May 9, 2016 | Written by: Allen Chan
Share this post:
In Part 1, we talked about how to apply some of the Watson Languages services to help you gain insights from information that are hiding in your customers’ communication. Now we are going deeper to discuss how to apply machine learning, which is the core of a cognitive system, to guide your business operations.
Apply machine learning to business operations
There are business rules governing most business operations. Some are explicitly defined, and some are applied by your knowledge workers (like claim adjusters, auditors and managers) as part of their daily work. In those situations, we can look at applying machine learning (ML) to help increase the efficiency of your operations by using a processing hub like Apache Spark. At recent conference bpmNEXT 2016, IBM Cloud Integration CTO Eric Herness demonstrated how the power of IBM Business Process Manager can be combined with Watson Visual Recognition and Apache Spark.
In this example, we modeled a simple automobile insurance claim process. Upon submission of the claim, we first ran through the Watson Visual Recognition service to extract information to classify the image to help with the decision.
Once the claims adjuster entered all the information, we ran a prediction with Spark. We used a pretrained statistical model based on prior claim results generated by the tracking points in the business model to provide an educated guess on whether the claim should be approved.
In this example, the analysis predicts a 81.25 percent chance that this claim should be approved. Over time, as the system becomes more mature with more and more data, one can provide a simple business rule that if confidence is over 90 percent, the claim can automatically be approved. In addition, we can examine the rules that were inferred by the system based on the existing set of input data, and ask the system to generate a new model based on re-sampling of the existing data with new data.
Beware of self-fulfilling prophecy
As you can see from the example, a properly trained machine-learning system can be incorporated to your business operations to help guide your knowledge workers to make the right decision, and in some cases, help you to make decision as well.
With any statistics-based machine-learning system, it is very important to feed it the right set of data representative of business operations. If the data is skewed in a certain way, it will most likely result in a system that favors certain outcomes. In the same vein, be careful feeding a system with data generated by the system itself, as that at could inadvertently guide (or should I say, misguide) the system toward a certain goal.
Having a good data scientist is very important in setting up a training program of a machine-learning system. With the Cognitive Business Operation Workshop we can bring the expertise to you to create a model that will suit your needs.
On the other hand, many Watson Services in IBM Bluemix have already been pretrained by IBM, so you can get started right away without worrying about training the system.
Want to learn more?
In your journey to adopt a cognitive business operation, take advantages of the many Watson services that are already available with IBM Bluemix, and combine them with IBM BPM to help transform your business operations. And remember, you’re not alone on this journey, and IBM is here to help. To learn more, click here. I would also encourage you to sign up for the Cognitive and Adaptive Business Processes Workshop.
Here are some additional links that you may find useful:
Watson Services – the existing set of Watson Services that are available for use
SystemML – an open-sourced machine-learning library, now part of an Apache Incubator project.