As an example of the disruptive potential, AI-powered applications estimated to add $13 trillion in value to the global economy in the next decade. But while many in the insurance industry see the value of the new technology, there is uncertainty as to how to leverage it to scale and evolve rapidly.
Advances in AI and ML will impact every level of the value chain, from product design and pricing to underwriting, distribution, administration and claims management, according to Andrew Cassel, VP of Data & Analytics, and Scott Ziemke, Director of Data Science at Vertafore. The entrepreneurs explain to Digital Journal that to successfully scale, companies should start off with a clearly-defined strategy and operating model, and structure roles around AI.
Digital Journal: How is the insurance industry being impacted by digital transformation?
Andrew Cassel: It’s changing everything — from taking old paper processes and converting them to digital and taking existing digital processes and figuring out how to leverage that data. I think about my dad, who sold insurance before he passed away. At the time, email was just emerging, and he would ask his assistant to print out all of his emails. The idea of going fully digital is a huge shift in the way many agencies, especially smaller, family owned businesses, have been operating. But, it’s an absolute necessity. Folks like my dad are retiring at a rapid rate — as many as 10,000 baby boomers turn 65 every day – and we have to replace that workforce. The next generations to come on board have grown up in the digital age and want a digital workplace, since they understand that technology allows for more streamlined job functions and frees up time to focus on customers. It’s going to be tough to recruit and retain those employees if the first day on the job, we ask them to log into a system that looks like Windows 95 and we’re still pushing paper for every task.
From the client perspective, they’re also demanding a digital experience. Insurance has historically been behind in this space. We need to evolve to meet their expectations by understanding how much digital they want and what that looks like. Digitizing and automating low-value, repetitive tasks will give customers the self-service options they want and give brokers more time to spend with clients, understanding their needs and how best to solve them.
DJ: Are most of the innovations coming from start-ups?
Scott Ziemke: There’s been more than $6 billion invested in InsurTech over the past year alone. But, while there’s been a lot of money flowing to startups, you also have established organizations driving real innovation, especially when it comes to data analytics. Given Vertafore’s history and the volume of data we have at our disposal, we can do things that startups can’t yet, because it takes decades to collect the kind of data that allows us to spot trends and opportunities. The models that are built by data scientists will increasingly be commoditized – the real value and power is in the data that you have available to train those models. It will be interesting to see how these new and novel tools will pan out when they don’t have the data at their disposal.
Another advantage that organizations like ours have is that we’ve also built some very deep and robust partnerships with our customers. We work together with them on innovations to develop solutions that the entire industry can benefit from. The advantage here is that the parties involved are well-established and well-funded to carry out the work, rather than being at the mercy of capital investment. We’ve also built strong relationships with the carriers who are doing very sophisticated data science, giving us access to those resources as well.
DJ: How are insurance companies using big data?
Cassel: Insurance is all about defining risk and assigning a financial value to mitigating it. Carriers want as much info about the end-insured and what exposures they might have in order to decide if they should insure them and at what rate. To do so, they gather historical transaction data and information about the individual, such as their credit report as an indication of risk, their home location, etc. All of this gets fed into complex models to assign an economic risk value. In this way, big data can help carriers more accurately define the risk in order to home in on the economic value and set premium rates accordingly. That helps the end-insured get the coverage they need at a price that makes sense.
For agencies, big data allows them to do a better job of servicing their customers by better assessing the risks they have and helping them to mitigate those. For example, if I have a client who’s a restaurant owner opening a new store, the client might come to me for general liability coverage. But, when I look at the data, I might see that my customer is significantly exposed to other risks as well, like cybersecurity and data theft. Armed with industry data, I can better assess their overall risk, suggest the right coverages and then leverage my carrier network to get them the best value.
On the other side of the coin, let’s say I’m an agency that’s done very well selling commercial auto coverage in Massachusetts, but I’m ready to expand. Where do I go? Vermont? New Hampshire? With access to the market data, I can see where the opportunity is and where my carriers have an appetite to write new policies, which helps me move in the right direction.
DJ: How is artificial intelligence changing InsurTech?
Ziemke: AI is a pretty loaded term. In practice, it refers to disciplines like machine learning and predictive modeling. These tools allow us to identify and automate certain routine human behaviors, interpret the meaning of text or spoken word, and pull context from image files. For example, we can use predictive modeling to determine whether an end insured is likely to renew their policy with an agency, or if they are likely to take their business elsewhere.
DJ: What can insurance companies do with AI?
Cassel:One of the most valuable things we can do with AI is to automate and simplify insurance. This is something carriers, agencies, and customers want. There are a lot of extremely manual tasks that happen from a broker’s standpoint: generating documentation, loading data into the system, forwarding documents, etc. These tedious tasks can be automated, eliminating time-consuming activities and making room for more valuable efforts. AI, and Machine Learning specifically, can help us identify which of those tasks can be automated, and which are best left to a human to perform.
In the same way the AI in my GPS learns my behavior — that every weekday I go to the office in Denver, but on Monday, Wednesday and Friday, I go to the gym first and then to work — AI can learn insurance behaviors too. For example, when an agent kicks off a renewal workflow in an agency management system, a learning algorithm can determine which steps are routinely taken, and over time can begin automating those steps.
I think we’ve finally reached a point where the proverbial “hype cycle” around AI is settling down and we’re now seeing practical applications rather than concepts. It’s exciting to see the potential for how these predictions and automated workflows will enhance the agent/end-insured relationship.
DJ: Can insurance companies sometimes get AI implementation wrong? Why does this happen?
Ziemke: One of the biggest risks of any modeling effort is in introducing bias to the models results. There’s a great deal of effort that goes into making sure our models don’t include skewed data that can cause bias toward certain groups. This is where having access to a really large data set that spans the industry is super helpful. If you train a model on a single agencies data, it will in some ways be more accurate in that any process or data nuance that is present within that agency will be accounted for. But it will also be subject to substantially more bias and is at risk for experiencing a shock in the results at any change in process, personnel , business, etc.
Another trap is in expecting models to be 100% accurate. A famous statistician named George Box once said “All models are wrong, but some are useful.” You want to think about a predictive model as a tool that you can use to assist with decision making, not something that is going to make a decision for you full stop. One thing that machines are not yet good at is a full accounting of context, and you need a person to interpret and act on the output.
The biggest issue we see is companies spending a lot of time and money solving the wrong problems. We need to first understand what problems people want to solve, then figure out if we can solve them. But just because we can, doesn’t mean it will have the desired end result, or that we can do so cost effectively. Data science is costly, so it makes no sense to automate, simplify or predict the wrong thing. We must first figure out what’s the right problem to go after, then build data analytics and AI support around that. Taking a use case driven approach and having the proper heuristics in place for decision making is probably the most important business construct that is needed for AI initiatives to succeed.
DJ: How does AI-led insurance benefit consumers?
Cassel:The end goal is better service. AI enables carriers to get better at quantifying risk and developing products to mitigating those risks. As the data gets richer, and our analysis techniques improve, customers will get products tailored more for their specific needs. When agency customers need advice, AI not only helps predict risk and tailor a better package, but it also automates rote tasks, giving customers more quality time with agents, resulting in better service.
Just one example is the claims process. This is the source of a lot of end-insured’s bad experiences with insurance. AI can help fix that, not only by speeding the payment of claims, but also overcoming fraud issues and staffing shortages that hold up the process. It can also give customers better insight into how long it will take them to get paid, which sets expectations and keeps the end-insured in the know, all of which contribute to a better customer experience.
