The capability of generative Artificial Intelligence (AI) applications like ChatGPT suggests potential applications to many sectors including general insurance and its value chain. In this ICNZ Scholarship winning essay, AA Insurance’s Toby Kelly discusses.
Generative Artificial Intelligence (AI) applications have taken the world by storm in 2023. ChatGPT-3 from OpenAI reached 100 million active users in January, just two months after launch —making it the fastest growing consumer application in history. 
Generative AI features have quickly followed from leading technology giants, with Google releasing a similar GP-esque tool just 6 weeks later.  Private investment in AI technology start-ups has followed the hype, increasing 400 per cent compared to the first half of 2022.
The predicted impact on business is huge, with leading consultancy firm McKinsey describing a new paradigm for work that will touch every industry and role, potentially automating up to 70 per cent of work activities and adding $4.6T USD annually to the global economy from productivity gains.
For general insurers specifically, this essay will argue that the benefits of adopting generative AI across the value chain will outweigh the risks. The technology has the potential for wide-ranging applications that promise to increase productivity, decrease costs through automation and improve decision-making capability.
In the context of an industry that has been slow to adopt new technology from previous waves of innovation, the ability to realise these benefits will accrue for those with the appropriate existing technology, human capital and culture of experimentation.
What makes Generative AI transformative?
Generative AI is described by Accenture as,” using sophisticated algorithms and deep learning techniques…to create new content, insights, and solutions that were previously thought to be exclusively within the realm of human creativity.” 
Beyond the presence of creativity, the speed and precision with which applications can respond to a variety of media is particularly impressive along with the breadth of tasks they can undertake including creating and responding to text, music, imagery, video and voice, even when the AI hasn’t been specifically trained to do so.
Generative AI tools can exist both as a co-pilot feature to existing processes and systems such as drafting an email, or they can be programmed as part of other software as autonomous agents for example, directly responding to customer emails. This means there are a multitude of ways they can be used in a business context.
What impact could it have across the general insurance value chain?
Sales and service
Utilising Generative AI in sales and service has the potential to reduce costs and improve productivity, resulting in happier customers and more engaged staff. The degree of impact depends on how deeply the technology is integrated into customer interactions and staff processes, which will be an organisation-specific decision based on cost, risk and perceived customer and staff satisfaction.
AI powered chatbots are already utilised by general insurers triaging digital and phone-based customer enquiries to reduce costs to serve and decrease response times. However, current chatbots are widely considered limited in their ability to handle more complex tasks.
Generative AI powered chatbots represent a significant step up. Using company data, they can be trained to give bespoke responses that build upon the context of a conversation. They can also be given a specific organisational tone, which can match the values and behaviours of the company. 
Once integrated into systems, Generative AI tools could be used to augment human staff by providing analysis, transcribing call notes, and supporting the drafting of responses. Over time, they may improve and move into automating more complex workflows, like updating policy details.
The technology could also help new staff get up to speed using personalised training tools at considerably less cost than traditional approaches – think of Duolingo, a language learning platform, which has released a personal tutor feature in collaboration with Open AI. It costs roughly the same per month as an hour from a real-life tutor. AI’s ability to parse and understand huge amounts of the spoken word could also augment quality control processes for sales and service teams at contact centres.
While claim processing is already at various degrees of digitisation depending on the insurer, Generative AI tools promise a higher degree of automation if trained on prior claim data and outcomes. The potential benefits are speed, accuracy, consistency, and reduced costs.
For small value claims this degree of automation is likely to be desirable, whereas, for high value and/or complex situations, the technology is more likely to be beneficial for augmenting human activity, such as providing initial analysis and drafting responses.
There is also the potential for Generative-AI to be used for negative pattern matching on prior claims data to help flag potentially fraudulent claims. Claim fraud was estimated to cost New Zealand Insurers over $739M in 2021.
Any improvement in understanding claims drivers is likely to benefit risk assessment and underwriting decisions. Generative AI can help underwriters uncover patterns and trends which could augment existing risk models and improve productivity.
The degree of benefit that Generative AI will have for underwriting is likely to depend on how reliant the process is on the technology and how successful it is at improving underwriting functions. Initial perspectives are that AI will be a useful co-pilot for empowering underwriting teams by augmenting decision-making.
Product and marketing
For product teams, Generative AI could provide an internal “expert” which consistently evolves and updates products as they change in the required systems and answers key queries from staff. The tools could also be used to re-write or draft policy documentation, ensure compliance with regulations, or analyse competitor products. For new product development, Generative AI could offer a source of ideas and create detailed concepts to enable more extensive user testing before releasing products.
For marketing teams, Generative AI could reduce the time spent drafting communications or creating new digital assets and imagery. It might reduce reliance on agency partners or internal design resources. For organisations with rich data sets on customers and prospects, Generative AI is likely to enable more personalised targeting and performance insights, creating a flywheel to drive conversion.
What are the key risks?
1. False responses
Generative AI technology can create false responses outside its training data. For example, in one instance, a lawyer used Chat-GPT to write an opinion, and it was later found out by the presiding judge that the tool had generated false case law to bolster the work. In another, a Swiss Insurer, Helvetia, has a digital assistant utilising Open AI’s technology, but a disclaimer deems responses are “not legally binding” due to this issue. These sorts of bugs need to be ironed out before the technology could be used in a more integrated manner for areas like claims and underwriting, or without a degree of human co-pilot oversight.
Generative AI models are a reflection of the data they are trained on and can often amplify any biases present. For example, a popular image-generation model called Stable Diffusion caused controversy when a Bloomberg report found it was over-emphasising racial and gender disparities, in areas such as careers and criminality.
Such bias could potentially result in inequitable outcomes or incorrect recommendations in claims and fraud decisions. Testing results from models will be important to assess and address the possibility of bias.
3. Black box
Building on the above issues, Generative AI models suffer from what is known as the ‘black box’ problem. We understand the data that goes into them and the responses the algorithms come up with, but we don’t know enough about the decision-making process. This is part of the way the technology is designed — it is supposed to come up with its own types of pattern recognition – however in regulated environments like insurance, compliance requires decision-making processes that are aligned to regulatory guardrails. This risk is likely to support the co-pilot deployment model.
4. Data privacy
Currently, Generative AI tools are being used in informal ways that can occasionally put confidential customer and company data into external datasets. This data is then used to train the models of the future. Several corporates including JP Morgan and Verizon have banned Chat-GPT use by staff for this reason. For any enterprise use, general insurers need to be clear on how the information staff input will be used by the technology provider in the future.
5. Negative customer responses
Customers are exposed to an increasing number of digital experiences across industries — and some prefer it.
However, any ceding of the customer experience to Generative AI needs to be cognisant of customer preferences. For example, while chatbots are already widely adopted by companies recent studies highlight significant customer dissatisfaction, including perceptions of cost-cutting. Companies will therefore need to tread carefully to ensure that the technology delivers benefits to customers as well as to themselves.
How to implement
Those adopting this technology will need to be cautious, but willing to take small risks and ensure they are consistently getting it right internally while aligning with regulatory and customer expectations before proceeding further.
Firstly, insurance businesses need modern existing technology to enable them to work with Generative AI most optimally. The risks of staff using confidential data in third-party web applications have been highlighted. A study showed 68 per cent of staff lied to bosses about using Chat-GPT, so while it is important culturally to support this innovation, it’s also important to educate staff on its risks.
Companies are wise to embrace the new releases of Windows and Google office software to give staff integrated access when possible. If insurers’ core operating systems are not modern, it is highly unlikely they will be able to successfully leverage Generative AI to the degree anticipated. The features and interoperability would be expected from modern cloud applications and are unlikely to be supported by the likes of legacy policy, claims and CRM systems.
Technology + Human Capital
For building more complex tools, eg. automated claim processing — insurance companies will not only need modern platforms but strong enterprise data capability to record, label and process claims data before training the AI on it. This kind of work is going to require AI expertise to fully understand and realise the new technology, internal technology experts are needed to streamline existing data and integrate it into current platforms, alongside business stakeholders who can ensure alignment with current processes and requirements.
Culture of innovation
McKinsey surveys show that most organisations achieve less than one-third of the impact they expect from digital transformation. Even with modern technology platforms and the talent to leverage Generative AI, it is still necessary to have a culture that encourages innovation. Without such a culture, it is unlikely companies will be successful.
As with any emerging technology, companies need to be able to experiment with the technology in the organisational context to learn what works and what doesn’t. The number of risks we have discussed shows companies should be cautious, testing significantly before rolling out live solutions.
A leadership culture of accepting failure encourages teams to push boundaries, taking smaller risks to learn before scaling up. The roadmap to unlocking the value of new technology is not a straight line of investment to business value, so patience and risk-taking are required.
While the benefits might take longer to be realised and the cost of significant experimentation might be larger, this approach will increase the likelihood of meaningful success over the long term. Companies should test Generative AI in the most obvious areas, building organisational capability with the technology before trying more complex solutions.
The nature of Generative AI means that the models are ever-improving as new training data becomes available and training methods are iterated. The release of GPT-4 in March 2023 was estimated to be almost five times better than its predecessor.
This essay has argued that the benefits of adopting Generative AI outweigh the risks for general insurers, and it is more important to grasp these benefits given this context.
Having the right technology and internal capability to begin with, while embracing a culture of innovation and taking an experimental approach will give those who do implement the technology the best chance at successfully leveraging it. There is also a risk of doing nothing – ignoring the opportunity to innovate with potentially transformative technology hands an advantage to your competition.
However, technology alone is unlikely to be the sole reason for the success of general insurers in the future — the $5 trillion USD lost in the Dot-Com era highlights the mistake of blindly embracing new technology without viable business models. That is not a sustainable long-term strategy.
General insurance is an industry that will always be grounded in the real world alongside an increasing number of ones and zeroes. Customers will continue to respond to those companies that deliver the best experiences at the most crucial times.
This requires human empathy beyond the purview of current AI — 44 per cent of customers still prefer to interact with a human being once a policy is in place. If insurance constitutes a promise to be there for customers when things go wrong, technology can only go so far in supporting that delivery.
Deloitte research shows purpose-oriented companies have higher productivity and growth rates, along with a more satisfied workforce who stay longer. To win in the future, Generative AI is likely to be a piece of the puzzle but not a silver bullet. Most general insurers are likely to have access to similar technology from software providers.
Those businesses that succeed will likely be the ones with the most engaged employees who build and deliver the best experiences; aligned with a customer centric, value led way of doing business.
 ChatGPT sets record for fastest-growing user base - analyst note | Reuters, (Krystal Hu, February 2023)
 Google’s AI chatbot Bard seems boring compared to ChatGPT and Microsoft’s BingGPT - Vox (Shirin Ghaffary, March 2023)
 Generative AI Is Generating Hype But Can It Generate Value? (forbes.com) (Lawerence Wintermeyer, Oct 2023)
 What is Generative AI & Why is It Important? | Accenture (accessed November 3, 2023)
 ChatGPT's future: 5 jaw-dropping things GPT-4 can do | CNN Business (Samantha Kelly, March 2023)
 AI-powered messaging: The uptake of chatbots in insurance (hannover-re.com), (Johannes Seebacher, May 2022)
 All you need to know about Generative AI Insurance Chatbots (masterofcode.com) (accessed November 2023)
 How to Train Your ChatGPT — A Marketer’s Guide (hubspot.com) (Ankit Vora, October 2023)
 Duolingo Max Uses OpenAI’s GPT-4 For New Learning Features (accessed November 2023)
 https://www.leewayhertz.com/generative-ai-in-insurance (accessed November 2023)
 5 Ways Generative AI Will Transform Claims | Insurance Thought Leadership (Jeff Gurtcheff, September 2023)
 Pitchbook Q2 2023 Emerging Tech Future Report, Generative AI (Robert Le, accessed November 2023)
 An Interview with Jamie Yoder | Insurance Thought Leadership (accessed November 2023)
 How Insurers Can Use Generative Artificial Intelligence (oliverwyman.com) (accessed November 2023)
 Marketing and sales soar with generative AI | McKinsey (accessed November 2023)
 A lawyer used ChatGPT and now has to answer for its ‘bogus’ citations - The Verge (Wes Davis, May 2023)
 AI-powered messaging: The uptake of chatbots in insurance (hannover-re.com), (Johannes Seebacher, May 2022)
 Generative AI Takes Stereotypes and Bias From Bad to Worse (bloomberg.com), (Leonardo Nicoletti & Dina Bass, accessed November 2023)
 Employees are entering sensitive business data into ChatGPT | SC Media (scmagazine.com) (Stephen Weigand, March 2023)
 Creating Customer Service Bots That People Don’t Hate (hbr.org) (Johannes Boegershausen, Noah Castelo, Christian Hildebrand, and Alexander P. Henkel, October, 2023).
 68% of ChatGPT Users Admit Hiding It From Their Boss (tech.co) (Isobel O’Sullivan, February 2023)
 Using AI to speed up the insurance claims journey | InsurTech Magazine (insurtechdigital.com) (Alex Clere, July 2023)
 Advantages of Innovation - 2023 Most Innovative Companies | BCG (accessed November 2023)
 Dotcom Bubble - Overview, Characteristics, Causes (corporatefinanceinstitute.com) (accessed November 2023)