November 1, 2018

Amongst the major challenges that come up when discussing how to implement artificial intelligence (AI), and any type of advanced analytics, into insurance processes is the issue of how it will impact on employees and skills.

Is AI a string of code, a process, a department, a way of thinking, or a part of the product design and delivery? How does it fit into the concept of the lifetime value of the customer? Where should it sit in an insurance company?

Looking around at recent developments, the technology has reached a point where there is a lot of exploration with a certain amount of hype, and just a few examples of real machine learning going into services.

Allianz is launching a proof of concept for automating the estimation of motor repairs from photos at the accident scene. Zurich began by automating the review of 10-40 page medical reports flowing into health insurance claims. It managed to reduce the average processing time of 58 minutes per case for a manual claims handler, down to five seconds with straight-through processing, and saving the company $5 million per year in productivity gains and reducing claims leakage.

There are many pilots and working groups, looking to deploy ML and faster processes to specific tasks, and one by one a few live examples of automation (with a human-defined algorithm) are reaching the market. Automation rules. It is a step up from manual case handling and traditional statistical modelling deployed by actuaries. However, it’s not the same as machine learning or pure AI, which is a much more transformative proposition (with a machine-defined algorithm in a loop of feedback and improvement).

But taken as a whole, it’s a technology that is going to allow us to refocus insurance employees on technical tasks and analytical work related to data management and things like customer insight, but less on repetitive admin tasks. Within insurance providers we are seeing more cross-functional teams and dedicated innovation teams, set up to test and implement pockets of AI.

Consumers meanwhile are becoming more digitally-driven, and more self-aware about how they can pick and choose solutions, but with a diminishing attention span. It is all creating a new model for the insurance contract and the customer relationship. We are just at the start of the journey.

The recent Insurance Analytics and AI Europe conference in London, gathered together over 300 insurance leaders from all over the region working in data initiatives, underwriting, claims, digital transformation and actuarial roles. The conclusion from the meeting was that there are some skill sets missing for advanced analytics, and outstanding questions about how to balance prediction and improvements in the predictive models versus the need to build a governance structure, the checks and balances.

Data structures and data availability are part of the AI puzzle

The governance aspect is a weighty issue, considering that there are ethical and regulatory parameters. For example consider how to set a machine running, making decisions that could affect pricing, with possible cross-selection effects or unintended consequences for gender (or even using height of an individual as a proxy for gender), which could be an unintended effect of completely unrestricted AI. These are the areas that are going to need a monitoring process and a new business culture about humans and machine-derived algorithms working together.

The discipline of procuring and processing the right data cells necessary for AI, convincing consumers of the currency of their own data, and fixing the data noise and the data errors from the past, are going to become all-important.

Notwithstanding recent improvements in governance and regulation, in the final analysis, there is still not a clear compact between consumers and the insurance industry about the use of personal data. There’s still a group of customers who don’t completely trust insurers, and who feel the new efficiencies and pricing benefits only ever result in benefits for insurers, and profits.

In fact, the real answer is that the business benefits of advanced analytics are, and they should be, a win-win for both consumers and for the industry.

“Data quality is a challenge. It is now much more important than just a few years ago,” commented Jon Holtan, COO, Head of Product and Price at SpareBank 1 Skadeforsikring of Norway.

“Data quality is creating issues with inputs from devices such as health devices and self-declared data from customers. So there’s a need for checks and normalization that makes it very difficult for an individual to cheat the system,” he added. “The established wisdom is that data quality is an issue. On the other hand AI offers opportunities to better interpret the dirty data, though there could be regulatory issues around that….What we are missing most are the managers with experience in AI and analytics. These are the leaders that will influence the rest of the organisation.”

Many insurers are already up-skilling their people into new tasks, to be able to take part in the new RPA tasks, with automated elements of claims and customer service, getting involved with scoping AI projects. But how to ensure the machines are answering the right questions? Does it create a process that’s in the interests of the consumer and the business? How to test and scale AI? Most insurers are already making aggressive strides into these areas.

“We need to agree that AI is a very sensitive topic,” commented Somesh Chandra, Chief Health Officer & board member MAXIS & AXA EB Partners.

“A lot of people start off thinking about jobs or other threats, so there is really a need to talk about change management….There’s a feeling that the technical knowledge or the jargon can take over [the organisation]. But for the customer, the complexity should just be behind the scenes. We can make AI just a buzzword, if we don’t think about the people aspect, and what we actually want to do.”

There’s no single ‘Swiss Army Knife’ with technology

Approaching AI as a tool that can be applied to any part of the insurance business, there was agreement in the conference that there’s no single, wonderful machine that will solve the problems. Don’t look for the ‘Swiss Army Knife’ that can do everything.

In terms of change management, there’s a need to deploy honest communications with staff and stakeholders, and embedded teams into different parts of the business with analytics expertise.

There is a lot of data, whether structured, unstructured or poly-structured, in the insurance space and available to automate the machine learning pipeline. But for most insurance providers, other than the very largest players, machine learning is still a missing concept. There are still a lot of bottlenecks with data and barriers to adoption.

Still more questions to be asked

The pace of change is accelerating, and with a software development analogy, there is a race to the finish line. There’s a sense that if it takes a human time to code these solutions, with the right data management framework, a machine could be coding them faster.

In insurance organisations, this approach of data scientists is rubbing up against the role of IT teams, and there’s also some friction with the customer-facing digital teams. But we can expect to see data scientists finding their role in the insurance organisation, and finding these correct protocols.

Gero Gunkel, Head of AI for Zurich, commented that there’s a need for multiple supplier prototyping. There is no one-size fits all approach. The main lessons learnt for insurers to date are that technology and the AI test lifecycle are just a part of the puzzle.

“We learned that the machines with the human handler together generates the best performance,” he said. “AI is increasingly reaching human-level cognitive abilities but they still need to learn a lot more….In some areas there are issues where there is insufficient data to build a model. There are pockets of big data, but it is not everywhere.”

AI raising the ‘usefullness’ to the customer

At the Insurance Analytics and AI Europe conference, Nils Mork-Ulnes Head of Strategy, Science in the AIG global data science team, described the success AIG has had with embedding AI into products and micro-processes that are very close to the customer experience, rather than as a ‘big bang’ approach in disparate functions. This AIG team recently consolidated its projects to two or three that are headed to become live products, effectively new, innovative user experiences with an “inform and assist” model, increasing the usefulness of insurance to the customer.

He commented that for AIG, the modern digital approach to products was a “must do” exercise, combined with the modular product concept, and creating the ability to move quickly.

“Fundamentally, we think that AI is about everything digitising,” said Nils Mork-Ulnes.

“Rather than general AI it is the small cognitive tasks the technology is proving very good at. Algorithmic approaches help deliver customer satisfaction, embedding AI into the product….But without the right data even the smartest data scientist will have limited impact. With the internal structured data it is easier, but as you go further out to outside sources, you get a lot more unstructured data. So this is not something you can do overnight, to enrich the data.”

Tensions between internal team goals

Sometimes you can have a tension between the goals the business sets for the data, and what the legacy systems will allow. “There can be a conflict between our willingness, and not being able to provide the structured data,” commented Anne-Sophie Grouchka, Chief Customer Officer at Allianz France.

AI can hit all parts of the value chain for insurance. It drives next-generation underwriting and pricing. Everybody is now talking about AI being the future for pricing and underwriting, leading to dynamic, real-time pricing. Within a few years the customer’s behaviour will directly influence the delivery and the pricing model, as it started to do, for example, in motor telematics. The predictions for growth in analytical horsepower and computing power are huge. And there’s talk of the ‘democratisation of AI’, with greater self-service for analytics by non-experts and all kinds of job roles in insurance. Many types of open source AI tools and code bases are now being eagerly taken up by some insurers.

As Jensen Huang of NVIDIA said in the now infamous interview for the MIT Technology Review, “Software is eating the world. But AI is going to eat software.”

It’s a journey that is already creating tensions between different interests and different KPIs within insurance companies. There’s a need to showcase what AI and advanced analytics can do, not just by creating an innovation team or a data science team and positioning them to the side of the organisation. It’s a matter of grabbing hold of problems and fixing them. Create a discussion around blending data science into the insurance organisation, as we at LexisNexis® call “embedded analytics” and three ‘Ts’ of advanced analytics working together: techniques, technology and talent. Gather the broader interest and move forward.

Follow the link to the LexisNexis Risk Solutions website to find out more about how we support insurance providers.