Demantra Engine Tuning
Demantra Engine Tuning
Demantra Engine Tuning
Executive Summary
Many companies continue to struggle with poor forecast numbers even after implementing Demantra Demand Management (DM) or Advanced Forecasting (AFDM) modules. The promise of a more accurate forecast post- Demantra implementation remains unfulfilled as demand planners are forced to resort to manual overrides for large number of records leading to a lengthy forecast review process causing significant delays in the overall demand management process.
The most common reason for poor forecast numbers generated by Demantra is the engine not being tuned to take into account client specific data set requirements.
Ideally, during the implementation, a lot of time needs to be spent on analyzing the data set and configuring the engine parameters while keeping the client specific data model in mind. Unfortunately, it has been observed that Demantra engine tuning exercise is accorded least priority and often left for until after go-live period.
Many a times during the Demantra implementation, both the consultants as well as the business users are so focused on catering to requirements related to worksheets, series, workflows etc. that they tend to take better forecast accuracy out of Demantra for granted and ignore to do the due diligence to tune the Demantra engine.
Also, since Demantra engine tuning being a specialized skill; it requires indepth understanding of various factors that contribute towards better forecast accuracy and understanding of various engine parameters that need to be setup for better results.
Though this is a specialized area and should be performed by highly qualified and experienced consultants but, users of Demantra and demand planners should also be familiar with the different factors that can influence the forecast accuracy.
There are several factors influencing Demantra forecast accuracy but some of the most important ones are listed below:
Demand Data Profiles
Nodal Tuning
Causal/Promotions
Forecast Tree
Proport Function
Demand Data Profiles
The first step towards a better forecast out of Demantra is to know the various demand profiles that apply to the client's business.
The demand data pattern could be intermittent, regular, smooth etc. and the knowledge of these demand patterns to the various products would help setting up the Demantra use the right statistical model for forecasting.
Oracle Demantra utilizes different statistical methods and algorithms to project demand into future. Demantra DM model uses eight statistical methods whereas Demantra AFDM uses fourteen different methods for the statistical forecasting. Both Demantra DM and AFDM modules uses Bayesian approach for generating the final forecast for a specific item-location combination.
The Bayesian approach combines the results of individual models. Each model is evaluated, and each model in turn tests a number of subsets of system and user-supplied causal factors. All combinations of models and subsets of causal factors are assigned weights indicating their relevance. Every combination contributes to the final forecast according to its weightage.
Therefore, having an understanding of the demand patterns of your products could help you apply the correct forecast method to the item-location combination in Demantra that will improve the forecast accuracy considerably.
e.g. if you already know that there is a product line that exhibits intermittent demand patterns only, then turning off other forecasting models for this combination could significantly improve the forecast accuracy as the other forecasting methods will not contribute to the final forecast number.
The following forecasting models are used by Demantra:
Regression
Regression
Log (log transformation before regression)
CMReg (Markov chain selection of subset of causal factors)
Elog (uses Markov chain after log transformation)
Exponential smoothing
Holt
Bwint
Intermittent Models
CMReg for Intermittent
Regression for Intermittent
Croston
Time Series Models
ARX and ARIX
Logistic and AR Logistic
Other Models
BWint (a mixture of regression and exponential smoothing)
Nodal Tuning
One of the reasons for poor forecast accuracy for the clients using Demantra Demand Management (DM) module is that the statistical methods and algorithms apply either to all the combinations or not apply at all. There is no flexibility to choose statistical models specific to one particular combination different from the rest of the population even though the demand pattern exhibited by that item-location combination might be different from the rest of the combinations. This proves to be a major constraint during the forecast tuning exercise for the clients of Demantra DM module.
This constraint is overcome in the Demantra AFDM module which provides advanced analytics capabilities through Nodal tuning feature.
Nodal Tuning is a powerful functionality available in Demantra Advanced Forecasting and Demand Management (AFDM) module.
Nodal Tuning lets the demand planners pick and choose the statistical models that engine should apply to a particular item-location combination for generating the system forecast and also allow setting the engine parameters for that combination.
Nodal tuning also allows fine tuning the Demantra engine parameters specific to the combination.
This feature provides a tool in the hands of Demantra experts to fine tune the engine for better forecast accuracy. This feature along with the knowledge of demand patterns type as mentioned in the previous section would allow users to enable only those forecasting models that fit the demand pattern. This improves the forecast accuracy considerably.
Causal Factors/Promotions
One need to be very careful while modeling causal factors into Demantra. If the Data Model has many causal factors and promotions, they tend to dilute the base line forecast and result into a highly skewed forecast.
A good practice of introducing causal factors into the model is to first start with no causal factors and promotions data to generate a baseline forecast out of Demantra. Once the baseline forecast is tuned, other causal factors should be introduced one by one keeping in mind the effect of introduction of any causal factor to the baseline forecast.
This way effect of causal factors on the baseline forecast can easily be tracked and analyzed and anytime introduction of a causal does not seem to have desired effect, it should be turned off.
Forecast Tree
The forecast tree determines which item/location aggregation combination the engine will forecast at. The engine examines each level in the forecast tree and validates if there is enough sales history data available for forecasting or if the forecast generated have sufficient accuracy at that level. In case the validation fails, the engine moves on to next level and continue with the validation phase until it finds a level where it can generate a forecast.
In case the engine ends up forecasting at a higher level of aggregation in the forecast tree, the forecast is split to the lower levels.
Forecast tree is a system configuration that has a direct bearing on the forecast accuracy.
This is one of the first setups that need to be done after careful analysis of the sales history and after discussion with users. The forecast levels should be meaningful to the business users and it is recommended to have between 3 and 6 levels that the engine can traverse and forecast.
It is useful for the forecast tree to include the level on which accuracy is measured, if possible.
Proport
Proportions are very important and are used during the aggregation of the forecast from the lowest level to higher levels and de-aggregation of the forecast produced at the higher level to the lower levels.
The final output of the Demantra generated forecast could be very different depending upon the proportions.
The proportions are calculated and stored during the sales history data load. Several parameters control the calculation of proportions.
One of the parameters that influence the proportions is the amount of the sales history data that the system uses to calculate the proportions. The proportions calculated based on 12 months sales data would be different from those calculated based on 6 months historical data. Therefore, proper setting of this parameter is crucial to the calculations of the proportions which in turn influence the final forecast.
Conclusions
The Demantra engine tuning is a complex exercise and there is no one-fit-all solution for it.
A major engine tuning exercise should be undertaken every couple of years and whenever there is a change in the demand pattern of the products. The tuning exercise needs to be tailored to client specific Demantra implementation but having awareness of the factors that influence the forecast accuracy would go a long way in improving the forecast accuracy further.
All About Oxford Street Hotels In London Arsenal Vs Manchester City Live Epl Direct Tv Uk Broadcasting Energetic vehicle FC Vehicle Modular Offices Easy, Quick, and Affordable Why You Should Try Fraxel Laser Treatment For Removal Of Your Acne Scars Succeed With This Search Engine Marketing Tutorial Choose Best Tire And Good Quality For Your Vehicle The Freedom You Get From Cape Car Hire Meet Elvis, King of the Road And Customer Service Education Teacher Extraordinaire A Convenient Way To Travel With Airport Car Rental Service There Are Still Market For Bankruptcy Car Loans Enjoy Beautiful Cedar City Utah Even More with Satellite Broadband