11436 SSO

The New Predictive Analytics: Proactive and Precise

Jun 10, 2014

In the first installment of this series, we examined technology advances in predictive analytics, the role APIs play, and the benefits for businesses and IT.

In this second and final post, we’ll discuss some critical features in predictive analytics toolsets and how Apigee Insights technology helped improve customer service at Independence Blue Cross.

Real-time, adaptive, and user friendly

The most important objective for predictive analytics is to drive greater precision in predictions. This requires machine learning on big data. So the capability to handle big data is important. This means handling of real-time data, text data, and structured data at the level of fine-grained events.

Watch out for approaches that work on samples since you lose the long tail of signals. Apigee Insights is built on Hadoop for scale and also includes powerful real-time and text processing.

Equally important are machine-learning based algorithms that deliver adaptive and precise learning from patterns in data. Apigee Insights is based on graph data structures we call GRASP; they enable us to dramatically increase our ability to find time-based patterns. We have built adaptive machine learning algorithms on GRASP that result in precision through adaptation even when the world around us keeps changing.

Making predictive analytics tools consumable and manageable for developers, IT, and business is also critical. Apigee Insights supports development by JavaScript app developers and offers features for managing models and interfaces that make it easy for data scientists and data engineers to understand and control predictive analytics.

Apigee Insights includes machine learning on big data and brings together several highly differentiated technologies. GRASP is our secret sauce; we designed this proprietary graph database for large scale pattern-finding to increase the precision of machine learning. We built GRASP on top of the Hadoop file system to handle the scale of big data.  

This enables us to solve the very hard problem of efficiently finding patterns in fine-grained behavior data. We also have language-independent text analytics that allows us to convert unstructured data to structured events so that it can be used in GRASP.  

Adding value, proactively

Predictive analytics changes customer service from reactive to proactive. We worked with Independence Blue Cross (IBC), a major health insurance company, to help them make their customer service better.

We did this by identifying customers who are likely to be unhappy so that customer service reps could reach out to them rather than wait for them to call. The way this works is that now there is a new business process–Apigee Insights generates a list of people who should be called every morning, customer service reps reach out to these people and also record what issues they solved for the customer. Recording the results is feedback for machine learning allowing it to adapt and get better by learning from both true positives and false positives.

Pro-active customer service is an example of a business transformation that fundamentally changes competitiveness.

We deployed Apigee Insights for IBC on a HIPAA-compliant cloud-based service. IBC used unstructured data, including call center notes, and structured data, including medical claims, pharmacy claims, customer complaints and grievances, and customer attributes.

Apigee Insights combined the unstructured and structured data to create GRASP graphs and then used machine learning to score customers for satisfaction. Our machine learning is essentially micro-segmentation that results in very precise scores that, in this case, represent customer satisfaction.

Understanding the future, keeping sight of business value

The key for IBC and other businesses is solutions that make or save money. Enterprise architects need to both understand and educate the business in reducing the technological advances required to attain business value.

Technology organizations who behave like order takers will fail to educate business on new possibilities and their companies will lose to competitors. Disruptive changes require an equal partnership between technology and business.

Previous business intelligence has been about understanding the past. It is left to the human who consumes business intelligence to make decisions for the future and this is affordable only for very high-value decisions.

Predictive analytics is about predicting the future. It fundamentally works by using the past as a predictor of the future. The business value is in automating decision making in areas where human decision-making is too expensive or error-prone.

Image: Nina Matthews Photography/Flickr


Scaling Microservices