Generative AI in Practice is InsTech’s monthly newsletter dedicated to the use of generative AI in insurance.
The InsTech Perspective
Coding mistakes and prompt engineering
Almost a year into the release of ChatGPT and the practical business use cases are starting to emerge. We cover some of these in today’s issue, such as chatbots and assessing claims. But generative AI isn’t just about extracting information or creating human-like responses to clients’ questions.
Programmers previously used communities such as Stack Overflow to get help from other friendly coders when they got stuck writing code. Now ChatGPT is offering a faster alternative for both professional and amateur software engineers to write new code or identify and fix errors. An article in Techmonitor.ai suggests that coders prefer this to the human helpers in Stack Overflow but warns that ChatGPT gives the wrong response over 50% of the time. Should we be worried?
We’re doing a bit of coding ourselves and trying out ChatGPT (Matthew Grant reveals his renewed interest in that dark art here). It’s quite impressive. It’s true that at times ChatGPT makes mistakes but these usually occur if the question (known in GenAI terms as the “prompt”) is too complicated, too vague or lacking in context. ChatGPT’s errors are usually small and can be quickly fixed by copying the error message back into ChatGPT and asking for help. Like all emerging technology we need to be realistic about what can be achieved in a short time and recognise the important role of humans in the loop. Writing code without making errors is like trying to learn to ski without falling over. It’s a natural part of the learning process. Just make sure you have a soft landing.
Use case of the month
Distribution at scale
Distribution, the term insurers use to describe how they sell insurance to customers, is one of the areas we’ve seen insurance organisations implementing generative AI today. Although insurance products in some countries are sold online through websites and aggregators, in most cases the sales process involves manual back-and-forth communication between client, broker and insurer through emails or phone calls.
For personal lines insurance, a simple example of how to streamline this process is using online chatbots through which consumers can learn about products and get covered. These have existed for a few years, but integrating generative AI can make them more capable at answering difficult questions with human-like responses. Chatbots are also being used for commercial insurance, with cyber insurance MGA Cowbell now providing a GPT-powered chatbot that helps brokers and policyholders receive guidance on risk assessment, cybersecurity and the claims process.
Generative AI can also be used as an internal productivity tool for insurers and brokers looking to scale up their marketing and sales. In this issue, Zywave’s CEO explains how the company is helping brokers and insurers write marketing emails with an AI-powered tool, whilst a broker reveals that it uses ChatGPT to help explain different insurance products to its customers.
Guest Perspective
MIC Global: 4 focus areas for GenAI and 2 live use cases
Harry Croydon and Jonathan Croydon from start-up insurer MIC Global, which launched a Lloyd’s syndicate in 2022, explain how MIC is using generative AI for customer service and operations. They also suggest four immediate areas of focus for the insurance industry.
What the insurance industry is doing with GenAI
QBE Ventures’ Head of Emerging Technology Alex Taylor sees the most immediate potential to use generative AI for extracting insights from unstructured data in submissions and claims.
Crawford & Company is using generative AI to triage claims automatically in a pilot in the US.
Australian life insurer TAL is testing generative AI use cases in its internal helpdesk and creating marketing content through a partnership with Microsoft.
Malaysian insurer Tune Protect is using ChatGPT-powered chatbots to engage with customers in the claims process.
Tata Capital, the financial services arm of Tata Group, has enhanced its existing online voice chatbot with OpenAI’s GPT. Customers can speak to or message the bot to resolve queries related to insurance or other financial products.
The Auto Club Group, a US automobile association that sells insurance, will use generative AI to allow its agents to retrieve information about insurance policies. The company is using Salesforce’s AI Cloud product to do this.
Scale-up insurer Lemonade says it can now deploy “fully compliant generative AI capabilities at scale” as it looks to improve operational efficiency. There are differing opinions about Lemonade’s business model.
Canadian broker Mitch Insurance is using ChatGPT to help create concise overviews of complex coverages and pitch products to clients.
News from the InsTech network
At InsTech’s next evening event in London on 19 September, we’ll be joined by Cytora and Google Cloud to see a demo of generative AI in action digitising insurance submissions.
Zywave, whose software helps insurers and brokers distribute digitally, is using generative AI to help companies with marketing emails and HR advice.
A video from Eigen Technologies demonstrates how an underwriter could use generative AI to answer questions about a complex submission.
Salesforce has launched a tool to enable companies (such as insurers) to use their own AI models within Salesforce applications and train models on their proprietary data.
Instabase’s blog post shows how LLMs can help actuaries analyse large volumes of data accurately and efficiently for portfolio risk analysis.
EXL says insurers are keen to deploy customer-facing ChatGPT-style chatbots, but should start with internal use cases for generative AI first.
Consumers of all age groups have “high trust levels in generative-AI-based interactions”, according to a Capgemini report that surveyed 10,000 consumers in 13 countries.
EY reports that it believes most property and casualty insurers will prioritise claims management when adopting generative AI.
Generative AI can be used by criminals to make phishing more convincing, resulting in more effective cyber attacks, according to Chubb.
Indico Data, which uses generative AI to help insurance organisations ingest unstructured data, has joined the InsTech network. Indico’s report shows how it helped MetLife unlock value from its unstructured data.
Other news and insights we’ve seen
63% of insurance executives are investing in AI and machine learning in the UK, compared to just 50% of their US counterparts.
Insurance Thought Leadership’s Paul Carroll writes about the ten biggest mistakes in AI strategies that companies should learn from.
Who is responsible for subsequent liability if directors use generative AI to help with corporate decision making? Law firm Kennedys writes about the risks of AI for D&O insurance.
How InsTech has used AI this month
InsTech’s research team maintains a database of insurance technology companies, called ATLAS. We use ATLAS to help our insurance corporate members identify potential clients and partners. We’ve now developed our own generative AI tool to keep the database updated and increase the capacity of our research team (allowing them to write more newsletters like this one). Every time someone at InsTech comes across a new company, they enter the web address of the company into “AtlasBot”. If the company is already in our database, AtlasBot returns the relevant entry. If not, it searches third-party data and the company’s website and uses OpenAI’s GPT model to write a draft database entry. The AI-generated company profile is automatically populated into ATLAS and labelled for the research team to review and finalise.
Generative AI in Practice is InsTech’s monthly newsletter dedicated to the use of generative AI in insurance – you can sign up for free here.
