Artificial intelligence (AI) seemingly has been discussed everywhere over the last few years, and now it’s made its way into the commercial insurance industry. Organizations are using AI and machine learning for everything from streamlining operations to offering more personalized care and better customer service. There is an increasing sense of urgency about getting started on the AI journey. The question is how. Do you develop a custom solution in-house or purchase a third-party solution already on the market?
At first blush, the temptation to build can be strong—after all, you can design exactly what you want for your specific environment. But, in reality, it’s hard to accurately weigh the perceived benefits of a highly customized internal platform against the time and cost requirements of purchasing a tested third-party solution. To help figure out the best course of action for your organization, I’d like to share some criteria that may guide you.
Developing a quality AI-based platform that effectively addresses specific needs requires a dedicated team. To build this team in-house, your organization will need to hire more than just data scientists. Full deployment of a new solution requires product managers, software engineers, data engineers, data scientists, operational experts to develop process and operational workflows, staff to integrate data models into operations, people to manage onboarding and training of the employees who will ultimately use the solution, and staff who can quantify value generation. It’s also important to have all these members operate as one unified team instead of spanning various organizational groups that are not 100 percent aligned.
For some organizations, this may not be a big deal. For others, the process of recruiting, hiring, training, managing and scaling down staff is one of the worst and often most prohibitive parts of embarking on the AI journey. If it’s too daunting to put together a team with the necessary skills, opting for a third-party solution that already has this figured out could be the way to go.
What types and how much data does your organization currently pull? If you can glean industry-leading insights and possess a treasure trove of information internally, you may want to keep it under lock and key, developing new ways to access and analyze it in-house. But this is usually the exception rather than the rule due to the complexities involved in the insurance industry. Even very large organizations with a high number of claims may lack a preponderance of data on a particular feature, injury or litigation scenario. An external vendor, however, could have data aggregated more broadly to cover all situations. External AI vendors draw on a wealth of anonymized and aggregated data from both public and private sources. This means data models can be trained more quickly and accurately.
This is an area where in-house development wins. Your organization can build something from the ground up completely specific to your needs at every turn. If you opt for a third-party solution, there are some constraints that you have to adhere to. However, it’s important to think of customization not just at a point in time but also across the entire life of the AI solution. While you might be able to build exactly what you want right now, if you don’t have continued focus, the solution rapidly will become obsolete. This brings us to the next point.
Just because an AI-based solution is created and implemented doesn’t mean the work is done. It is, in fact, just the start of a journey that requires a dedicated team focused obsessively on the problem. These solutions need to evolve at a fast clip or they will become irrelevant. Models need to refresh, and platforms and software need to be updated, maintained and optimized.
When planning for this in-house, factor in both the staff and time involved to refresh models, fix bugs, or add new fields or features. If you go the third-party route, ongoing maintenance and improvements typically are included in the cost or subscription. If you feel uncomfortable dedicating an internal team to the project on a continuing basis, it might be better to go to a third party.
When it comes to security, in-house platforms have an edge because data is not shared outside of the organization. While you still have to ensure that your networks, systems and endpoints are carefully managed, you are in control. When evaluating third-party vendors, it’s important to check their security credentials and processes to handle data. They need to be as good as your internal processes (if not better) with clear evidence of tight controls through certifications like SOC 2 Type II, HIPAA and HITRUST.
Time to Capture Value
There is a race going on to bring down cost structures dramatically. This is driven by the premium pressures in the market. The primary way to improve combined ratios is by pushing on operational efficiencies. Time matters. It’ll be helpful to think hard about how you could capture value quickly. Ask yourself how much time it will take to:
- Assemble the team.
- Receive data and set up a data pipeline.
- Design the solution.
- Build the solution and create a testing infrastructure.
- Operationalize the solution.
- Design and implement a way to track value.
- Continuously iterate on the solution.
Your ultimate decision may come down to some basic math. Once you’ve narrowed the list of potential outside vendors and received their quotes—which typically include an ongoing fee that covers hosting, support, performance and additional improvements—you can compare those quotes to your estimate for building a solution internally. In calculating this estimate, factor in staffing, training, infrastructure and hosting costs as well as ongoing maintenance and improvements as previously discussed.
I hope these guidelines assist you in making the decision on how to best bring AI into your organization. There are pros and cons to both building and buying. The trick is to prioritize your needs and what is actually feasible and realistic for your company to ensure that the end result more than justifies the means to get there.
First published in Data Science Central.