Adding intelligence in the Forecasting process to support Supply Chain Segmentation
Most advanced planning systems provide a suite of statistical forecast models and different parameters for the Forecaster to further fine-tune the models and in addition they also contain Best-Fit models that perform multiple iterations to choose the forecast model and its parameters that provide the lowest Forecast Error specified. There is however no easy way for the forecaster to assign these models to different segments that is available out-of-the-box. As a result, the Forecaster ends up reviewing the forecast models for every part every forecast cycle and then changes the assignment, given a change in the strategy for a particular segment.
I had discussed in my earlier blog the importance of evaluating the value added by the Forecasters by comparing the accuracy of their forecasts with that of a Naïve Forecast or with that of the results from a Best-Fit forecast algorithm that are available in most Advanced Planning systems. We had also looked in my other blog a key assumption that underlies most statistical forecast models - The pattern identified in the historical data will repeat in the future and hence can be extrapolated. We have seen that this however is always not the case as a result of both external activities [driven by Market, Competitor and regulatory and social environment] and internal activities [driven by Demand Shaping, Product Lifecycle]. There are means to identify structural shifts in demand patterns and the alert the forecaster accordingly, that we explored in the blog, where we had discussed the usage of one of the mechanisms - Tracking Signal. However, these are more reactive in nature and hence we see a need to incorporate the impact of these activities in the forecasting process, where the Forecaster can add immense value by providing the additional intelligence in line with the Supply Chain Segmentation Strategy.
We have seen the following approach as an effective means of incorporating the Supply Chain segmentation information in the Forecasting process:
1. As part of the initial implementation, Forecaster can develop a complete suite of different forecast profiles that are differentiated by the assigned Forecast Models [like Moving Average, Linear Regression, Trend and Seasonal, Croston, Shape Modeling] and the forecast parameters assigned [like history horizon, smoothing constants, periods for seasonality, trend dampening]. The forecaster creates these forecast profiles in close coordination with the entire Supply Chain team to achieve the different strategies set for different segments [for e.g. Products with responsive strategies need to react quickly to demand changes and hence will warrant a shorter historical horizon; Products that are in a particular lifecycle and have another lifecycle change in the forecast horizon will need to be forecasted using Shape Modeling algorithms; Products for the educational channel will need to factor in seasonality and this list can go on and on]
2. Once the Forecaster creates all the required Forecast Profile, the next task for him is to assign these Forecast Profiles to the respective SKUs and create a Forecast Strategy Matrix for them to be forecasted accordingly. These rules can also include an effectivity date, wherein post the effectivity date the default forecast profile assignment would kick in. These rules needs to be maintained centrally and should lend themselves for further automation within the Advanced planning systems to generate the statistical forecasts
3. Forecasters take note of the changes to the Supply Chain Segmentation strategies in the monthly S&OP meeting and then revisit the above table to assesses if they need to develop any new forecast profiles or if they can extend one of the existing forecast profiles to a different segment
4. As part of the Forecast review, the forecaster would also want the feedback from the Demand Planner [a different model chosen than the one the Forecaster had assigned] to be incorporated into this matrix. By introducing this closed-loop feedback cycle, the additional intelligence from the Demand Planners can be incorporated in the forecasts and thus help improve the overall forecast accuracy
While we are all aware of the basic principle of forecasting that the forecasts will always be wrong, there is merit in adding the business and market knowledge to the statistical forecast models and hence improve the overall effectiveness of the demand planning process. While most advanced planning systems offer a wide variety of statistical forecast models, they do not provide out-of-box functionalities to support this critical process of incorporating intelligence into the statistical models. We have helped some of our clients by developing custom solutions that automate this process and seamlessly connect with the backend processing in the Advanced Planning systems and thus help the forecasters to be more effective and productive in achieving higher Forecast Accuracy. I would like to hear from you on your experience and challenges that you face in this particular area.