What it takes to solve the AI paradox

Recently, I've seen thought-provoking studies highlighting an uneasy AI reality: despite surging investment and experimentation, many firms haven't seen big returns.

McKinsey calls this the “GenAI paradox”: some 80% of companies report adopting GenAI yet just as many have yet to realise bottom-line impact. Similarly, MIT-affiliated researchers have documented how 95% of enterprise AI pilots fail, producing zero return for organisations. 

This isn't a complete surprise. Productivity gains, even from breakthroughs, don't happen overnight. With AI, the paradox may feel sharper given rapid adoption raises our expectations.

Still, what these studies really highlight for me is that fragmented, standalone AI pilots won't yield results. Instead, companies should deploy this technology in their core processes, reimagining how these can work in a human-AI collaboration. Moreover, they should prioritise AI initiatives that can scale fast while aligning their tools with business and societal expectations.

With Swiss {ai} Weeks from 1 September to 5 October, with hackathons, workshops, and casual meetups to catalyse new ideas, the time is right to consider progress so far – and discuss resolving this paradox, in re/insurance and beyond.

At Swiss Re, we see essential building blocks as crucial to unlocking AI’s potential. The first is applying technology to real business challenges, with initiatives aimed at delivering tangible value for us and our clients, not simply innovating for the sake of it.

Just as critical is Swiss Re's robust data foundation, built on decades of investment in common platforms and enriched by more than 160 years of risk knowledge. For a decade, we've invested in reducing fragmentation, unifying systems, streamlining data flows, and eliminating hurdles that could hinder scaling AI initiatives. Our data and analytics platform, used by over half of Swiss Re employees, supports two dozen business domains, group functions, and specialised departments, managing thousands of inbound feeds with petabytes of information.

Further, reusable design patterns also promote scalability across business lines, regions, and client segments. We've all seen impressive GenAI demos that get stuck in pilot mode. At Swiss Re, we emphasise use cases proven in one team, geography, or dataset, but scalable across businesses, regions, or client segments.

Finally, robust governance frameworks ensure data assets are trusted and fulfil regulatory requirements. Ultimately, governance is both an enabler and a safeguard, helping us confidently integrate data into AI applications while ensuring compliance. This includes integrating humans in decision-making loops, which I see as key to building and keeping trust.

For re/insurers, where critical information is locked in emails, submissions, contracts, and claims files, AI turns fragmented, unstructured data into machine-readable assets. This addresses pain points like inconsistent submissions and accelerates decision-making as we focus on processes where AI can fundamentally change how we work.

As Swiss Re's AI journey develops, tangible examples of our strategy are emerging as we focus on core processes like underwriting and claims to avoid getting sidetracked by niche projects.

In Swiss Re Corporate Solutions, for instance, our ClaimsGenAI platform automates and streamlines handling of over 40,000 claims annually. Built on decades of data, ClaimsGenAI identifies patterns signifying fraud or recovery opportunities that manual reviews may miss. Clients benefit from more accurate claims attribution, avoided deductibles and reduced premium pressure.

Our AI-powered Underwriting Ease and Life Guide Scout help life insurance clients make faster, more accurate risk assessments. They both leverage Swiss Re’s holistic data resources, are scalable, and adhere to our governance approach. By augmenting human expertise, they help industry professionals focus, analyse and make better judgements for faster resolution, less friction, and better customer outcomes.

A holistic AI approach embeds AI into core processes, redesigning interactions. This goes beyond deploying one-off applications by combining AI techniques like GenAI, machine learning, rules-based automation, and optimisation models into coordinated systems.

Managing these interactions demands mature AI platforms with orchestration capabilities, trusted data foundations and strong governance. With these in place, we can take AI beyond today's prompt-response paradigm to one where agentic AI systems plan, act, and collaborate with other systems or humans to achieve business goals.

For our industry's complex workflows and legacy systems, agentic AI promises to be a game-changer, helping streamline operations without requiring wholesale overhauls. At Swiss Re, we're experimenting with these approaches and seeing encouraging results in manual, data-heavy areas like contract management and submission handling.

As we ride this next wave, I'm pleased McKinsey and MIT have gotten us talking about AI's challenges. It shows that a mature conversation is unfolding about moving from AI experimentation to scalable, impactful, responsibly governed, data-driven systems. I'm eager to hear your views during Swiss {ai} Weeks as we explore how to turn today’s paradox into tomorrow’s productivity.

Versions of this article were published previously in Insurance Day and finews.ch