(Image credit: Adobe Stock.)
It is easy to get excited about what artificial intelligence (AI) offers the insurance industry. It has been a bright, shiny object on and off for a few decades now. This time around, it feels like the glitter is turning into gold, especially for insurance carriers and self-insureds that are deliberate and comprehensive with their approach.
AI has taken its lumps from initiatives that have overpromised and underdelivered in past cycles. And while AI has taken the blame and created a generation of skeptics, much of that blame is misplaced. Integration, or lack thereof, of well-built AI algorithms and decision models with a company’s core technology infrastructure is the leading factor associated with AI initiative failures.
Some of the blame falls on the data scientists that build AI models. They are prone to thinking that “it’s all about the model.” It’s easy for data scientists and the executives funding them to fall into this trap. After all, mining insights from the massive amounts of data insurance companies possess has been every CEO’s dream. And often, the data science team has no problem creating insights from a company’s data. However, they often don’t stop to think and ask, “Are these insights useful?” And more importantly, the question, “How can these insights be implemented given our legacy technology?” is one that no one seems to ask until everyone is ready to go.
This article will address the second part of this problem—how to integrate artificial intelligence with legacy technology. Integrating AI into core insurance technology platforms presents several challenges. Here are some key issues and potential solutions:
Data Quality and Availability
Challenge: AI uses large volumes of high-quality, structured data to function effectively. Insurers might have data quality, availability or inconsistency issues, which can affect AI models’ accuracy and effectiveness.
Solution: Depending on the situation, this data must be near real time, semi-structured and accessible to all approved users. To ensure the quality of the data from now on, companies should invest in data cleansing, standardization and enrichment, as well as in developing data governance frameworks and data management strategies to ensure data quality and consistency.
This work is time-consuming and painstaking, but it is the foundation for effective AI systems. If not done thoroughly and adequately, one is simply building a house of cards that inevitably will collapse.
Challenge: For AI to work, data must pass through it to generate the desired output. Legacy systems can potentially incorporate these dataflows and decision models. However, doing this can be a difficult, costly proposition as most of these systems were not developed to be easily augmented. Yet, its effectiveness is severely hampered or destroyed if the AI cannot be set in the standard workflow.
Solution: If the legacy system makes use of APIs (some newer ones do), then an effective strategy can be to manage dataflows, AI models and the downstream input they generate with the creation of “micro-services” that interface with each other and the core systems via APIs.
This, unfortunately, is rarely the case with legacy systems. As new types of data repositories need to be built, so too does an ecosystem that can access real-time data, act on it with AI and connect back to the legacy system in the appropriate place in the workflow.
Designing and building this ecosystem requires fresh thinking, new skills and collaboration among the involved factions within IT and from the other business functions. It’s best done in-house but is usually outsourced because internal resources are scarce, and skill levels may not be on par with what is needed. Unfortunately, involving outside entities raises the cost of building this new work and can lengthen its completion time.
Skills Gap and Talent Acquisition
Challenge: Integrating AI into core insurance platforms requires a skilled workforce, including data scientists, AI and data engineers, architects, and other specialists. These experts can be challenging to find and retain.
Solution: Insurers should first invest in upskilling existing employees. Many employees are excited to learn new technologies and languages, especially if they can be used on the job.
Collaboration with universities and educational institutions is another fruitful way of finding fresh talent. Internships, scholarships, grants and graduate student collaborative projects are some examples of a thorough approach in this regard. One positive trend in this space is today’s independence from proximity to these universities to have productive relationships.
People with demonstrated abilities in any of these newer disciplines are in high demand, so creating attractive career paths and actively managing them is critical. Breaking some molds about standard practice in your company may be necessary. In my experience, the most talented people in this space can create 10-times or more value than the average performer. Yet, due to HR policies, their pay differential is probably not more than 15-20 percent. Not all these employees are motivated solely by money, but they quickly understand the 10-1 performance dynamic and will want to be treated as elite employees.
These issues are new to most insurance companies and IT departments and should not be ignored. You can’t build world-class applications with average performers, at least not in the time it takes to win.
Change Management and Cultural Resistance
Challenge: Integrating AI into core insurance platforms requires new skills and significant changes to existing workflows and processes, which can lead to employee resistance. Inevitably, more focus will be directed on the latest things being built. The CEO will talk about it to investors and the press. Internally, interest among non-IT professionals will be high. At the same time, some factions will form in IT, dismissing the investments in AI and all that support it as wasteful and infeasible.
If not appropriately managed, this resistance can manifest itself in an us-vs.-them culture, with “us” being everyone supporting the core operations of the company and “them” being the young (mostly) staff who don’t know how things in the company operate. Such a culture, or a variant, can be deadly to all IT initiatives as non-productive squabbling takes everyone’s eyes off the ball.
Solution: Insurance companies should foster a culture of innovation in which everyone can participate. It is critical for senior leaders, not just from IT but also from the business units on the consuming end of AI, to be unified in their vision and support.
Investments in AI can be divisive, especially when resources are stretched, and every employee should be helped to find how they fit into the “new world.” At a large carrier I worked for that was trying to innovate faster and more significantly, we talked about how continuous investment in “strengthening the core” was essential to keep us competitive. In contrast, investments in “creating the future” helped us have a robust one. Time and effort were expended to allow all employees to see themselves as essential to these equally important efforts and that they could be engaged in either. Encouraging continuous learning and sharing accomplishments and plans should be done to move the organization ahead together. A deliberate and transparent change management strategy can help ensure a smoother transition.
An Alternative Approach
These are just some of the challenges companies reliant on legacy systems and processes must address when implementing AI-driven strategies.
An alternative approach to building these systems inside an insurance carrier’s four walls is to leverage technology providers to create AI as software as a service (SaaS). The carrier supplies data to external entities and receives actionable information in real time or near real time. SaaS solutions can help insurance companies incorporate AI capabilities without investing heavily in infrastructure, development or maintenance.
There are several advantages of using SaaS AI tools in a legacy system environment:
Seamless integration. SaaS platforms often provide APIs and pre-built connectors that can be used to integrate AI capabilities into legacy systems with minimal effort.
Scalability. SaaS solutions are designed to be scalable, allowing insurance companies to accommodate changing business needs and user loads without significant investments in additional hardware or software.
Cost-effectiveness. SaaS platforms typically operate on a subscription-based pricing model, which can be more cost-effective than building and maintaining in-house AI solutions. This allows insurance companies to access advanced AI capabilities without significant upfront investments.
Faster deployment. SaaS AI solutions can be deployed quickly, allowing insurance companies to leverage AI capabilities in their legacy systems without delays associated with building custom solutions.
Access to cutting-edge technology. SaaS providers often update their platforms regularly, ensuring their clients’ access to the latest AI technology and features.
Reduced IT burden. With SaaS AI solutions, the service provider handles software updates, maintenance and security, freeing up insurance companies’ IT teams to focus on other strategic initiatives.
Expert support. SaaS providers typically offer customer support and assistance in implementing AI capabilities, which can be invaluable for insurance companies with limited in-house AI expertise.
In my experience, there are two major hurdles to overcome when deploying SaaS tools.
First is establishing the proper data flow between the carrier and the SaaS provider. Typically, the SaaS company will want to train its AI on the carrier’s data, thus requiring five years or more of policy, claim, billing and underwriting data to be extracted from core systems and delivered to the provider. This is, by far, the steepest hurdle to overcome and may be something the carrier’s IT and data teams have not been asked to do before. But where there’s a will, there’s generally a way. Fortunately, the SaaS team has experience doing this with numerous other carriers and can help the carrier’s crew.
The second major hurdle in an SaaS AI implementation is obtaining approval from IT and data security officers. Chief information security officers (CISOs) have demanding jobs. They are accountable for maintaining appropriate controls on where a carrier’s data goes, who has rightful access to it and assuring proper use. They must look at external threats (hackers, cyber terrorists and the like) and internal threats (unauthorized access to sensitive data and data leaks to the outside world from sophisticated and unsophisticated users).
CISOs take their jobs very seriously and are suspicious of external vendors that will gain access to their company’s data, especially when the data “leaves the building” and resides in the cloud.
Fortunately, CISOs are becoming well-versed in new technologies, so they can be open to approving an SaaS provider if they meet several sets of stringent security protocols. Likewise, SaaS providers have garnered experience with such protocols and generally have effective ways of maintaining them such that a CISO is willing to approve the arrangement. More CISOs than ever are engaging with SaaS providers to enable collaboration rather than find ways to thwart it.
We’ve looked at AI implementation through two lenses, first incorporating it within existing legacy systems and second implementing AI as software as a service. I hope this has helped you think not just about the formidable challenges of either approach but mostly about how to overcome them. I genuinely believe we are entering the “golden age” of artificial intelligence, where leveraging it in a business setting will be table stakes for competing in the future.