预测机器学习：联合利华如何改进预测.pdf

1、2024 Databricks Inc.All rights reservedPredictive ML:Predictive ML:How Unilever is How Unilever is improving its improving its forecastingforecastingBill Tsiligkiridis,Data Science Lead,Unilever EuropeBill Tsiligkiridis,Data Science Lead,Unilever EuropeOzge Koroglu,Data ScientistOzge Koroglu,Data Sc

2、ientistGaius Noordhoek Hegt,Data ScientistGaius Noordhoek Hegt,Data ScientistPatrick Van Dalen,ML EngineerPatrick Van Dalen,ML Engineer12024 Databricks Inc.All rights reserved What we are trying to solve for Overview of solution Working towards sustainable maintainance Key Learnings Q&A2AgendaAgenda

3、2024 Databricks Inc.All rights reserved2024 Databricks Inc.All rights reserved3ForecastingForecasting(Its difficult business)(Its difficult business)2024 Databricks Inc.All rights reserved4Examples of typical Examples of typical forecasts in CPGforecasts in CPGDemand VolumeTrade investmentSupply Cha

4、in costsMarket GrowthPrice GrowthMarket ShareBusiness GroupCategoryBrandSKUCountryChannelCustomerVarious levels of granularityExternal forecastsInternal forecasts2024 Databricks Inc.All rights reservedVery expensive to generate at scaleAlso expensive to update frequentlyInconsistent set of assumptio

5、nsHuman BiasHistorical data availabilityForecast dependencies&compounding errorsSolution maintainanceExplainabilityManual/Expert forecastsManual/Expert forecastsAutomated/Machine Learning forecastsAutomated/Machine Learning forecasts5The challenges of forecasting The challenges of forecasting Either

6、 approach will have its own set of challengesEither approach will have its own set of challenges2024 Databricks Inc.All rights reservedEasy experimentationDeployment processSustainable maintenanceEasy experimentationDeployment processSustainable maintenance6Key Technical RequirementsKey Technical Re