No announcement yet.

Best Forecast plan?

  • Filter
  • Time
  • Show
Clear All
new posts

  • Best Forecast plan?

    Hello Gents.

    Am not sure if this is the right forum for my question or not, and am sorry in advance if this post has to be moved somewhere else.

    We all know how important is forecast for us (buyers), to insure accurate flow of items and does indeed have a direct role in controlling inventory ..

    Till the moment, no forecast schedule has been 100% accurate- at least those I experienced. My question is: what is in your opinion the best forecast schedule and plan, procedures formulas best proved their efficiency for you so far.. Any advice in this matter will be highly appreciated!

    Best regards

  • #2
    Re: Best Forecast plan?

    Hi Hassan

    You are correct in saying that no forecast or demand plan is 100% correct. Forecast error represents an area where the improvement potential is generally very significant, typically between 30 to 60%.

    It is a crucial process because it directly drives all subsequent supply planning activities. If a demand plan is inaccurate, no matter how perfect the supply planning process, it will inevitably generate the wrong results, as it
    will not be aligned with actual demand.

    No matter how sophisticated the algorithm used to generate statistical forecast, there are two major reasons why the result can prove unrealistic and misleading:
    • First of all, historical sales data often are heavily polluted. Such pollution comes from different factors, such as promotions which are not explicitly recognized, abnormally large customer orders, stock-out situations during which demand existed but could not be fulfilled, and acceleration in the
    sales volumes at the end of the quarter because of heavy discounts granted to customers. If the impacts of these causal events are not taken out of the historical data, then the resulting forecast will obviously be wrong.
    • Secondly, many future events are likely to influence the demand pattern -- a price increase targeted for a certain date, a promotion that is not comparable to ones performed in the past, and changing market conditions.

    The difficulty is that all this information on future events or trends is not the
    property of a single department within the company, but rather comes from a large number of different individuals belonging to different functions and regions.

    At the end of the day, the combination of polluted historical data and lack of consideration of future causal events can severely limit the ability to generate good forecasts.

    • The first challenge is the ability to clean up historical data with regard to all causal events.
    • The second challenge is the ability to select the right forecasting technique. This is a challenge because there are many statistical algorithms available (mean average, double exponential smoothing, triple exponential smoothing, Box Jenkins , etc) and the identification of the right forecasting technique in a specific situation is not always obvious.

    The solution is to use the ‘focused forecasting’ logic (also known as “pick best”) which works the following way:
    • First, you select a set of forecasting techniques that have a potentially good fit with the demand pattern you want to forecast.
    • Second, a “forecast” is then generated over the recent past with each forecasting technique, for instance over the previous year. The result is a set of demand curves that can be compared to the actual sales.
    • Third, the focused forecasting logic then automatically identifies the demand curve that has the least deviation with the actual sales to select the most adequate forecasting technique that will be applied to forecast future sales.

    • The third challenge of the demand planning process -- and certainly the most complex one to solve --is to capture all intelligence influencing future demand. As mentioned earlier, information on promotions, pricing modifications, phase-in and phase-out of products, evolution of market conditions, evolution of the sales potential in major accounts, etc, are typically scattered across the company and actually belong to many different people in the organization.
    • The fourth challenge is the ability to compare all these inputs. This is a challenge because the information on future demand is generally presented in different “languages”.
    • For instance, a sales manager will look at future demand by region. His concern is not whether he is going to sell product X, Y or Z, or whether he is going to sell to customer A, B or C , but the sales quantity he will achieve in his territory.
    • A marketing manager will have a different perspective on future demand: as he is responsible for a certain line of products, his focus will be on product families rather than on territories or customers.
    • An account manager will look at future demand from yet a different perspective: here the focus is neither by products, nor by geography; it is by customers.

    These three examples among many others demonstrate that future demand needs to be represented along multiple dimensions. The question is: how do you manage to compare these inputs that are provided in very different formats?

    • Finally, the fifth and last major challenge of Demand Planning is to reconcile all these different perspectives on future demand, in order to generate a consensus forecast that will correspond to the single demand plan that will drive all supply planning activities.

    In summary, the demand planning reality is as follows: various people work at various levels of the dimensions using various time increments different data to create different plans.

    To reconcile all this “biodiversity” into one single demand plan, there needs to be a process to choose one demand plan expressed in one demand dimension and then to translate this plan in all the other dimensions at all levels of aggregation.

    This process of disaggregating and re-aggregating a demand plan along multiple dimensions requires the use of a multi-dimensional database, which is indeed an absolute prerequisite to perform a fully optimized demand planning process.


    I hope this give you something to think about. Also keep in mind that the moment you start executing your "perfect plan" you have to start re-planning to incorporate the actual deviation from your plan as time passes.

    So, it is actually not how well you plan, but how well you re-plan.
    David van der Walt


    • #3
      Re: Best Forecast plan?

      Thanks for the reply David. That was a very nice illustration for the difficulties facing forecast.
      You mentioned through your reply : "many statistical algorithms available (mean average, double exponential smoothing, triple exponential smoothing, Box Jenkins , etc)"

      whould you please suggest a source to have a detailed overview for these methods?

      Thank you


      • #4
        Re: Best Forecast plan?

        Hi Hassan

        The point I wanted to make is that the statistics is (IMHO) less important than the data, processes and procedures around the forecasting process.

        For more detail on the actual algorithms, I suggest either a session on Google, or speak to the sales guys at your friendly IT vendor.

        SAS, i2, Business Objects etc. all do forecasting products.

        David van der Walt