.e., subjective or objective.  The subjective method is most often used by production and inventory planners. The subjective method is sub-divided into Judgmental, Delphi and Experimental. Objective methods are mainly Causal analysis (e.g. – Demand of Cold drinks is driven by temperature), Time-Series (e.g.- When the demand has certain patterns such as seasonality) and Machine Learning.

Now let’s consider some of the objective methods of demand forecasting:

Time Series Analysis

Time Series analysis is generally used when the dependent variable is a function of time. In Demand forecasting, the demand is a function of time hence time series analysis can be used. Time series is used for mid-range forecasts for products that have a long history of demand. A time series has the following components:

  • Level: Level is the value where the demand hovers. It is like the y-intercept of a straight line.      
  • Trend: It is usually the rate of change of demand over time. Trend can be explained by taking the analogy of a straight line. It is like slope of a straight line.
  • Seasonality: It is the repeated cycle around a fixed time period.  
  •  Cyclicity: It is a pattern which repeats in abrupt time intervals.
  • Residuals: It is the random variations that remains after removing the above components from a time series.

The time series model can be an additive or multiplicative model created by a linear combination of the above five components.

Let’s look into some simple time series models:

Naive Forecast:
This is the simplest method of forecasting in which the demand for the next period is forecasted to be the demand of the last period.

Ft+1 = Dt

Cumulative Forecast
In cumulative forecast, the forecast for the next period is the mean value of last periods demand. All the data are given equal weightage. It is a stable forecast but is not responsive. The forecast is given as:

n-period Simple Moving Average
In n-period simple moving average method, the mean of last n periods is used for getting the forecast. This method is used when the demand is fairly constant over time. The forecast is calm and stable. The forecast is given as                                           

n-period Weighted Moving Average

In n-period Weighted Moving Average method, more weight is given to the recent history than the past history. More responsive forecasts are obtained with this method. The forecast is given as:

Simple exponential smoothing

The general idea in exponential smoothing methods is that the value of the data degrades over time. It is used for stationary demand (i.e., the mean and variance of the demand is constant over time). It is used for time series with level only. 

α is the smoothing constant which lies between 0 and 1. The greater the value of α, more is the forecast responsive and volatile. The lower the value of α, stable is the forecast

Holt’s Model

Holt’s model is an upgrade over the simple exponential smoothing model. It considers both the level and the trend.

Forecasting metrics
There are two methods to validate the quality of the forecast: Bias and Accuracy. Bias is the ability to continuously over-predict and under-predict. Accuracy is the closeness of the forecast to the actual demand. The various methods of calculating errors is as follows:

  1. Mean deviation(MD):

MD = Summation of (Actual Demand – Forecast)/number of observations.

  • Mean Absolute Deviation(MAD):

MAD = Summation of |Actual Demand – Forecast|/number of observations.

  • Mean Square Error(MSE):

MSE = Summation of (Actual – Forecast)2 / number of observations

  • Root Mean Square Error(RMSE):

RMSE = Root (Summation of (Actual – Forecast)2)/number of observations

  • Mean Absolute Percentage Error(MAPE):

MAPE = (|Actual – Forecast|/Actual) X 100

Accuracy = 1 – Error metric

Process of demand forecasting

  1. Data Gathering: In data gathering process transactional data is collected (Sale-in data) in the required data format. Other data is also collected. (For e.g. – Predecessor-Successor linkages for price fluctuations, promotions data, etc.) Data cleaning is also done in this stage of the process.
  2. Data Quality Assurance: In data quality assurance, the gathered data is shaped according to the required forecasting level. The data is validated against the actual sales that has occurred. Consolidation of the data is also done in this phase. Data analysis is done to de-enrich the data (i.e., removal of outliers and promotions data).
  3. Building the model: In this phase of the process, Time series or Machine Learning models are prepared according to the data.
  4. Running the model: The model is run and parameters are tuned. Best fit model is chosen and ensemble method is used for model selection.
  5. Result analysis and validation: In result analysis and validation, accuracy of the forecast is calculated and the forecast is presented.

Importance of getting accurate forecasts

The following problem arises due to inaccuracies in forecast:

  1. Lost Sales: This generally occurs due to under forecasting.
  2. Inventory stock-outs: Stock-outs in inventory can occur due to under forecasting. Shortages in inventory can occur due to less quantity of items being forecasted as opposed to the actual demand.
  3. Excess and obsolescence of inventory: Excess and obsolescence of inventory can occur due to over-forecasting.

Leave a Reply

Your email address will not be published. Required fields are marked *