构建多租户全球机器学习SaaS平台.pdf

1、 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.I S V 3 0 6Building Multi-tenant global ML SaaS platformSean StauthQlikGlobal Director,AI and Machine LearningVin DahakeAWSSr SA Manager,ISV PE(East)2025,Amazon Web

2、Services,Inc.or its affiliates.All rights reserved.Agenda1.Why forecasting at enterprise scale is hard2.The journey to a global ML SaaS3.Architecture Deep Dive4.Where we are today5.Lessons for ISVsQlik Predict 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.The Machine Learning Op

3、portunityEvery business runs on forecastsDemand,supply chain,staffing,and financials all depend on predicting the future.Accurate forecasting drives profitability,efficiency,and resilience.The opportunity is massivePredictive analytics is a$14B market growing 23%annually,reaching$40B by 2028.*Foreca

4、sting and demand planning represent a$3B+segment where most tools still miss multivariate complexity.Qlik is built for thisWith 40,000+global customers,Qlik is uniquely positioned to operationalize ML at scale.Qlik Predict brings GPU-accelerated multivariate forecasting to business users through a n

5、o-code SaaS platform.The Challenge:Why Time Series Forecasting is Hard 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Past Approaches Come Up ShortSignificant Feature Engineering Architecting lag features is time-consuming and labor-intensiveTraditional Approaches are Univariate

6、Miss real-world complexity,resulting in lower accuracy and limited usabilityExternal Inputs MatterDemand is inextricably linked to other factors and often left unaccounted 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Traditional Univariate Forecasting Fails Under Real-World Con