In the past, we’ve explored the need for insurance companies to adapt to millennial buying preferences through customized policy offerings and a more personalized customer experience. AI could help to these ends, but how could insurance carriers reach this point of AI transformation?
According to a study by Accenture, 82% of insurance executives believe their companies need to innovate in order to maintain a competitive edge in the coming years. This article highlights AI and data science strategies insurance carriers could implement to generate an ROI from early AI projects and innovate in their industry with AI.
Insurance Nexus is hosting the Insurance Nexus AI and Analytics USA Summit 2020, May 12 – 13 in Chicago, an event that the company claims will foster a dialogue between innovation experts on AI and data science in insurance. Experts at the event will discuss the importance of company agility when innovating with new technology.
This article includes insights from speakers at the 2019 Summit, including Lee Ng, Vice President of Innovation at Travelers, Glenn Fung the Chief Research Scientist of AI & Machine Learning at American Family Insurance, and Ted Stucky, Managing Director of QBE Ventures, along with insights from our AI executive guides.
For more on unlocking AI ROI and innovating with data science in the insurance industry, download Insurance Nexus’ latest white paper: Agility Is The Key To Successful, AI-Powered Innovation In Insurance
Buying AI Products in Insurance
When it comes to adopting AI in insurance, a carrier has two choices: buying from an AI vendor or building an AI application in-house. The former isn’t necessarily required for an insurance firm to innovate in the industry.
The carrier could buy an AI application from a vendor, and if it integrates it well into its workflows to cut costs, drive revenue, or venture into a new business opportunity, the carrier could see a powerful ROI without having to build anything from scratch.
Shane Zabel, Head of AI at Raytheon, highlights this dynamic well in our interview with him on our AI in Industry podcast. He says:
The way to think about…your AI technology space is figure out where you fit in the market…and what part of that market…to you want to be…leading in. There are a lot of components in AI and ML, and you may not want to tackle them all in-house.
One example would be your data strategy. If you are selling some type of marketing product…You don’t want to go off and build your own data lake technology. That might be a great place to partner because that is not your core discriminator. It’s an enabling technology that you have to have, but its not something you are going to build a [part] of your business around….You may want to partner with all of the technology, and…your discriminator is that you are an integrator.
Zabel suggests that companies could buy their AI technology and leverage it to position themselves in the market better. For example, an insurance carrier could buy a machine vision application that allows customers to take a picture of their cars’ damage and immediately get an estimate on how much the insurer will pay them to fix it.
The insurance carrier could build such an application in-house, but it could also buy it and focus more of their attention on marketing the product to attract new customers. This makes sense in some cases, but there are risks involved in buying AI products from vendors:
Risks to Consider When Buying AI in Insurance
AI Vendors Reposition Themselves Often
AI companies often reposition themselves as they seek to find their best product-market fit for their technology. There is no guarantee that an AI vendor will stay in business or continue to work on the specific kinds of algorithms they are at the moment. This could leave an insurance carrier that buys from the vendor without a way to troubleshoot or update the product it bought, especially if the carrier doesn’t employ in-house data scientists or doesn’t have leadership that understands AI at a conceptual level.
Vendor-Bought AI Products Still Require a Lot of Work to Integrate
Oftentimes, buying an AI product from a vendor still requires a lot of work on the part of the client insurance carrier before it’s operational. Even if the product comes 90% effective “out of the box” and is a mostly pre-trained algorithm, the insurance carrier is still very likely to incur steep integration costs.
Specifically, the insurance industry may struggle to integrate AI products not just because of technical challenges, but business challenges. Stucky recalled an instance in which 80% of the time a company spent integrating an AI system was spent integrating it into existing business workflows:
The actual models themselves worked incredibly well out of the box, interpreting doctors’ handwriting and the different forms coming in. There was 99% accuracy. It wasn’t a process of ‘let’s force all the [users] to use the same form.’ That’s never going to happen. Let’s just deal with what we’re being given and make the best of it.
The first half of Stucky’s remarks highlights the corporate agility benefits “buying” may provide insurance carriers. Buying an AI product may mean buying a product that is capable of solving a business problem at least at the technical level.
However, the second half of Stucky’s remarks reveal a challenge that the insurance industry faces in incentivizing its employees to trust that an AI product will solve the problem they’re having any better than their existing system.
It’s difficult for departments in large enterprises to pivot from the way they do business toward a new method that may or may not be better than the methods they’re used to.
At the very least, building an AI project requires that an insurance carrier foster the kind of culture that may allow employees to transition to a new, AI-enabled way of doing their work in a way that’s less abrupt and more palatable. This cultural dynamic is discussed in more detail later in the article.
Innovating With AI Toward A Desired Result
It is also imperative to innovate toward a specific business problem. For example, an insurance carrier may want to reduce the time it takes for customers to receive a payout for their claims.
Next, the carrier would ask whether the kind of improvements it wants to make actually require AI or not. Oftentimes, a business can successfully cut costs or drive more revenue in a particular area without artificial intelligence.
The carrier may, for example, only be looking for a modest reduction in the time it takes customers to receive their payouts, and this could be done by streamlining a workflow, implementing robotic process automation software (which is rarely ever AI), or digitizing claims forms with old-school optical character recognition.
Avoiding AI “Toy Applications” in Insurance
Embarking on an AI initiative for AI’s sake is a recipe for disaster and a waste of money. When a company builds an AI product just to “check the AI box,” it is building what we at Emerj call “Toy applications.” When it comes to starting an AI initiative, Ng remarks:
You have a tolerance for failure but no tolerance for incompetence. You need a lot more discipline to innovate than just to run a project because, unless you’re very clear in your mind about how you’re setting up that experiment, you are wasting your time.
Measuring the ROI of an AI Project in Insurance
In order to innovate with AI to solve a specific business problem, an insurance carrier must first come up with a hypothesis that AI might enable some cost reduction, increase in revenue, or new business opportunity. Specifically, the carrier needs to determine measurable metrics of success for the AI project. If the goal is to reduce the time it takes customers to receive a payout on their claims, these metrics may include:
- A measure of how fast customers are receiving their payouts with the AI system in place vs. how fast they received payouts before the AI system was in place
- A measure of how accurate the payouts customers are receiving when the AI system is in place are vs. how accurate they were before the AI system was in place
- A measure of ROI on the AI system, taking into account the system’s integration costs and how much overhead was reduced by implementing the system
Incompetence, as Ng puts it, would be starting an AI project before thinking about how to measure success. “Your key goal is to learn, pivot, and decide if you want to put in an even bigger investment,” Ng explains. As such, if an AI project prevents an insurance carrier from learning, then that carrier is “incompetent.” However, If the carrier discovers the AI project was unable to reach the carrier’s target ROI the way it hypothesized, the project is a failure, but the carrier is not incompetent.
From this point, the carrier can learn from the project’s failure. It can iterate on the project or cancel it, still having gained the knowledge that what they tried didn’t work and the skills to try again with AI in the future.
Developing Critical Capabilities for AI Adoption at Insurance Carriers
Ng’s remarks get to one of the core returns on investment that companies often overlook when taking on AI projects, and that is the ability for a company to develop Critical Capabilities for adopting artificial intelligence well into the future. When a company develops its Critical Capabilities, it builds the skills, resources, and culture needed to leverage AI in the coming decade and beyond.
One of these skills is the knowledge of how many data scientists will be required for a given project and the foresight to grow data science talent from within the company. Companies cannot bring in a consulting firm and expect them to operate with the full context and strategic “know-how” of an actual employee at the insurance company. Insurance carriers will want to foster the growth of this AI talent in someone in-house, someone who has the carrier’s interests in mind and who is aligned with the carrier’s motives.
Initial AI projects done well also build the Critical Capability of establishing core data access, storage, and quality. AI initiatives need a lot of well-maintained data, and insurance carriers need to realize there is an incredible benefit to a project that allows the carrier to get better at cleaning, harmonizing, and storing its data.
An AI project may “fail” because the insurance carrier realizes its data storage methods are not sophisticated enough or that it hasn’t collected enough data to achieve any ROI from the project. Such a failure could lead the carrier to build new data infrastructures and creating new data collection methods, and as a result, its AI next projects become that much more likely to generate an ROI.
When it comes to Critical Capabilities, one key weakness specific to the insurance industry is that it doesn’t necessarily have a culture of innovation at its core. It is an industry built on internal policy and strategic planning; iteration involves constantly shifting and recalibrating ideas. These two ideas are opposed. According to Ng:
One of the biggest changes is to iterate as you go along. This is anathema to many carriers as they are more used to developing business plans and then executing, questioning any deviations as they occur. Innovation does not follow this neat plan.
This may be the most difficult challenge for insurance carriers to overcome in the coming decade as AI’s potential unlocks within the industry, and it’s a challenge that leadership will in large part need to solve for themselves first.
Concluding Thoughts on Innovating With AI and Data Science in Insurance
An insurance carrier that establishes its Critical Capabilities for AI adoption early could build a product or customer experience that is truly innovative in the industry.
That said, establishing Critical Capabilities is not innovating in and of itself; a foundation of strong Critical Capabilities at an insurance carrier merely provides an AI initiative with the best chance of success.
True innovation in the insurance industry will come from a top-down AI strategy that starts with AI champions on the leadership team. These leaders will steer their companies toward experimenting with AI projects, measuring their success, and learning from their failures. In doing so, they will build a more agile organization with a culture that encourages iterating with new technology to solve business problems and win market share.
This article was sponsored by Insurance Nexus and was written, edited and published in alignment with our transparent Emerj sponsored content guidelines. Learn more about reaching our AI-focused executive audience on our Emerj advertising page.
Header Image Credit: The Digital Enterprise
