Analytics technologies have traditionally told users what had happened in the past using the data captured in transactional systems. Then, predictive analytics and machine learning began giving users forecasts for the future based on what happened in the past. Now these previously separate application and analytics worlds can merge time. Application developers can hook into the power of data science by using Azure Machine Learning model APIs. They can begin creating application platforms that learn from their past behaviour.
Traditional One-way Analytics Systems
The capabilities of analytics systems, using technologies such as SQL Server and Power BI, have undeniably improved massively in the last few years. However, their role has mostly always been to get an opinion to end users about what had happened in the past. Even as newer predictive technologies such as data mining and machine learning appeared, their outputs were often limited to analysis tools given to end users. Any guidance analytics systems gave typically needed humans to implement it. These systems could be considered “one-way”.
Modern Two-way Analytics Systems
Today’s advanced analytics systems are becoming “two-way”. Microsoft Azure’s Machine Learning service now provides applications with access to its models through APIs. The same systems that provide machine learning models with their source data can now get real-time insights from them. Application systems have “closed the loop” and can now automatically learn from their past behaviour - shown with the red line below.
In-application Decision Making
Accessing Machine Learning APIs is about helping applications to make better real-time decisions based on historic patterns rather than hard-coded business rules. Imagine a customer gets the chance to enter free text while logging a customer service query. Based on its pre-programmed rules, the application might think everything is fine just from combination of ticked boxes and drop-downs options selected. However, sending their text to a Machine Learning model might help sense something actually isn’t quite right and have someone calling the customer before they’ve even left the web site. Although this example may not feel complex, that’s a good thing. It’s an example of how analytics can constantly help applications become more effective whereas previously they’d be relying on infrequently updated business rules.
Azure Machine Learning Resources
Microsoft publish a range of ready to use Machine Learning APIs in the Cortana Analytics Gallery. From the text analytics and sentiment analysis used in the previous example - to customer churn detection - to speech and face analysis. However, Microsoft with its Project Oxford is broadening the scope of its APIs to include human behaviour analysis as well as more traditional customer, financial and operational mathematical models. Machine learning will soon become as much an application developer’s toolbox as a data scientist’s.