Let’s be honest for a second. If you’ve worked in insurance for more than a week, you’ve probably had a moment where you’ve stared at a screen, waiting for a legacy system to do its thing, and wondered, “How are we still using this?”
These core systems are the heart of our industry, but they often feel like ancient relics. They’re a tangled mess of code written decades ago, with business rules that are barely documented and custom patches that only one person who retired five years ago truly understood.
We’ve all known for years that these platforms need a major overhaul. But the sheer cost, risk, and complexity of a full-blown modernization project? It’s enough to make any executive’s blood run cold. So, we kick the can down the road. Again.
But what if we’re finally on the cusp of a real solution? A recent report from the folks at McKinsey’s Financial Services Practice just dropped, and it’s got some fascinating ideas about something called "agentic AI." And the numbers they're throwing around are pretty staggering—we're talking productivity gains of up to 90% in some areas.
No, that's not a typo. Let's dig into what this actually means.
So, Why Are We Still Stuck with These Old Systems?
Before we get to the AI part, let's talk about why this problem is so hard to solve in the first place. It’s not just about writing new code. In fact, that's often the easiest part.
Think of your core system like a historic house that’s been added onto for 50 years. There’s old wiring next to new wiring, plumbing that goes nowhere, and a few load-bearing walls that nobody wrote down in any blueprint. You can’t just start knocking down walls without understanding how the whole thing is held together.
That’s exactly what happens in insurance. According to McKinsey, a huge chunk of time and money in these modernization projects goes toward things like:
- Discovery: Trying to figure out the thousands of undocumented product rules and actuarial settings buried in the old system.
- Data Conversion: Moving data from the old format to the new one without breaking everything.
- Testing & Reconciliation: Making absolutely sure the new system produces the exact same results as the old one.
- Cutover: The terrifying moment you flip the switch, praying it all works.
This process is so painful that it creates what McKinsey calls a “double-bubble” period. That’s when you’re paying to keep the old, creaky system running while also paying for the massive team and resources needed to build the new one. It’s a financial nightmare, and it’s why so many of these projects stall out.
Okay, But Isn't This Just Another AI Buzzword?
I get it. We're all a little tired of hearing about AI. But this "agentic AI" is a bit different from the tools you might be thinking of.
This isn’t just a developer “copilot” that helps a programmer write code faster. Think of it like this: a copilot is a really smart assistant. It can suggest the next line of code or find a bug for you. It’s helpful, but the human is still very much in the driver's seat.
An AI agent, on the other hand, is more like a project manager you can delegate a whole goal to. You can tell it, “My goal is to migrate this block of life insurance policies from System A to System B.”
The agent then gets to work on its own. It breaks that big goal down into smaller tasks, figures out what tools it needs, and starts executing. It can read the ancient code from the old system (even archaic languages most people don't know anymore), translate the logic into plain English, map the data, and even run tests to make sure everything lines up.
This is where the truly mind-blowing gains come from. McKinsey says an agent can do in days what might take a human expert months or even years. Imagine being able to reverse-engineer an entire product's logic without having to hunt down a subject matter expert who’s been with the company for 30 years. That’s the promise here.
How to Actually Make This Work: Three Big Shifts in Thinking
Of course, you can't just unleash a bunch of AI agents on your most critical systems and hope for the best. The McKinsey report highlights three crucial shifts in mindset that insurers need to make to capture this value.
1. Think Modular, Not Monolithic
First, don't try to build one giant, all-powerful AI to do everything. Instead, build a library of smaller, specialized agents.
Think of it like a set of Lego bricks. You might have one agent that’s an expert at reading COBOL, another that excels at validating data, and a third that’s a pro at running reconciliation tests. You can then snap these modular agents together in different combinations to tackle different parts of the project. This approach gives you way more control, makes the outputs easier to audit, and lets you reuse your "Lego bricks" across different projects.
2. Treat Modernization as a Portfolio, Not a Single Project
This might be my favorite insight. Because these AI agents can be reused, the cost of modernizing the next product or system can drop significantly.
This changes the entire equation. Instead of one massive, terrifying, all-or-nothing migration, you can start to think of modernization as a portfolio of opportunities. Maybe you use the agents to completely migrate your main auto policy admin system. But for that small, legacy block of business on the side, maybe you just use them to selectively rewrite a few key applications. It gives technology leaders a flexibility they've never really had before.
3. Redesign Your Governance for a Human-AI Team
Finally, and this is critical, you have to remember that a human is still in charge. This isn't about letting AI run wild, especially in a regulated industry like ours.
This means building new workflows with human oversight baked in. You need clear "human-in-the-loop" approval points where a person signs off before the agent moves to the next stage. You need rock-solid traceability, so you can see exactly how a requirement led to a specific configuration and the test that proved it worked. And you need clear rules for how you validate and manage these AI models.
It’s a big change in mindset, for sure. But for the first time, it feels like we might have a tool that’s actually up to the massive task of untangling our core systems. It’s less about a "big bang" and more about smart, continuous, and manageable improvement. And honestly, that sounds like a much better—and less terrifying—way to finally move our industry forward.
