2025: The Rise of Professional-Grade AI in Insurance

Last year, I called 2024 the Year of the Copilot. I also made a broader prediction: we were approaching a necessary shift - away from AI hype and towards systems that actually enable professionals to do their best work.  

In 2025, that shift didn’t just arrive - it accelerated rapidly. 

Just as importantly, the industry began to draw a much harder line between what works in AI and what doesn’t. 

Today, the question is no longer whether you can use AI for a given task.  It’s how you are driving material value from AI - and what systems you’ve built to deliver impactful, reliable, and repeatable outcomes.

We have now entered the Professional-Grade AI Era. 

Separating the Wheat from the Chaff

Last year, I predicted that: 

• Improvements in large language models would continue - but that the leaps would slow.
• We would see a surge in “POCs to nowhere”. 

Both came true in 2025. 

While foundation models became more accessible and capable, the incremental gains began to decelerate.  Scaling laws continued to deliver gains, but the biggest breakthroughs increasingly came from reasoning models - which often come with significantly higher inference latency and cost.  At the same time, AI luminaries ranging from Yann LeCun to Andrej Karpathy cautioned that the current generation of LLMs remains at least a decade away from the promised future of autonomous agents.

Meanwhile, enterprises poured resources into AI pilots that looked impressive during demos - but failed to survive contact with reality.

A widely cited study from the MIT Sloan Management Review  suggests that up to 95% of enterprise AI initiatives failed to deliver intended business value. We have seen this firsthand across insurance: demos that wow, only to whimper in production.

This was not surprising. 

A critical challenge with AI - especially transformer-based large language models - is that they are non-deterministic. You can provide the same input and get different outputs.  This unpredictability makes AI difficult to control and even harder to trust in high-stakes, highly regulated environments like insurance. 

AI demos impress but real environments demand accuracy, explainability, and resilience. 

‍AI Systems, Not Science Projects

In 2025, a wave of quick-fix AI applications emerged across domains ranging from law to medicine to insurance. For example, many tools promised insight through simple document ingestion and LLM parsing. 

The reality in insurance is that simply feeding policy or submission documents into ChatGPT, asking a question, and hoping for the best is not enough to drive real improvements in GWP growth or reductions in combined ratios.

In production, these approaches often deliver 50-70% accuracy - sometimes proudly marketed as success. But delivering 70% correctness on tasks that demand 99.5%+ is a non-starter in underwriting. The stakes - and the standards - are simply too high. 

This was an important lesson of 2025: Models alone can’t deliver business value. But AI systems can. 

2026 and Beyond: Professional-Grade AI

Looking ahead, the most advanced organizations will not think in terms of single models. They will deploy layered, hybrid systems: combining probabilistic models and deterministic logic, layering controls and fault tolerance, automatically orchestrating workflows to decide when to automate and when to bring in a skilled human, and continuously learning from outcomes. 

This approach will become dominant in 2026.

We will see AI infrastructure begin to resemble systems engineering in fields like aviation, medicine, and civil engineering - disciplines that have long known how to operate effectively under uncertainty. The goal will be to make AI not just intelligent, but stable, resilient, and explainable - and, when necessary, able to fail gracefully.

Going forward, it isn’t about smarter models, it's about smarter systems - systems that detect issues, correct themselves, and improve over time. It’s about building trustworthy systems that work every day, for everyone, in production. 

That’s exactly what we’ve built at Kalepa.

Kalepa’s 2025 Highlights: AI That Delivers in Production

At Kalepa, we have always taken a systems-first approach. This year, we advanced that vision across every layer of our AI Underwriting Platform: 

• Launched the next generation of the Kalepa AI Ensemble Engine: A purpose-built framework that understands context and intelligently selects the right model or method at the right moment to deliver the most accurate recommendation to an underwriter. This infrastructure doesn’t just improve performance; it adapts around uncertainty - deciding when to automate, when to augment and how to balance both.
• Expanded strategic tools for underwriting and distribution leaders: Enabling leaders to effortlessly analyze their portfolios and automatically receive recommendations on product, rate, and appetite opportunities - all based on empirical data in real time. 
• Expanded full lifecycle support: Adding AI-native functionality for quoting, binding, rating, document intelligence and forms management, making Kalepa the hub for end-to-end underwriting workflows.
• Earned industry wide recognition: Including the 2025 InsurTech100 and AIFinTech100, as well as being named one of Wellfound’s Top 10 AI & ML Startups, alongside leaders such as Anthropic, Writer, and Weights & Biases. 
• Delivered Professional Grade AI to global insurers: From tier-one carriers to innovative MGAs, our partners are turning to Kalepa to centralize data, improve risk selection, bind more policies, and drive portfolio profitability.  See how we did this for Bowhead Insurance. 

Each win shares one common thread: Professional-Grade AI, tested across millions of risks and delivering measurable results to the world’s top insurance organizations.

Raising the Bar for AI in Insurance

Across the industry, insurance leaders are waking up to the risks of poor AI - and the upside of doing it right. The shift from models to intelligent AI systems is already underway, and it will define underwriting success in the year ahead. 

At Kalepa, we’re proud to stand at the forefront of this shift - building the infrastructure, tools, and outcomes that help our partners to not just keep up, but to lead.

We’re excited for what’s ahead in 2026!

-Paul Monasterio, CEO and Co-Founder at Kalepa

‍