forecasting basf’s custom material demand using abc/xyz
TRANSCRIPT
Forecasting BASF’s Custom Material Demand Using ABC/XYZ Analysis
Team 7Travis Greene, Patrizia Mach, Triguna Ashin Wijaya , Li-Cheng Pan
↑Zur Angabe der Klassifizierung (VERTRAULICH etc.)
bitte dieses Textfeld im Folienmaster verwenden
BASF - chemical company
↑Zur Angabe der Klassifizierung (VERTRAULICH etc.)
bitte dieses Textfeld im Folienmaster verwenden
Business Problem
● Intermittent Demand● Forecasting, demand planning and inventory management is key
↑Zur Angabe der Klassifizierung (VERTRAULICH etc.)
bitte dieses Textfeld im Folienmaster verwenden
Forecasting Goal
Monthly 2-month ahead forecasts
Quantity of units shipped per month
December January February March
↑Zur Angabe der Klassifizierung (VERTRAULICH etc.)
bitte dieses Textfeld im Folienmaster verwenden
Project Workflow Overview
Remove negative demand
Aggregate into “Monthly” demand
Remove series without orders in test
period
1
2
3
Data Cleaning
Separate in ABC-XYZ
Get test/RMSE for every group
Choose group forecast method based on best
test RMSE
Refit full series and forecast based on
desired horizon
4
5
6
7
Grouping Series
Forecasting
Choose group forecast method based on best test
RMSE
Refit full series and forecast based on
desired horizon
6
7
↑Zur Angabe der Klassifizierung (VERTRAULICH etc.)
bitte dieses Textfeld im Folienmaster verwenden
Data Description
Training: 10/2012 - 08/2017 (59 months) Testing 09/2017 - 08/2018 (12 months)
Original data 108,324 rows, 826 materials, 73 months of data
After cleaning 58,575 rows, 825 materials, 71 months of data
↑Zur Angabe der Klassifizierung (VERTRAULICH etc.)
bitte dieses Textfeld im Folienmaster verwenden
Methods
Pareto Principle
ABC-XYZ Analysis
Importance
Fore
cast
abili
tyXYZ Analysis
ABC Analysis
Results
↑Zur Angabe der Klassifizierung (VERTRAULICH etc.)
bitte dieses Textfeld im Folienmaster verwenden
The best performing models tend to underpredict in January/April and overpredict in August
EvaluationETS test set forecast errors (A)
August ‘18
Jan ‘18April ‘18
↑Zur Angabe der Klassifizierung (VERTRAULICH etc.)
bitte dieses Textfeld im Folienmaster verwenden
Deployment and Maintenance
Deployment through Business Analytics Department on R Shiny server. Employees to log in and check forecasts regularly.
Business Analytics
Team
R Shiny Server
Factory Managers
Production Decisions
Maintain Log inReview Forecasts
AnnualRegrouping and re-evaluating forecasts of existing materialsAnd grouping of new material products
↑Zur Angabe der Klassifizierung (VERTRAULICH etc.)
bitte dieses Textfeld im Folienmaster verwenden
Recommendations
● Confirm whether 2018 order volumes correspond to behavioral pattern change
● ETS model underpredicts for January/April and overpredicts August
● Ensemble in current model limited to seasonal naive
● Capacity estimate 90% of historical high instead of actual facility threshold
Limitations
● Combine technical forecast with production facility experience
● Weight “importance” by unit cost● Tie in external material information
depending on nature of product e.g. steel price indices
● Experiment with additional forecasting techniques in development e.g. variants of Croston’s methods
● Explore 2-month test set period