AI in Insurance: Transformation Starts Below the Surface

Source: Originally published by Handelsblatt Live: “Three Questions with Christoph Krieg, Chief Strategy Officer at Peak3.”

What should a board of directors be most concerned about regarding AI today — and what is often underestimated?

I'm answering from a technology perspective. Anyone who views the successful scaling of agentic AI as an iceberg sees the AI ​​agents as merely the visible tip above the water's surface. What lies beneath determines success or failure: the agent orchestration platform, the data platform, core systems with AI-native interfaces, an AI-enabled development platform (AI SDLC), and curated AI knowledge bases.

This is precisely where the most significant underestimation occurs: Many executives treat AI as a matter of use cases and model selection; however, it is first and foremost a question of the foundation. Anyone who believes that agentive AI can be scaled without these technological foundations is bound to be disappointed. Pilot projects may excel, but scaling fails due to the legacy infrastructure. My thesis: Agentive AI should be the final impetus to finally tackle the long-overdue modernisation of legacy systems systematically…and not the next excuse to postpone it further.

Where is the gap between AI promises and measurable added value currently greatest?

Every consulting firm and every tech company touts the value proposition of agent-based AI. Many practitioners in the industry view this with a justifiable degree of cynicism. This is because the promise is usually conceived from scratch: a new, ideal operating model, drawn on a blank sheet of paper. The reality is long-established legacy processes, historically grown organisations, heterogeneous technology landscapes, regulatory requirements, and uncertainty.

From a technological perspective, the idea of ​​organisations largely driven or significantly augmented by AI agents is simply unrealistic as long as existing systems cannot efficiently provide their data and functions, as long as they cannot transact in real time, and as long as they cannot scale their parallel processing to the extent required by a hybrid workforce of humans and agents. It is precisely at this juncture, at the seam between the glossy vision and the real-world machinery, that the gap is widest today. It can only be closed through uncomfortable investments in the foundation.

How will we know in two years that AI has truly "arrived" in the insurance industry?

We will recognise it by three things. First, when AI moves beyond the selective optimisation of individual process steps and instead rethinks entire end-to-end processes, when efficiency gains (at least in individual processes) are no longer measured in 10-30 percent, but in 70-90 percent. Anything below that is automation in old clothing, not necessarily transformation. And third, when the executive suites have recognised and accepted that they can no longer postpone the fundamental transformation of the underlying systems.

In other words: AI hasn't arrived when everyone has a co-pilot. It has arrived when the first insurers have the courage to rebuild their engine rooms for AI…and the others realize they've fallen behind.

Coming up: Christoph Krieg will be at the Handelsblatt AI in Insurance Conference on 16–17 September 2026 at ERGO Group in Düsseldorf, Germany. Learn more and register: Handelsblatt AI in Insurance Conference.