您现在访问的是微软AZURE全球版技术文档网站,若需要访问由世纪互联运营的MICROSOFT AZURE中国区技术文档网站,请访问 https://docs.azure.cn.

预测建模和影响客户行为Predictive modeling and influencing customer behavior

数字经济中有两类应用程序: 历史预测There are two classes of applications in the digital economy: historical and predictive. 通过使用历史数据(包括几乎实时的数据),可以仅满足许多客户需求。Many customer needs can be met solely by using historical data, including nearly real-time data. 大多数解决方案主要侧重于当时聚合数据。Most solutions focus primarily on aggregating data in the moment. 然后,他们以数字或环境体验的形式处理并共享该数据。They then process and share that data back to the customer in the form of a digital or ambient experience.

与历史建模相比,历史建模是预测建模。In contrast to historical modeling is predictive modeling. 但什么是预测建模?But, what is predictive modeling? 预测建模使用统计信息和已知结果来处理和创建模型,这些模型可用于在原因内预测将来的结果。Predictive modeling uses statistics and known results to process and create models that can be used to predict future outcomes, within reason. 由于预测建模更为经济高效且易于使用,因此客户需要提前思考体验来提供更好的决策和操作。As predictive modeling becomes more cost-effective and readily available, customers demand forward-thinking experiences that lead to better decisions and actions. 但是,这种需求并不总是建议使用预测解决方案。However, that demand doesn't always suggest a predictive solution. 在大多数情况下,历史视图可以提供足够的数据,使客户能够自行决定。In most cases, a historical view can provide enough data to empower the customer to make a decision on their own.

遗憾的是,客户通常会获得一个 myopic 的视图,该视图基于其直接环境和影响的影响作出决策。Unfortunately, customers often take a myopic view that leads to decisions based on their immediate surroundings and sphere of influence. 随着选项和决策的增长,myopic 视图可能不满足客户的需求。As options and decisions grow in number and impact, that myopic view may not serve the customer's needs. 同时,作为假设在规模上得到证明,提供该解决方案的公司可以看到数千或数百万的客户决策。At the same time, as a hypothesis is proven at scale, the company providing the solution can see across thousands or millions of customer decisions. 这种大图方法使您可以看到广泛的模式和这些模式的影响。This big-picture approach makes it possible to see broad patterns and the impacts of those patterns. 预测建模能力是一项明智的投资,当对这些模式进行了解时,就需要对客户做出最佳决策。Predictive modeling capability is a wise investment when an understanding of those patterns is necessary to make decisions that best serve the customer.

预测建模的示例,以及它如何影响客户行为Examples of predictive modeling and how it influences customer behavior

不同类型的应用程序和环境体验使用数据进行预测:Different kinds of applications and ambient experiences use data to make predictions:

  • 电子商务: 电子商务网站根据其他类似的使用者购买的内容,建议可能需要添加到购物车中的产品。E-commerce: Based on what other similar consumers have purchased, an e-commerce website suggests products that may be worth adding to your cart.
  • 调整了事实: IoT 提供了更高级的预测功能实例。Adjusted reality: IoT offers more advanced instances of predictive functionality. 例如,假设程序集线路上的设备检测到计算机温度上升。For example, suppose a device on an assembly line detects a rise in a machine's temperature. 基于云的预测模型确定如何做出响应。A cloud-based predictive model determines how to respond. 根据该预测,另一台设备将减慢程序集行的速度,直到计算机可以冷却。Based on that prediction, another device slows down the assembly line until the machine can cool.
  • 消费者产品: 移动电话、智能在家甚至车载,都使用预测功能,这些功能根据诸如位置或当天的因素进行分析,以建议用户行为。Consumer products: Cell phones, smart homes, even your car, all use predictive capabilities, which they analyze to suggest user behavior based on factors like location or time of day. 当预测和初始假设对齐时,预测将导致操作。When a prediction and the initial hypothesis are aligned, the prediction leads to action. 在非常成熟的阶段,这种对齐方式可以使产品(如自行驾驶汽车)成为现实。At a very mature stage, this alignment can make products like a self-driving car a reality.

开发预测功能Develop predictive capabilities

始终提供准确预测功能的解决方案通常包括五个核心特征。Solutions that consistently provide accurate predictive capabilities commonly include five core characteristics. 五个核心预测建模特性如下:The five core predictive modeling characteristics are:

  • 数据Data
  • 洞察力Insights
  • 模式Patterns
  • 预测Predictions
  • 交互Interactions

每个方面都是开发预测功能所必需的。Each aspect is required to develop predictive capabilities. 与所有出色的创新一样,开发预测功能需要对 迭代的承诺Like all great innovations, the development of predictive capabilities requires a commitment to iteration. 在每次迭代中,已将以下一个或多个特性成熟,以验证日益复杂的客户假设。In each iteration, one or more of the following characteristics is matured to validate increasingly complex customer hypotheses.



如果客户在 生成 中开发的假设包含预测功能,则所述的原则可能适用。If the customer hypothesis developed in Build with customer empathy includes predictive capabilities, the principles described there might well apply. 但预测功能需要大量的时间和精力。However, predictive capabilities require significant investment of time and energy. 如果预测能力是 技术高峰,而不是真正的客户价值,则建议你将预测延迟,直到客户假设已大规模验证。When predictive capabilities are technical spikes, as opposed to a source of real customer value, we suggest that you delay predictions until the customer hypotheses have been validated at scale.


数据是前面提到的特征中最 elemental 的。Data is the most elemental of the characteristics mentioned earlier. 开发数字程序的每个学科都将生成数据。Each of the disciplines for developing digital inventions generates data. 当然,这些数据有助于开发预测。That data, of course, contributes to the development of predictions. 有关将数据引入预测解决方案的方法的详细信息,请参阅 使用数字发明变得大众化数据 和与 设备交互For more information on ways to get data into a predictive solution, see Democratize data with digital invention and Interact with devices.

各种数据源可用于提供预测建模功能。A variety of data sources can be used to deliver predictive modeling capabilities.


行业专家使用有关客户需求和行为的数据来通过对原始数据的研究来开发基本业务见解。Subject matter experts use data about customer needs and behaviors to develop basic business insights from a study of raw data. 这些见解可找出所需的客户行为 (或) 不希望出现的结果。Those insights can pinpoint occurrences of the desired customer behaviors (or, alternatively, undesirable results). 在预测过程中,这些见解可帮助确定可能会产生积极结果的潜在相关性。During iterations on the predictions, these insights can aid in identifying potential correlations that could ultimately generate positive outcomes. 有关启用主题专家开发见解的指导,请参阅 使用数字发明变得大众化数据For guidance on enabling subject matter experts to develop insights, see Democratize data with digital invention.


人们始终试图检测大量数据中的模式。People have always tried to detect patterns in large volumes of data. 计算机旨在实现此目的。Computers were designed for that purpose. 机器学习通过检测这类模式(一种由机器学习模型组成的技能),加速了这种寻找。Machine learning accelerates that quest by detecting precisely such patterns, a skill that comprises the machine learning model. 然后通过机器学习算法应用这些模式,以便在算法中输入新的数据集时预测结果。Those patterns are then applied through machine learning algorithms to predict outcomes when a new set of data is entered into the algorithms.

使用见解作为起点,机器学习可开发和应用预测模型,以使数据中的模式变成资本。Using insights as a starting point, machine learning develops and applies predictive models to capitalize on the patterns in data. 通过多次迭代、测试和采用迭代,这些模型和算法可以准确预测未来结果。Through multiple iterations of training, testing, and adoption, those models and algorithms can accurately predict future outcomes.

Azure 机器学习 是 Azure 中的云本机服务,用于基于数据构建和定型模型。Azure Machine Learning is the cloud-native service in Azure for building and training models based on your data. 此工具还包括一个 工作流,用于加速机器学习算法的开发This tool also includes a workflow for accelerating the development of machine learning algorithms. 此工作流可用于通过视觉界面或 Python 开发算法。This workflow can be used to develop algorithms through a visual interface or Python.

对于更可靠的机器学习模型, Azure HDInsight 中的 ML 服务 提供了基于 Apache Hadoop 群集构建的机器学习平台。For more robust machine learning models, ML Services in Azure HDInsight provides a machine learning platform built on Apache Hadoop clusters. 此方法可以更精细地控制基础群集、存储和计算节点。This approach enables more granular control of the underlying clusters, storage, and compute nodes. Azure HDInsight 还通过诸如 ScaleR 和 SparkR 之类的工具来提供更高级的集成,以便基于集成和引入数据创建预测,甚至可以使用流中的数据。Azure HDInsight also offers more advanced integration through tools like ScaleR and SparkR to create predictions based on integrated and ingested data, even working with data from a stream. 使用 " 航班延迟预测" 解决方案 ,可以根据天气情况预测航班延误。The flight delay prediction solution demonstrates each of these advanced capabilities when used to predict flight delays based on weather conditions. HDInsight 解决方案还允许企业控制,如数据安全性、网络访问和性能监视到操作模式。The HDInsight solution also allows for enterprise controls, such as data security, network access, and performance monitoring to operationalize patterns.


构建和训练模式后,可以通过 Api 应用该模式,这会在数字体验交付过程中做出预测。After a pattern is built and trained, you can apply it through APIs, which can make predictions during the delivery of a digital experience. 其中的大多数 Api 是根据数据中的一种模式从经过训练的模型生成的。Most of these APIs are built from a well-trained model based on a pattern in your data. 随着越来越多的客户将日常工作负荷部署到云,云提供商所使用的预测 Api 将导致采用速度快得多。As more customers deploy everyday workloads to the cloud, the prediction APIs used by cloud providers lead to ever-faster adoption.

Azure 认知服务 是由云供应商构建的预测 API 的示例。Azure Cognitive Services is an example of a predictive API built by a cloud vendor. 此服务包括内容审核的预测 Api、异常检测和个性化内容的建议。This service includes predictive APIs for content moderation, anomaly detection, and suggestions to personalize content. 这些 Api 已准备就绪,可供使用,并基于众所周知的内容模式,Microsoft 已使用这些模式来训练模型。These APIs are ready to use and are based on well-known content patterns, which Microsoft has used to train models. 其中每个 Api 都基于你要送入 API 的数据进行预测。Each of those APIs makes predictions based on the data you feed into the API.

Azure 机器学习 使你可以部署自定义构建的算法,只需根据自己的数据创建和训练。Azure Machine Learning lets you deploy custom-built algorithms, which you can create and train based solely on your own data. 有关 Azure 机器学习部署预测的信息,请参阅 将机器学习模型部署到 AzureFor information about deploying predictions with Azure Machine Learning, see Deploy machine learning models to Azure.

有关为 Azure HDInsight 上的 ML 服务公开开发预测的过程的信息,请参阅 设置 HDInsight 群集For information about the processes for exposing predictions developed for ML Services on Azure HDInsight, see Set up HDInsight clusters.


通过 API 提供预测后,可以使用它来影响客户行为。After a prediction is made available through an API, you can use it to influence customer behavior. 这会影响到交互的形式。That influence takes the form of interactions. 与机器学习算法的交互发生在其他数字或环境中。An interaction with a machine learning algorithm happens within your other digital or ambient experiences. 当通过应用程序或体验收集数据时,它将通过机器学习算法来运行。As data is collected through the application or experience, it's run through the machine learning algorithms. 当算法预测结果时,可以通过现有体验与客户共享此预测。When the algorithm predicts an outcome, that prediction can be shared back with the customer through the existing experience.

了解有关如何通过调整后的 现实解决方案创建环境体验的详细信息。Learn more about how to create an ambient experience through an adjusted reality solution.

后续步骤Next steps

通过发明和创新方法专业,你现在已准备好学习如何以客户理解的方式进行构建。Having acquainted yourself with disciplines of invention and the Innovate methodology, you're now ready to learn how to build with customer empathy.