AI: The Next Battleground, Part 2 Kevin Mason, EVP Solutions | May 21, 2019

Insurance Technology Systems

In Part 1, we discussed how data and artificial intelligence will be the next battleground for insurers. We covered some high-level definitions using Microsoft’s AI stack and took a deep dive into machine learning. We identified two additional components of the AI stack that make heavy use of machine learning: cognitive services and robot(ic) process automation. In this post we take a closer look at these and wrap up with some strategic thoughts about AI in general. 

Cognitive Services: Sense and Respond

Cognitive services are a set of components Microsoft provides that are directed toward high-level artificial intelligence functions. Making heavy use of machine learning, the components are grouped into 6 areas:

  1. Vision to provide capabilities for video and image analysis
  2. Speech for speech recognition
  3. Language for understanding words and sentences
  4. Knowledge for research management
  5. Search for applying machine learning to search
  6. Anomaly detection for the detection of rare or unusual items

Cognitive services have a wide range of uses, spanning from driverless cars to website chat bots. In almost all cases, cognitive services makes heavy use of machine learning. For example, Google uses Captchas (challenge response tests used to determine if the user is human) to train their machine learning algorithms which are used for a variety of purposes1

For insurers, the immediate use case on the sales side is for chat bots. Here, the technology is less exotic. It winds up being a keyword match in which the user enters key words like “auto” or “workers comp” and the program responds with the correct information. (Think of it as a series of If-Then-Else statements). The other use case for carriers is for claims processing in which photos are sent in to the insurer for damage assessment. There are a number of third-party databases that contain photos of automobile accidents, for example, that can be used to start training machine learning algorithms.

Eventually, cognitive services will be widespread, interacting with users in a variety of different ways throughout the value chain. It bears repeating, however, that at the base of this technology lies the machine learning capability we discussed in the last post. The other technology cognitive services will leverage heavily is robot(ic) process automation.

Robotic Process Automation: It’s Not Science Fiction

Robot(ic) Process Automation (RPA) sounds like a science fiction story but is much simpler than that. It is software designed to handle a high-volume repeatable task without the need for human beings. As with cognitive services, it makes heavy use of machine learning to guide decision making. Quoting from CIO.com, “businesses can automate mundane rules-based business processes, enabling business users to devote more time to serving customers or other higher-value work. Others see RPA as a stopgap en route to intelligent automation (IA) via machine learning (ML) and artificial intelligence (AI) tools, which can be trained to make judgments about future outputs.”2

The benefit is obvious. RPA can be used to automate repetitive task that humans used to do. This allows insurers to more effectively use their employees while at the same time reducing human mistakes and errors. While estimates vary as to the range of jobs that will be eliminated by RPA, the bottom line is that insurers must use RPA to match efficiency gains realized by their competitors.

One obvious use case is in the automation of underwriting rules associated with low touch or no touch policies. If an insurer wants to sell low-cost Business Owners Policies (BOP) – that carry premiums of $1,0000 or less – profitably, they need to do so with almost no human intervention. Anytime an underwriter gets involved, the cost associated with that action will mean the insurer will lose money on the policy. This intensifies the need for no touch or low touch. Solving this puzzle requires RPA in which the underwriting guidelines and decision-making process are encoded in machine learning algorithms which are then trained by historical data. Once trained to the level required, RPA takes over and either reduces or eliminates the need for the underwriter to be involved. This then frees their time to work on more high-value activities.

Final Thoughts

Data and artificial intelligence are technologies that have a very high probability of making a major impact, not only on insurers’ businesses but on the lives of people in general. In dealing with this technology, here are some useful tips to remember as you develop your strategy.

  1. Every technology has limits and AI is no exception. While machine learning does some things remarkably well, it still cannot match humans in decision making efficiency for complex decisions for which there is little data or firm fixed rules. (Think Brexit and whether any algorithm could figure that out). Do not underestimate human creativity.
  2. Understand that every major vendor is building out AI capabilities. It is not a question of if they will become mainstream, it is a question of when. If you do not adopt these technologies, you will be left standing at the curb watching as the driverless cars pass you by.
  3. Be aware that during the creation and training of machine learning models, biases may creep into the models.3 Because of the complexity of these models, these biases may not be apparent or visible. Understand how your models are built, and be on the lookout for biases that may be incorporated into them.

1     https://medium.com/@thenextcorner/you-are-helping-google-ai-image-recognition-b24d89372b7e.
2     https://www.cio.com/article/3236451/what-is-rpa-robotic-process-automation-explained.html.
3     https://searchenterpriseai.techtarget.com/definition/machine-learning-bias-algorithm-bias-or-AI-bias.

Kevin Mason has worked in many aspects of software development since 1981, including roles as product strategist, software development methodologist, project manager, and technology architect for companies such as Cincinnati Bell Information Systems, SHL Systemhouse (now part of EDS), AGENCY.COM, and GENECA. He joined Instec in 2008 and is responsible for development associated with all products. He holds a BA in Political Science, from the University of Iowa and an AS in Computer Science.