The existing forecasting system faced challenges in calculating the supply chain cost scientifically and therefore subject to individual assessment thus leading to higher forecasting time and manual errors. The Adjustments done to the base forecast took approximately 9 Man days every month to finalize supply chain costs.
The problem statement was approached with below three steps-
- Opportunity assessment: Data collection from MRF files extracted from Forecasting tool. Identified the various drivers impacting supply chain costs in collaboration with the client
- Data Validation and Scoping: Data validation happened over multiple iterations to verify the low accuracy numbers in for Supply chain cost Categories were narrowed down based on the consistency of data (free from erratic behaviour)
- Model Building:
a. Past 3 years of data was used in the mathematical modelling and was separated into
i. Training Window
ii. Back testing Window
b. Based on industry research, the team compiled a list of internal/external variables which would have a significant correlation between these accounts. Further exploratory data analysis was performed to establish their impact quantitatively
i. 8 algorithms are used and tested in the Back Testing window
ii. Best fit algorithms are selected dynamically
iii. The result obtained is compared with the first real values inside the testing window.
iv. Once finished, another iteration can start, increasing the historical data in one month as illustrated in this diagram
The solutions comprising of Time series Forecasting and Analysis, help the client in monthly P&L forecast submission at category level for Supply Chain costs accounts. The model was able to forecast the net revenue, trade discounts, production cost, distribution cost, supply chain cost, overhead cost with very high accuracy but was not very successful with Marketing cost which is not history dependent and very dynamic in nature. However, this was handled by creating a feature to capture upcoming promotions and campaigns at product hierarchy level to capture the marketing cost in forecasts.
