Demand Forecasting is one of the three components of an organization’s demand planning, being the other two planning and management. Forecasting concerns the expect demand given a plan of action.
Which means, making a prediction on the size of demand expected given a marketing and business plan.
Forecasting is composed by three levels:
- Strategic Forecasts: capacity planning, investment strategies.
- Tactical Forecasts: sales plans, short-term budgets, inventory planning.
- Operation Forecasts: production, transportation, and inventory replenishment.
Each of these levels have a particular timing. The strategic is developed trough years, tactical takes weeks or months and operation only takes hours or days.
Demand Forecasting Methods
We can divide forecasting methods into subjective, usually used by marketing and sales, and objective, used by production and inventory planners. On subjective there are judgmental and experimental, and objective causal and time series.
Judgmental refers to sales force surveys, or opinions by experts that deal with those scenarios. The basic premise here is that someone knows the truth. Experimental refers to sampling local and extrapolating. For example, focus groups or marketing tests.
On the objective side, casual means that there is a underlying relationship ou reason that can be perceived. Time series, on the other hand, relates to historical patterns in demand.
All these methods have their uses and functions. You need to evaluate each scenario to identify which one is the best for your purpose. But regardless of the chosen method, you need to secure the quality of your forecast.
To make sure your forecast is optimized, there are two key concepts to pay attention. Bias, which means a tendency to over or under predict and accuracy, or the closeness to the real value of observations.
A good habit is combining more than one method, because none of them can capture both dimensions. So, if you can utilize more than one, you will be more secure that your data is not biased and have a good accuracy.
Key aspects of Demand Forecasting
Whenever you need to use forecasting models, you must keep in mind that forecasts will never be perfect. Which means that you shouldn’t rely only on points of forecasts.
You can incorporate ranges and track past errors so you can perceive drifts and changes in your demand. With past errors you have a better perception on how close your forecast can be.
Another thing you must consider is the time horizon of your prediction. Shorter forecasts are more accurate than longer. That happens because it’s easier to predict events that happen closer to us.
A good idea is trying to postpone the forecast to a closer date. That results in better understanding and more accurate data.
Demand Forecasting techniques
There are a lot of techniques to make demand forecast, we will talk about the most used ones.
Time Series Analysis
It’s a very popular technique, especially for mid-range forecasts. As it makes predictions based on historical demand, it needs a lot of data and records. We can cite three major models:
- Cumulative: Every data is included. The result is calm, and changes are slow. But is a very stable model.
- Naïve: Only the latest data matter. The result is a nervous and volatile forecast. This makes the model very responsive.
- Moving Average: The data included is selected. This is a middle ground between the two other models.
This model will treat data differently depending on its age. The basic idea of the model is that data lose value over time, then you need to weight more newer observations than old ones.
To determine the value of data, the model uses the alpha factor. It will set a weight from newest information to oldest. Then, the alpha indicates:
- If alpha is bigger than one, the forecast is nervous and volatiles.
- If alpha is bigger than zero, the forecast is calm and stable.
There are a lot of products and forecasting techniques and they differ from one to another. You need to evaluate your demand to find the best suited one, keeping in mind that a good forecast should not be biased and have a good accuracy.
As we said, to do this the best way is to combine more than one model. Most models need history to rely on. If your demand doesn’t have historical data, you can search for similar products and try to establish a strategy based on that.
So, in sum, demand forecasting is part of demand planning and management process. Range forecasts are more precise than point ones, aggregated have more accuracy and less biased and finally, short time horizons are better than long ones.
With an inventory optimization tool you will be able to define the best stock policy per product with intelligent rules. Get to know Supply Brain.