It's little wonder that "AI governance" is a term on everyone's lips wherever business technologists gather. With great power comes great responsibility, and we are all aware of the tremendous capacity for change that is inherent in AI, and agentic AI especially, but also the potential for rogue agents and human errors that can cause catastrophes. That goes double for an industry like insurance that has so much systemic impact for individuals, enterprises, countries, and economies, and, as is appropriate, is highly regulated as a result. The good news is that I believe insurance can move very fast so long as players don't neglect necessary defensive measures.
I stress the relevance of AI governance specifically to agentic AI because it is so much cleverer and more capable of true business reengineering than generative AI. Generative AI lets us parse content and ask and answer questions with blistering speed, but agentic AI provides an opportunity that is without precedent. It has the reach, performance, and access to resources to create better organizational (and inter-organizational) workflows and processes.
The results of this are amazing even to veterans of the sector that may feel they have seen everything. I'm referring to the removal of human drudgery, as the gears move faster and act with greater accuracy, alongside the chance to delight customers and partners, and liberate people to work more closely with customers while igniting creative, collaborative thinking.
Barriers to progress
But there are challenges and constraints. Making relevant data available to AI agents involves a certain level of sensitivity and understanding of data protection, privacy rights, and ethics.
Bill Gates is surely right when he argues that:
"We have no experience with a technology that can be adopted quickly or that can think and move like a human. When the PC came along, it took twenty years to significantly change how we worked because the software had to be developed, the price had to come down, and people had to learn how to use the tools and incorporate them into their business processes. AI, on the other hand, runs on the devices we already have, and it uses natural language. We don't have to adapt to it because it can adapt to us. It can watch the same training video that is used to train human workers and learn from existing data."
Gates says this new AI era will be "one of the most turbulent times in human history," and he calls for "human-reserved" roles to ensure that chaos does not ensue, thanks to bad prompts or machines acting in ways we had not anticipated.
I echo all of these thoughts, and I think we can have it both ways. That is, transformation without trauma.
Crash barriers must be in place
Essentially, the remarkable autonomy of AI agents means we need more guardrails, and we will always need the 'human in the loop' to oversee what our brilliant digital minions are doing.
But we can't be Luddites. A concise shortlist of the advantages of agentic AI includes turbo-charged productivity, better-prepared (and auditable) executive decision-making, faster claims processing, and customer experiences that delight customers and encourage them to evangelize on behalf of the insurer serving them.
And it's hard to overstate the scale of what is possible in the agentic AI world. This isn't about trimming a fraction of a cent or penny off costs, and neither is it about incremental change. We are talking about new insurance products coming to market in a couple of days (or, say, a month when marketing campaigns and everything else is aligned) rather than six, eight, or 12 months. That is the sort of acceleration and cost arbitrage that even the dotcoms didn't see with the advent of e-commerce.
So, we have an opportunity for seismic change in how insurers operate and accelerate. Brokers are more likely to go in steps, but we are still talking about genuine transformation, offering faster response times to inbound sales queries and service desk calls, as well as an ability for brokers to absorb sharp growth without adding to the employee roster.
There are other benefits. Standardization of processes becomes simpler, and continuous improvement will occur organically as agents learn. And of course we know that agents excel in other ways that humans don't: their lack of mood swings, health issues, need for holidays (or even breaks), and so on.
Building trust
As processes become refined and we become more trusting of our robo-companions, we will be able to confidently rely on agents to act in autonomous ways, with the human in the loop always there as a check and balance and ultimate arbiter.
To reach that point, though, we need to school our agents by providing them with the institutional knowledge the insurance sector has built up over years, decades and centuries. By creating a super-brain of knowledge management and making that available in machine-readable data, we can reduce the human element so that we can act more like observers and referees seeking out anomalies and red flags rather than micro-managers.
We will reach the point where we only require human oversight for bigger deals or to handle exceptions and 'black swan' events. My estimate is that we will reach that stage in up to four years; although, such is the rate of progress, it could be more like one year.
Guardrails will still apply and that's why we advise customers to spread workloads across agents to validate stability and create a separation of duties. Just as we wouldn't want to have one member of staff only holding the keys to the kingdom, we must use agents in tactical ways, restricting their movements and allocating unique tasks so the risk of a single point of failure is eliminated.
We also want alert human beings who use their experience to detect risks as opposed to the narrow focus of an agent on task completion. We will rely on a combination of humans and agents to understand regulatory change and what are likely to be the next steps of industry lawmakers.
We need to ensure the insurance industry can manage risk, and there is an equitable relationship between technology provider, buyer, and end-user. Today, we have a legislative grey area, so we must be in listening mode and have the sorts of interpersonal meetings that are not (yet!) the domain of agents.
Similarly, we must be honest with our people. Agentic AI will take some jobs, but it will enable others, so we need to create a learning environment where new skills can be added.
And finally, insurers must be certain their choice of AI technology providers is good. Not all providers in what has become a febrile space will last, and not all will have the ethics or reach across the IT ecosystem to succeed as long-term players.
We need to understand the trade-off at play. The power of technology always comes with a price tag that goes beyond the cost of acquisition and maintenance. Technology is not a silver bullet, and it can be used for good and bad. We need to move fast to gain an early-market advantage but without sacrificing protections that require the human touch.
