As the use of artificial intelligence (AI) becomes increasingly prevalent throughout the insurance industry, efforts are underway to ramp up collaborations with solutions providers even as the debate rages around whether to use open-source options or develop proprietary solutions, how to grapple with a variety of ethical concerns, and what stance regulators will ultimately take.
At a very practical level, the choice between outsourcing the use of AI or developing solutions in-house requires careful balancing of a range of considerations, including how easy or difficult it might be for insurance providers to develop technology-heavy solutions, how effective these solutions will be given the high bar of industry-specific knowhow they require, and the increasingly onerous need to protect customer data.
Regardless of all these considerations, AI has the potential to upend the insurance industry.
Adopting a personal touch
The insurance industry used to be high-touchpoint and interpersonal, relying on communication between agents and customers. Pricing, underwriting and claims management all required a lot of face time and energy from multiple parties. The adoption of AI solutions could go a long way towards further automating all these processes and many others.
On the positive side, AI solutions could drastically speed up how insurers gather and digest the large amounts of data they generate, tapping into algorithms to reduce the manual effort from agents. And this could generate plenty of value. A study by McKinsey suggests AI could add US$1.1 trillion in value to the insurance business, said Howard Kwong, Chief Digital Officer at Prudential Hong Kong, speaking during the Digital Insurance Conference APAC 2023 in Hong Kong on June 27.
As in most other industries, the most common use of AI in insurance for the time being is general-purpose AI, which can be used to generate marketing materials, emails, images and videos. It can also be integrated into office productivity tools, said Andy Chun, Regional Director – Technology Innovation, at Prudential plc, during a panel discussion on the open AI ecosystem for insurance.
The low-hanging fruits for AI are in areas such as customer service with applications like chatbots or to generate regular reports, said Chun. AI can help with research, digesting industry-specific papers and articles, and generating reports to help professionals catch up on industry trends.
Chun also believes there are three more areas in which generative AI can help with customer journeys.
“From reaching out to find the right customers, marketing and recommending products to these customers, and underwriting, AI could even help with claims, fraud detection and streamlining non-customer-facing functions,” Chun said.
AI can also help develop new products and services that are too labour intensive, such as personalised products.
Technology development can also be strengthened with the use of AI.
“We have to reimagine that entire software development lifecycle, what role AI can play in the use case? Could be cogeneration, or planning or code review, or even maintenance,” said Kwong.
Proprietary vs open-source
There are two main pathways for the industry to develop AI solutions. One is taking the proprietary approach, using in-house tech teams to build solutions, sometimes from the ground up or licensing proprietary software developed by third parties. The other alternative is to use existing open-source engines.
The source codes of open-source software are open to users to modify and distribute freely. Proprietary software, on the other hand, is tailored and limited for use by an entity, and cannot be distributed or modified without expensive licenses.
There are already some ready-to-use open-source AI solutions for firms to adopt, said Efstratios Tsougenis, a director responsible for AI & Tech Risk CoE of Prudential plc. Big companies such as Facebook and Microsoft have put a lot of investment in some of the models and they have been commercially provided to the companies.
A lot of work has already been done in research into large language models (LLM). There are about 230,000 open-source models out there and around 45,000 datasets for use to train your own model. This could lead to greater use of open source solutions that might be more focused on insurance regulations, products and processes.
Decisions, decisions
Ultimately, the decision to use proprietary or open-source solutions really depends on the specific use case.
“If we look at claims, especially over claims, probably we need a proprietary model with proper health and medical data,” said Gary Ng, a data engineering and data analytics thought leader who participated in the conference.
There are a number of factors to balance out when determining whether to use proprietary or open-source AI solutions, said Kwong, including the sensitivity of the data. Protecting customer data is a key concern for insurers.
No matter which approach companies consider, each one has its own advantages.
“With open source, we have transparency and openness. You feel that you are in control of your data, but at the same time, with proprietary you might get a bit more advantage on maintenance and security,” said Kwong.
Business strategy is another important consideration.
“If we are looking at this issue as a potential insurer, looking at use cases that are fairly supportive of what we do on a day-to-day basis such as market analysis, then maybe a proprietary model will be a good fit,” said Kwong.
However, service providers whose end users are insurers, then open-source AI may be useful to build solutions customized (and cheaper) solutions.
How about both?
As far as Chun is concerned, the future is likely to be “hybrid” and include a mixture of open source and proprietary solutions.
“People will select depending on the nature of the application,” said Tsougenis.
However, there are still risks involved in the adoption of AI solutions, both open source or proprietary.
Tsougenis thinks one concern is transparency. For proprietary solution, the issue relies heavily on the trust the company has with the solution provider. If the company is using an open source solution, the burden of certifying is part of the company’s responsibility.
“It’s still a burden on the company itself, to convince the regulators that this particular open source (solution) I’m using satisfies the regulatory requirements,” said Chun.
A new way of collaboration
However, because big insurance providers don’t always have the capabilities to develop technology solutions, they tend to collaborate with start-ups using new business models, not like traditional one-off service purchases, to develop their solutions, said Kwong.
“Testing the model will be much more engaging. There should be more discussions and assessment on the model. So this means we need to work closely with the startups in a way that both sides could be benefited,” said Tsougenis.
Kwong thinks this process could generate valuable lessons.
“We should not do these projects in a typical project management way. This is not a spectator sport. You have to play. You have to participate.”
Kwong would engage engineers and data scientists in projects because “this allows you to develop that knowledge and that maturity, that you can make decisions down the road and not entirely have to depend on the suppliers…. That’s self-education.”
