A large software solutions provider, specializing in supply chain management and retail planning solutions, wanted to upgrade its legacy product solution and to build an autonomous supply chain.
The legacy IT product contained a forecasting framework that would generate time series forecasts with a high degree of accuracy provided there are no external influences. TheMathCompany partnered with the IT provider to introduce advanced AI and ML capabilities within the framework that accounted for external influences like unplanned events, extreme weather conditions, etc. to deliver accurate results.
TheMathCompany developed an ‘Intelligent Demand Forecasting’ framework that used a three-layered approach.
1. Baseline forecasts were generated using Lewandowski algorithm. An adaptive algorithm, it allowed for tweaking and fine-tuning the α, β, and γ (data smoothing factor, trend smoothing factor, and seasonal change smoothing factor).
2. The second layer was to identify short-term trends. Conventional algorithms result in high short-term demand sensing errors. Pattern recognition methods were leveraged to identify short-term trends.
3. The third and final layer was Impact Attribution. Impact of historical events on various products and stores was determined through attribute modelling. It was applied to forecasts with a feedback loop.
- Lewandowski algorithm improved baseline forecasts by about 0.5%.
- Pattern recognition improved the accuracy of identifying short-term trends to 96%.
- Application of historical impacts determined through attribute modelling through a feedback loop improved overall recommendation and accuracy.
A systematic breakdown of the forecasting framework and introduction of advanced capabilities allowed the legacy system to move towards becoming an integral part of the autonomous supply chain.