Original text here from Patrice Bernard (LinkedIn)
Executives at financial institutions — and even more so at insurance companies — dream of what artificial intelligence promises, yet those ambitions are hard to reconcile with their deep-rooted aversion to risk. So when Allianz announces a strategic partnership with Anthropic, the scope of what it encompasses feels rather underwhelming.
From the outset, the tone is set: the collaboration’s goal is to establish a responsible use of AI, and the choice of vendor is determined, at least in part, by this essential criterion, which Allianz says it shares. The arguments put forward are familiar and difficult to dispute. Insurance operations demand a level of security and reliability that ensures regulatory compliance and preserves customer trust. But is their practical interpretation overly restrictive?
Subscribe to our newsletter:
To support its official communication, Allianz highlights three broad areas in which it intends to benefit from artificial intelligence — generative AI, in practice. The list holds no real surprises, forming a well-worn trilogy: assistance with software development, process automation (largely dependent on human tasks), and traceability of actions performed by automated tools. Employees, moreover, are expected to remain firmly in control of the applications.
This selection is easy to understand, as it aligns neatly with the industry’s ingrained culture of caution. In particular, the chosen implementation conditions lend themselves to continuous oversight, limiting as much as possible the potential missteps of machines. Unfortunately, they barely scratch the surface of the opportunities that AI typically opens up in the insurance sector.
Worse still, one can question the real benefits of the initiative. Several observers have begun to cast doubt on the productivity gains expected from software development projects, as AI-generated code often contains numerous errors and security flaws. Faced with such risks, insurers are likely to respond by introducing stricter testing and validation mechanisms, eroding a significant portion of the anticipated savings.
As for process optimization, I remain just as puzzled: a few years ago, specialized vendors were touting spectacular results from their RPA platforms. Did those efforts fall short? And if so, how will this new generation of solutions perform any better? Especially when the only truly viable path in this area, as I have argued before, would be to redesign processes from the ground up.
Finally, additional — and likely resource-intensive — efforts will be devoted to meeting transparency and traceability requirements imposed on AI systems. In the end, the initiative’s profitability is open to question… and one wonders whether Allianz conducted an objective return-on-investment analysis beforehand, or whether it is simply following the trend, carefully avoiding higher-value applications deemed too risky or too futuristic.
