Microsoft IT Showcase Optimizing predictive analytics using Cortana Intelligence Suite on Azure To improve the effectiveness of marketing campaigns, Microsoft IT developed a predictive analytics platform on Microsoft Azure that delivers customized models for a broad user base. With the power of the Cortana Intelligence Suite, data scientists and marketing teams are able to use a common platform to glean deep insights from customer profiles and enable seamless delivery of predictive model results into downstream sales and marketing automation systems. Business data at Microsoft Microsoft IT is continually working to develop logical and business-based methods to store and retrieve immense amounts of data. We developed a Microsoft Azure solution that uses predictive analytics to give our teams access to current, accurate, and relevant business-driven insights based on the broad set of sales and marketing data available in our environment. Using Cortana Intelligence Suite to benefit business Microsoft sales and marketing data is contained across multiple business systems and applications in our organization, from customer purchasing data in sales systems to marketing data in customer relationship management solutions. The Cortana Intelligence Suite on Microsoft Azure offers a flexible, integrated set of tools to manage big data and transform data into intelligent business action. The suite offers: Predictive analytics. Machine learning algorithms augment your decision making processes with proactive alerting and simplify decision making involving complex data problems. Reliable intelligence. Intelligent structures enable you to interact with data using natural language and familiar interfaces, including cognitive APIs for vision, speech, text, face, and emotion detection. Flexible building blocks. Customizable learning models, APIs, templates, and industry-specific partner solutions help you connect to and collect data of any volume or variety, both on-premises and in the cloud. Scalable and secure solutions. The Azure-based infrastructure means that your solutions can grow with your organization and provide an environment with integrated encryption, threat management, mitigation practices, and regular penetration testing. Using predictive analytics for sales and marketing At Microsoft IT, we used the Cortana Intelligence Suite to design an enterprise-level, scalable predictive analytics platform. The platform is used for developing and publishing predictive data models that incorporate data from across the entire sales and marketing lifecycle. The platform also helps our data scientists and business stakeholders capture business decision criteria from across the global scope of our business. It provides relevant data and metrics for many business scenarios, including: Customer targeting for sales and marketing. Sales, support, inventory, and demand forecasting. Marketing effectiveness assessment. Customer classification and segmentation. Sales performance assessment. Social media analytics.
Page 2 Optimizing predictive analytics using Cortana Intelligence Suite in Azure Building predictive analytics infrastructure in Azure The platform uses several Cortana Intelligence Suite components to deliver predictive analytics. This robust platform lets us deliver a scalable, reliable, and highly available service to users company-wide. The platform uses these Cortana Intelligence Suite components: Azure Data Lake. Azure Data Lake has powerful capabilities that simplify ingesting and storing data. Azure Data Lake is used as a repository for predictive analytics platform data, and it stores all the modeling variables that are used to address the business scenarios supported by the platform. Azure SQL Database. Azure SQL Database is used for relational data storage in Azure. Azure SQL is an important data storage station, and is used across several functional components of the platform, including storing data for use in BI delivery. Azure Machine Learning. Azure Machine Learning gives developers the tools to mine and explore big data in Azure. The platform used machine learning to build training and scoring experiments for all of the data models. Currently, around 50 different experiments cater to varied business problems, each with its own web service endpoint to provide a REST API that can be consumed by a wide range of devices and platforms. Power BI service. Power BI offers powerful visuals that represent data and metrics in meaningful and compelling ways. In the platform, data model results are made accessible through standard Power BI reports in SharePoint. Azure API services. Azure API services are used to provide an interface for different marketing and sales engines/tools to connect to relevant data models and access valuable information. Azure Data Factory. Azure Data Factory provides data management and integration services for data-based solutions in Azure. In the platform, Azure Data Factory orchestrates and manages the multitude of activities that occur across the platform in on-premises and cloud data sources. Azure Data Factory also has functions such as: Triggering Azure Data Lake scripts for data transformation. Transferring data from Azure Data Lake to Azure SQL Database. Running Azure Machine Learning experiments. Running data cleanup and storage scripts. Analyzing business data with a Cortana Intelligence Suite framework End-to-end data model generation and maintenance is available on the predictive analytics platform with key functions shown in the following figure: Figure 1: Data flow within the predictive analytics platform
Page 3 Optimizing predictive analytics using Cortana Intelligence Suite in Azure Each step in the process serves a specific function in the predictive analytics platform model. The process begins with raw data, and then it applies the platform logic and functionality to the data, which results in actionable business data. The following high-level steps describe the end-to-end process: 1. Ingest and transform. Data is checked for a recent refresh, and is refreshed from the original source, if necessary. Azure Data Lake stores data and provides the transform functionality. 2. Analyze. Once data variables are ready in Azure Data Lake, web service endpoints for each Azure Machine Learning experiment are called via REST API to produce model scoring outputs. The model scores are refreshed regularly using these REST APIs to maintain current data, and the model experiments themselves are retrained quarterly to help ensure that changing business dynamics haven t affected model predictions. The data experiments in Azure Machine Learning capture model performance drivers and translate how they might influence outcome prediction. These performance drivers help to translate statistical information into businessrelevant insights. 3. Store. After scoring is completed via the experiment s API, the results are transferred to Azure SQL, along with model variables and additional firmographic information that allow reports to be customized for better targeting and a clearer view of report data. The data is consolidated into a single table, where we can access it with tools or apps that incorporate data into our business processes. 4. Visualize. Reports that are specific to the data model are rendered in Power BI for viewing and consumption. The data is then transferred from Azure SQL to blob storage. 5. Consume. Azure API services provisions our model scores to various components of the platform where they can be consumed by other processes or applications. 6. Orchestrate and Manage. The end-to-end process is orchestrated and managed using Azure Data Factory. In this step, the data movement, transformation, computing, and storage across the various Cortana Intelligence Suite components are defined, scheduled, executed, and monitored to minimize resource requirements and maximize efficiency managing the growing number of assets on the platform. Additionally, a notification system informs admins and users of task and platform status. This platform approach to predictive analytics provides a standard foundation on which we can build our big data business intelligence capabilities. The platform gives us agility and scalability in our predictive analytics development and ongoing operations. The models in the platform identify marketing leads for potential purchases and score existing marketing prospects, helping us to accelerate the sales and marketing process and better manage our marketing automation journeys for our customers. Functionally, the platform provides significant big data processing capabilities, including: Automated model refresh and retrain. The refresh/retrain process for all models is automated and requires minimal human intervention. The automated refresh is agile and adaptable to model inclusions or exclusions. Centralized score repository. Teams across Microsoft can publish their model scores to a central catalog of assets, which serves as an interface for different marketing and sales engines/tools to connect to relevant models and access valuable information. Automated model performance monitoring. Statistical performance of all the models is monitored and archived after every refresh cycle, enabling you to assess requirements for rebuilding or retraining a model. Standardized presentation layer for easy model consumption. The built-in presentation layer allows you to browse through all models published on the platform from a one-click interface. The design of the predictive analytics platform is scenario agnostic. We are currently using it to generate models using sales and marketing data, but the APIs allow for support of other data scenarios, such as finance or business group operations. Model Development via Azure Machine Learning Model development involves the training and validation of multiple models (i.e., experiments) to inform selection of the model that best fits the business need. Azure Machine Learning (AML) is a powerful cloud-based predictive analytics capability that streamlines the model development experience for all data scientist skill levels. By focusing on
Page 4 Optimizing predictive analytics using Cortana Intelligence Suite in Azure AML-based experiments, the predictive platform offers modeling workspaces with best-in-class algorithms, an intuitive drag-and-drop interface, support for R and Python custom code, cloud-based collaboration, and the ability to easily deploy models as ready-to-consume web services (supporting both batch execution services and requestresponse services). The AML experiment catalog also serves to reduce duplication of effort that may otherwise occur when data scientists can t see what models have already been developed by their peers, which can be a common occurrence in a company of our size. Score Publishing via Azure SQL and Azure API services For all models managed on the platform, scores are published to an Azure SQL storage layer. In addition, a parallel onboarding service permits non-platform model scores (models not developed in AML) to be cataloged in the same Azure SQL storage layer, creating a single, centralized catalog of model scores, enabling: Improved discoverability of predictive assets Increased use of predicted scores in business applications Standardized model evaluation frameworks for both usage and impact A standardized web services (API) layer to support direct integration of predictive assets within modern applications. The single layer also makes it easier for application developers; they only need to point their applications to one source of predictive outputs. Without this platform, anybody that builds a model to integrate with these capabilities would have to work directly with the engineering teams that maintain specific predictive models. Model operationalization via Azure Data Factory Model operationalization involves refreshing features, rescoring (and periodically retraining) models, and republishing scores. The predictive platform automates these activities using Azure Data Factory (ADF). For each model managed on the platform, ADF: Schedules refreshes of relevant features in the variable mart. Orchestrates the batch execution of the variable mart features against AML endpoints for batch scoring and periodic retraining. Provides standard model evaluation frameworks for both use and impact. Configures model outputs for alignment to our publishing schema to ensure compatibility with web services and reporting. Future data analytics plans The predictive analytics platform gives our sales and marketing team valuable business data insights that are gleaned from our big data environment. And the Cortana Intelligence Suite in Azure gives us a scalable and reliable platform that allows us to focus on developing new solutions. We are currently defining architecture plans to take advantage of Azure Stream Analytics and Azure Event Hubs to enable real-time stream processing of internet of things scenarios, enabling us to stream millions of events per second into our data models.
Page 5 Optimizing predictive analytics using Cortana Intelligence Suite in Azure Conclusion Our predictive analytics platform enables an entirely new way to look at big data for our sales and marketing teams. The Cortana Intelligence Suite in Azure allows us to create accurate predictive models using an automated workflow, and our sales and marketing teams can quickly gain deeper insights into their business. The ability to create new predictive models based on those within the centralized catalog of experiments has reduced the amount of time required to build models, and made our predictive analytics environment more agile and efficient. For more information Microsoft IT microsoft.com/itshowcase 2016 Microsoft Corporation. All rights reserved. Microsoft and Windows are either registered trademarks or trademarks of Microsoft Corporation in the United States and/or other countries. The names of actual companies and products mentioned herein may be the trademarks of their respective owners. This document is for informational purposes only. MICROSOFT MAKES NO WARRANTIES, EXPRESS OR IMPLIED, IN THIS SUMMARY.