Overcoming Barriers to Implementing Insurance Analytics

“Analytics will transform the insurance industry.” That’s the promise you read in many industry publications, as well as from vendor partners and service providers. I believe we are well on our way to seeing evidence of this. For example, many organizations use advanced analytics to automate processing of new applications, build machine learning models that identify property damage to streamline claims processes, and use artificial intelligence to identify client needs and provide relevant recommendations. There is no shortage of big promises and big opportunities.

What we don’t hear enough about is how hard it is for most companies to capitalize on these opportunities. There are many reasons why companies who buy into the promise aren’t able to quickly find value from analytics. Here are some of the biggest barriers:

• Technology: Yes, investing in new technology requires money, but that’s not the challenge. Integrating new technology with legacy systems that many insurers are already scrambling to replace is the challenge. Building a computer vision model to identify roof damage is the easy part. Figuring out how to integrate real time model scores with batch legacy systems is the challenge. Companies who want to be successful have to find ways to work around the system integration challenge, which brings me to the next challenge…

• Process: When it’s not feasible or cost effective to integrate analytical solutions with systems, companies need to explore ways to integrate the products into their processes. For example, instead of flooding a model score to the system, build a user interface that allows an underwriter, for instance, to get the score by plugging in some information. That requires additional resources and a lot of change management to get production oriented, lean teams to allow for additional “clicks” outside of their standardized workflows. Which brings me to the biggest challenge of all…

• People: While overcoming technology and process is hard, it isn’t insurmountable. The biggest challenge of all is helping leaders who aren’t familiar with analytics to understand the value of the black box and give you permission to experiment with them. We are insurance companies; we know data. However, there has always been an “art” to the decisions we make in pricing, underwriting, claims, etc., and many people struggle to believe a machine or algorithm can replace it.

Insurance companies are on their way to realizing their potential, but there are still real barriers to achieving that goal that is hard, especially for data people, to overcome.

In order to make progress, I recommend the following:

1. Find change agents: There are leaders we all know who are driven by finding new ways to solve problems. Find those leaders and collaborate with them on how analytics can provide solutions to their problems. At COUNTRY Financial, the data science team has successfully partnered with Property/ Casualty claims to rebuild existing rules-based subrogation models using machine learning. The claims team has realized significant benefits over the old model and have been championing this use case with others.

2. Progress over perfection: Instead of worrying about making the biggest impact possible, pick something small to solve and then design effective experiments to prove the effectiveness and accuracy of the model. Ultimately prove little by little that the promise of analytics transforming the organization is indeed possible. While we have built models to completely automate underwriting decisions, it is intimidating for teams to believe in them. We start by automating simple decisions to help teams get on board with using analytics to make decisions for them.

3. Focus on decision support: It’s hard for anyone to be comfortable with what they don’t know and can’t see. Instead of focusing on decision automation (the holy grail for any analytics practitioner), compromise and focus on decision support. Our data science team partnered with Billing Operations to build models to automate billing decisions. We built a user interface to allow billing employees to input a billing number and return a decision from a model. They use this decision and the features flagged to make a decision, versus full automation. By enabling people to use the model result as an input to their own decision, you can build trust in the model and with the team.

The future for analytics is exciting. We are well on our way to realizing its potential, but there are still real barriers to achieving that goal that are hard, especially for data people, to overcome. We must step back and realize that most people don’t think the way that we do. Leverage success and change agents to help chip away at the resistance and uncertainty of the organization little by little. Over time, the results will be exponential and truly transformative.