Applications of Data Science in Microsoft Cloud Products
0 likes81 views
The document outlines the applications of data science in Microsoft Cloud products, focusing on various roles such as data scientists, machine learning engineers, and product managers. It emphasizes the data science lifecycle, including phases like project kickoff, data analysis, modeling, and deployment, while highlighting the importance of customer-centric approaches and experimentation. Key topics include analytics, machine learning models, customer segmentation, and the necessity of data quality and model explainability in enterprise settings.
Applications of Data Science in Microsoft Cloud Products
- 1. Applications of Data Science in Microsoft Cloud Products Lisa Cohen MSFT Cloud Data Sciences linkedin.com/in/cohenlisa
- 2. About me Applied Math Bachelor & Masters VS Languages & IDE PM Shipped 3 major releases VS Telemetry Mgr Product & business analytics Principal Data Science Leader Cross-functional org leader Experimentation, Machine Learning, Data Science, Data Vis, Prod Mgmt E2E Azure business, Help customers succeed on the cloud Sr Community PM Strategy, Communications, Keynote speaker UW Data Science Board member Author & Speaker Data science and career topics linkedin.com/in/cohenlisa medium.com/@lisa_cohen Community leader: Data Science, Women, Mentors, Managers, Book club
- 3. 3 Agenda • Ensuring success from the project kickoff • Data Science in action: • Analytics • Experiments • ML models • Enterprise considerations
- 4. Data Science at Microsoft Data Scientist Analytics & Inference Statistical analysis & experiments Machine Learning Scientist/Engineer Production Models Develop predictive & prescriptive models, MLOps Data Engineer Data Platform & Pipelines Build the data platform Product Manager Planning & Stakeholder Engagement Manage the data science process
- 5. Data Science Lifecycle Problem & Hypothesis Design Approach Data Acquisition & Exploration Analysis & Predictive Modeling Evaluation & Reviews Deployment & Socialization Data Science project kickoff: Analysis, Experiment, Model What new capability will this enable? What decision/action will you take? What’s the expected impact? Deliver what users want (vs what ask for) Leverage the power of data science
- 6. Analytics problems State of the business Exploration, distributions, data insights Define funnel, calculate LTV Establish goals (OKRs) Trends and root cause analysis Retention curves Cohort segmentation 48.3% 29.5% 22.0% 19.5% 15.0% 6.7% <1 1 2 3 4 5-6 Tenure (in years) Cohorting (Churn sample data) Confidence intervals (CSAT sample data) Funnel stages (Free account signup) May 2019 Mar 2019 Jan 2019 50% 60% 70% 80% 90% 100% 0 1 2 3 4 5 6 7 8 9 10 11 12 Retention Month Retention curves by cohort start month
- 7. Experimentation & Causal Inference Accelerate learning and innovation Online, application and program experiments Designing experiments Outcome metrics, control group, power analysis, duration Analyzing experiments Tracking primary and secondary metrics Considering bias, ethics and fairness Causal inference scenarios Usage Time User Engagement Control Treatment EXP Platform MS Research Hypothesis Design Experiment Run Experiment Analyze Experiment Act on Results
- 8. Understanding the customer Customer Segmentation Customer Journey & Service Combinations Awareness Doc Recommender, Attribution, Search Marketplace Search & Recommender Buy Lifetime value Azure Cost Management Engagement Service/Solution Recommender Anomaly Detection Nurture Program Recommender Support ticket classification & prioritization Retention Predict churn risk & support needs Acquisition First workload Propensity models Customer-centric approach, aligned with lifecycle stages: Machine Learning Models Predictive and prescriptive models, with actionable insights & recommendations Internal and external facing Production vs learning
- 9. The data science process for machine learning Business understanding Modeling Data acquisition & understanding Deployment Scoring, performance monitoring, etc. Data source On premises vs. cloud Database vs. files Pipeline Streaming vs. batch Low vs. high frequency Environment On premises vs. cloud Database vs. data lake vs. … Small vs. medium vs. big data Wrangling, exploration, & cleaning Structured vs. unstructured Data validation & cleanup Visualization Feature engineering Transform, binning Temporal, text, image Feature selection Model training Algorithms, ensemble Parameter tuning Retraining Model management Model evaluation Cross-validation Model reporting A/B testing Intelligent applications Web services Model store Customer acceptance https://docs.microsoft.com/en-us/azure/machine-learning/team-data-science-process/lifecycle
- 10. Data Science in the real world Value of data quality, data cleaning and data set improvements Skewed populations Model reliability, SLAs, data contracts, maintenance Model explain-ability and simplicity Done is better than perfect (consider time to market) Tuning models for business use cases Socialize and integrate into existing tools Causation vs correlation
- 11. 11 Summary Project Kickoff • Clarify business goals • Determine the decision or action Data Science Products • Analytics • Experiments • ML Models Data Science in the Enterprise • Data quality • Model explain-ability
- 13. © Copyright Microsoft Corporation. All rights reserved.
Editor's Notes
- #7: Telling the state of the business in numbers Illustrating the business scenario and modeling in the funnel
- #12: https://pixabay.com/photos/raise-challenge-landscape-mountain-3338589/
- #13: https://www.flaticon.com/free-icon/linkedin_174857 https://medium.com/@Medium