Have you been in one of those meetings lately? The ones where everyone is buzzing about AI? The excitement is palpable. We’re all picturing it: a fully autonomous AI “agent” that can handle a claim from first notice to final payment, no human hands required. Or maybe an underwriting agent that can assess complex commercial risk in seconds.
It sounds incredible, right? It’s the future we’ve been promised.
But here’s the thing. A recent, and frankly quite sobering, piece of research from MIT dropped a bombshell on this whole conversation. It found that a staggering 95% of AI pilots fail to deliver any real value.
Ninety-five percent.
That’s not a typo. It means for every 20 insurance carriers pouring money, time, and talent into a new AI initiative, 19 of them are essentially lighting that investment on fire. It’s a tough pill to swallow, and it probably sounds painfully familiar to some of you.
So, what’s going on? Is the tech just not there yet? Is it too complicated? Surprisingly, the answer is no. The problem isn’t the flashy, futuristic AI agent. The problem is that we’re all trying to build the roof before we’ve even laid the foundation.
We’re Chasing the Race Car, But Forgetting the Engine
Let me put it another way. We’re all obsessed with the idea of the "autonomous agent." This is the AI that thinks, decides, and acts on its own. It’s the self-driving car of insurance. It’s sleek, it’s fast, and it’s the thing that gets executives excited and makes for a great press release.
But to get that race car to fly around the track, you need more than just a shiny chassis. You need an engine, a transmission, wheels, a steering system… you need all the boring, unglamorous, but absolutely critical components working together underneath.
In the world of AI, these components are the “tools.” And this is where that 95% failure rate comes from. We’re skipping the hard work of building the tools and jumping straight to the agent.
So, What Are These “Tools” We’re All Ignoring?
When I say "tools," I'm not talking about fancy new software. I'm talking about the fundamental building blocks that allow any advanced system to function properly.
Think of it like this: your dream is to have an AI claims agent that can instantly process a photo of a dented car fender.
For that to happen, the AI needs a few things first. It needs:
- A way to get the photo: Is there a clean, reliable pipeline from the customer’s phone to the AI? Or is it stuck in an email inbox somewhere? That pipeline is a tool.
- A way to know who the customer is: Can the AI easily access your policy administration system to verify the customer and their coverage details? That secure, fast connection (an API) is a tool.
- A way to understand the policy: Is the policy language structured and digitized, or is it locked away in a 50-page PDF? A system that can read and interpret that data is a tool.
- A way to access repair cost data: Does the AI have a direct line to your parts database or a partner’s repair cost estimator? That connection is a tool.
These are the nuts and bolts. They are the data clean-up scripts, the simple automation bots, the APIs, and the organized databases. They are, let's be honest, the boring part of AI.
Building a solid API doesn't get you on the cover of an industry magazine. Cleaning up a decade of messy customer data doesn't sound as cool as "launching a cognitive agent." But without them, your shiny new agent is like a genius who’s been locked in a soundproof room. It might have incredible potential, but it can’t see, hear, or do anything.
Let’s Walk Through a Real-World Example
Imagine your team spends a year and a few million dollars building a sophisticated AI underwriting agent for commercial property. The goal is for it to analyze a new application, pull third-party data on flood risk and building history, and deliver a bindable quote in under a minute.
The agent is finally ready. You feed it the first application.
And… nothing happens.
Why? The team digs in and finds that the address from the application form doesn't automatically sync with the third-party flood risk database. Someone has to manually copy and paste it. Then they discover the system that stores building permit history is 20 years old and doesn't have an API, so the AI can't access it. The final quoting system requires three separate logins that the AI can't navigate.
The "agent" itself is brilliant. But it's been dropped into an environment with no tools to help it do its job. The project is declared a failure, and everyone walks away saying, "I guess AI just isn't ready for prime time."
But that’s the wrong lesson! The AI was ready. We weren't.
How to Be in the Successful 5%
So, how do we stop the cycle? How do we get it right? It’s about a fundamental shift in mindset. We have to fall in love with the boring stuff.
Instead of starting with the question, "What amazing autonomous agent can we build?" start by asking, "What's one tedious, repetitive task we can build a simple tool for?"
Maybe you start by building a bot that just extracts customer information from an ACORD form and puts it into the right fields in your system. That's it. It’s a small win. But it’s a solid, reliable tool.
Then, you build another tool that automatically checks the address against a public records database. Then another that pulls the customer’s claim history.
You are building the Lego bricks, one by one. Each tool works, adds value on its own, and makes your processes a little bit better. Over time, you’ll have a whole toolbox of these reliable components.
And then, one day, you’ll be able to connect them all together. That’s when your "agent" is born. But it won't be a fragile, high-maintenance science project. It will be a robust, effective system built on a foundation of proven, battle-tested tools.
It’s not as exciting, I know. It requires patience. It requires telling your CEO that the revolutionary, world-changing AI is still a year or two away, but that you’re building the plumbing for it right now.
But the carriers who take this approach—the ones who focus on tools first and agents second—are the ones who will succeed. They’ll be the 5%. And they’ll be the ones who actually build the future of insurance, brick by boring brick.
