How do you improve your company’s profitability in the fiercely competitive insurance industry? The answer might be sitting in your data.
Insurers have historically been slow to adopt new technologies. But now, as disruptive insurtech startups battle with incumbents for market share, smart companies are using artificial intelligence to gain an edge.
To understand how leading insurance companies are using AI, we gathered four industry experts to share their insights. Dan Donovan, Head of Customer Success at Shift Technology and Wenling Yao, Head of Business Intelligence at Clark joined Paweł Jarmołkowicz and Grzegorz Mrukwa from the Netguru Machine Learning team to discuss:
Packed with tips and actionable advice to help you get the most out of AI in your business, here are the key takeaways from that talk.
Applied successfully, AI has the potential to reduce costs, improve efficiency, and increase profitability. But how are leading insurance companies using it?
Dan explains that at Shift, their core focus is on fraud detection and claims automation.
“In the insurance industry, fraud detection, in particular, is a well-understood use case. You have an ability to build concrete ROI numbers and key performance indicators around it because if you're finding fraud and effectively investigating it, you're saving money and not paying lost dollars on non-meritorious claims.”
“And then there’s the newer area that's emerging in claims automation. Some of the biggest discussions we have are around the efficiency gains that insurers can realize by automating significant portions of the claims adjusting process,” says Dan.
Paweł agrees with the potential for efficiency improvements and mentions a lead scoring model he’s working on that gives a probability score to how likely or ready a client is to buy from you.
“You can maximize the efficiency and productivity of your sales team because they can focus on the most promising leads. Also, your marketing team can use this information to improve personalization and to focus on the highest probability leads or the segment that those leads represent, increasing ROI,” he explains.
Wenling reveals that Clark also uses data and analytics to apply a targeted approach. “The main goal for us is to make sure we are approaching the right customers at the right time and giving them the right recommendation,” she says.
Wenling also mentions that her company is mapping its communications content with customers to better understand when they’re interested in specific content.
Insurance lags behind other industries in terms of digital transformation. That said, many insurers understand that they have masses of valuable data that is being under-utilized and recognize the need to focus on AI.
“We try to take the low hanging fruit, build our first dashboards, and show this initial value to our stakeholders – to show them this can really work – and by doing so, we're making sure we establish credibility and get some initial buy-in for further steps.”
The key to finding the hidden gems in your data lies in having the right perspective. Often companies approach AI from the starting point of their data and what they can do with it. Paweł argues that it should be the opposite.
“You need a business problem you want to solve, and then we can check what data you need to solve it,” he explains. “We can see whether you have the data or whether it needs to be collected or acquired.
“The next step is validation – validating how feasible each use case is and what impact it will have on the business. With that, we can eliminate use cases that are either low feasibility or low business impact. And we end up with two or three use cases that we can further focus on.”
The key message from Paweł is that you can get anything from data, but for maximum value, you need a problem to solve.
Once you’ve identified possible use cases, try to benchmark them against each other before deciding where to invest.
Wenling explains that at Clark, she tries to create a culture where they can assign a main KPI driver not just to use cases for data science, but to every product feature in every stream. “I try to make use cases comparable using normalized business impact,” she says. “So, a use case that aims to optimize a registration funnel is comparable to a use case that aims to improve sales conversion rates.”
“By doing so, we're able to place every use case on the same level, so we’re not comparing apples and oranges. This makes our arguments more convincing to the business stakeholders.”
When you’re building a POC for an application, Grzegorz advises:
“We're trying to connect the dots around how the machine learning system should be measured to make sure that we’re able to deliver business value,” he says. “For the scenario in which you’re using machine learning, specify the KPIs that you want to optimize, then connect these KPIs with the machine learning system.”
Grzegorz recommends verifying the ROI you can obtain as a baseline. If the project has the potential to be profitable, you can proceed iteratively and manage it as a normal agile software development process, improving one KPI at a time.
Wenling explains how they minimize risk when testing their POCs at Clark. “We do a traffic split at the POC stage – only 10% of the traffic is used for the allocation assignment. We’re looking at a specific segment to assess the potential uplift if we implement this algorithm in production. We can then combine this into our business case and use it to estimate an ROI.”
Dan highlights the value of having a change management strategy.
“You can build and deliver the best models in the world from a machine learning and AI standpoint, but if you don't have a proper change management plan in place, it may not matter how good those models are. You can get great results, but if the customer doesn't understand how they're going to take action or how they're going to use the solution, you can run into problems,” he says.
To get started with AI, Paweł recommends, “Look at your processes where there are a lot of small repetitive tasks, simple tasks that take a lot of time and that need a lot of people. Those are a good starting point to think about whether machine learning can help to automate that task.”
Echoing previous comments, Wenling advises, “Always start from the business side, and start with ‘why?’. What kind of business KPIs are we trying to address?” Once you have this clearly identified, she then recommends considering the data and technology requirements.
Dan stresses the importance of evaluating your data internally and ensuring you have good data management processes in place. “Think about how you can bring your data into one place where you can effectively leverage it for multiple types of projects around AI, machine learning, and business intelligence so that you can move quickly and generate benefits much faster.”
He also suggests weighing up the opportunity cost of building and managing projects internally. “Look outside your four walls and leverage vendors who provide Software as a Service who can help drive positive benefits more quickly.”
As many panelists highlighted, the key to successfully getting the benefits from AI in insurance is to start with your business challenges. And remember that you know your business best.
“There are some very costly or very manual processes that could be fairly easily improved or automated with AI, or even with simple heuristic or other automation, not necessarily even AI.”
He stresses that it’s important not to get caught up with trends or what other companies are doing and miss an opportunity to improve processes in your own business.
For the full recording of Disruption Talks: How Leading Insurance Companies Make Use of AI, sign up here.