Can ChatGPT make insurance agents better salespeople? That’s the bet that one insurtech in Hong Kong is making, despite facing technological and regulatory questions.
PortfoPlus is a startup that launched in 2018 initially with a policy wallet that agents could use to aggregate a customer’s insurance products. That would give agents a holistic view of a customer’s insurance and wealth profile, which would give the agent an edge in providing advice and pitching products.
Since then the team has rolled out other tools designed to help agents and it has about 4,000 users who pay for its software-as-a-service, says Colin Wong, CEO and co-founder (pictured, center).
“ChatGPT is the next function,” he said.
Reinventing insurance agents
Wong has worked both as a salesperson at an insurance broker as well as an engineer at IBM, after studying computer science at Hong Kong University. He and his co-founders Phoenix Ko (pictured, left) and Hugo Leung (right) got the startup bug.
“We want to do more than help salespeople,” Wong said. “We want to change the traditional industry and how sales agents operate.”
Ultimately that means using technology to enable them to better serve customers rather than just sell products with high commissions. The co-founders know this is easier said than done, as they are salespeople too.
The nature of GPT artificial intelligence, however, has the potential to change the incentives to become more customer-friendly.
GPT’s kung fu
GPT stands for generative pre-trained transformer. It’s part of a family of language-learning models (LLMs) that use big-data inputs of sequential data (like the beginning of a sentence) to predict what should follow. It’s being used to produce text and translate languages. Software firm OpenAI was the first to introduce this commercially with its ChatGPT chatbot.
The power of GPT lies in access to vast data sets along with self-learning as more people use it. It doesn’t create original text, but rather amalgamates human writings to guess answers, but its authoritative tone makes it influential, even if sometimes its responses are contradictory or wrong.
The important part here is GPT’s reflection of a community of voices. The broader its usage, the better it self-learns.
For PortfoPlus, that means the more agents use GPT functions in their app, the better the responses.
“It’s like kung fu,” said Hugo Leung, co-founder. “Each agent is protected by a master,” which in this case means a big insurance company. “Each master fears exposing their disciples to the other schools. But now this could change, because the agents benefit by learning from one another.”
That doesn’t temper their competitiveness, but it does mean that the more agents use PortfoPlus’s ChatGPT plug-in, the better job it does for all of them. Ultimately, the “master” could become the app rather than the insurance company.
Modest beginnings
There’s a long way to go before that idea comes true, though.
The first use cases are incremental: use ChatGPT to generate messages for customers to help agents stay connected. For example, the chatbot can personalize a birthday message based on an agent’s profile of her customer in the PortfoPlus database.
“The core value of the agent is the warm touch,” Wong said. “Personalization helps them keep their relationship with their customers.”
That’s a modest start, but the technology could potentially become more important once it starts to be used by the end customers. PortfoPlus could help agents set up a website or app in which customers can ask ChatGPT about their own insurance needs.
Phoenix Ko, co-founder and head of business development, says customers are more likely to trust ChatGPT than an agent because people know that agents are biased in how they select products. ChatGPT, because of its natural tone and unscripted fluidity, can influence users.
“This will be a revolutionary sales assistant because it can help agents connect with clients and complete sales,” Ko said.
Personalization versus advice
The idea is not, however, to leave actual product choices to ChatGPT – which first of all is not a licensed broker (although one day…). Rather, the conversation would end in the app recommending the customer to an agent, who would come armed with the chatbot’s insights about the customer’s needs.
Wong says the startup is not ready to deploy this higher functionality. It is currently beta testing it with about 20 clients. The challenges are both technical and regulatory.
On the tech side, PortfoPlus has obtained the licensing to integrate ChatGPT via Microsoft Azure (Microsoft acquired OpenAI earlier this year for $10 billion). But heavy demand to use ChatGPT often leads to slow speeds and sometimes it crashes. So there’s a stability issue.
There’s also a tone issue. Users can set ChatGPT’s “temperature” from 0 to 1. Zero means the chatbot sticks to facts but its responses are formal and robotic. One means the chatbot unleashes its full creativity, making it far more engaging and a lot of fun, but prone to errors and fabrications.
Finding the balance requires more testing.
What’s regulated?
Then there is the regulatory issue. The Hong Kong Insurance Authority is likely to say that any entity that talks to clients with a view to influencing sales will count as a regulated activity.
To what extent this applies to a chatbot talking through a client’s needs is unclear.
Also unclear is how to draw the line between an engaging conversation that tees up an agent recommendation – and actual financial advice. PortfoPlus is a licensed insurance broker, which means it can facilitate transactions but it cannot provide advice. Setting the GPT’s temperature is part of this negotiation.
It’s early days but change is coming fast. Wong only heard about ChatGPT late last year. By January he had realized it could be vital to the startup’s future, but perhaps just as a marketing tool and some limited client engagements. Now the founders are looking at building businesses for agents, including training, and for customers.
Reinventing startups
The impact is going to be industry-wide, Wong said. “GPT is now plug-and-play. Any startup can use it with their own data sets, which means every startup is now an AI company.”
The competitive landscape is going to rapidly shift to those who are both good at fine-tuning GPT and that bring proprietary data to create a unique output. Until now, startups were shut out of this game because they didn’t have Microsoft-level access to datasets.
But with ChatGPT integrations, small companies can create unique products if they have niche but relevant data – in this case, PortfoPlus’s 4,000-odd agent users, whose activity on the app improves the GPT’s self-learning. And if they self-learn within a startup’s app, the users within that app mutually benefit.
“Today we serve agents and brokers,” Wong said. “In the future this will be about serving the interests of clients. It’s by tailoring services that you get around conflicts of interest.”
