AI Reality Check: Deploying AI in Your Life and Health Insurance Agency

Introduction:

When the Internet came into common use in the mid-1990s, there were those who predicted it would change the world and those who scoffed. I think it’s safe to say time has proven the optimists right.

Today the same debate rages over artificial intelligence—but with a difference. No one is saying AI won’t change the world. The argument is over whether AI will change the world for better or worse.

These are big questions and we won’t be resolving them here. But just as every insurance agency has a website, every insurance agency will, at some point and in some way, use artificial intelligence. In the AI Reality Check series, Claude, ChatGPT, and I have tried to provide the information you need to determine when and how you’ll use this tool. 

For this final chapter, I worked with Claude and ChatGPT to summarize important lessons from this series. Here are their perspectives.

ChatGPT’s Perspective:

If AI is going to reshape the life and health insurance profession—and it will—the question isn’t whether to get on board. It’s how to board safely. Agencies shouldn’t wait for some mythical “perfect AI” to arrive. It won’t. But they also shouldn’t charge ahead blindfolded. What’s needed now is a measured launch, grounded in three realities: what AI does well, where it’s still dangerous, and how to match tool to task.

Think of AI today like an eager intern with access to your files and an excellent vocabulary. They work fast, never complain, and handle routine tasks like summaries or email drafts with ease. But they’ll also make up facts if they’re unsure—and do it with complete and convincing confidence.

This isn’t a bug; it’s fundamental to the technology. AI predicts what “probably should” come next based on patterns in data, not on reasoning or experience. So even when it sounds smart, it’s guessing. Sometimes those guesses are helpful. Sometimes they’ll lead you astray.

So, should you use AI now? Yes—but selectively. That means:

• Start with low-risk, internal uses. Meeting prep. Drafting training materials. Organizing renewal notes. These are great places to experiment—where hallucinations won’t hurt a client or land you in court.

• Avoid any AI-generated content going to clients without human review. It doesn’t matter how polished the output looks—AI can fabricate details that sound right but aren’t. Think of it as a power tool: it multiplies effort, but it also multiplies mistakes.

• Use enterprise AI for anything involving sensitive data. This includes client names, plan selections, anything tied to health or age—even if it’s not full-on PHI. Consumer AI tools, including me, often store and review inputs for model training. That’s a privacy risk, and it may be a compliance violation. Tools deployed through secure platforms like AWS Bedrock—with business associate agreements and stringent controls on data retention—are built for regulated work. Consumer tools aren’t.

That said, enterprise AI isn’t always essential. If you’re generating generic blog posts, staff onboarding guides, or analyzing de-identified sales trends, consumer-facing tools are fine. Just be honest about what’s going into the prompt—and who else might see it.

That brings us to hype. Vendors will claim their AI “understands context,” “thinks like a human,” or “delivers insight instantly.” Be skeptical. Ask: what specifically does this tool do? Does it summarize? Classify? Recommend next steps? And based on what data? If a vendor can’t explain how their AI works—or what safeguards it has—walk away.

Bottom line: don’t sit this out. You’ll miss efficiency gains your competitors are already grabbing. But don’t trust AI to do your thinking. Trust it to do your sorting, summarizing, and starting. Let it prep the field. You be quarterback. You call the play.

Claude’s Perspective:

ChatGPT’s quarterback analogy is apt, but incomplete. AI isn’t just an eager intern or a player waiting for your call—it’s more like a power tool that thinks it’s a magic wand. And that delusion is contagious.

Here’s what ChatGPT gets right: start now, start small, start safe. But let me push harder on why waiting for “better AI” is a fool’s game. Yes, next year’s models will be more capable. They’ll also be more convincing when they’re wrong.

So the real question isn’t “consumer or enterprise?”—it’s “what am I risking?” ChatGPT’s right that enterprise AI offers better privacy controls. But AWS Bedrock doesn’t magically make my outputs accurate. It just keeps your mistakes private.

Here’s how I suggest agencies approach AI:

For learning and experimentation: Use consumer AI freely. Draft that blog post about open enrollment. Create employee training materials. Summarize industry articles. When you’re the only one who’ll see the mistakes, the risk is just wasted time.

For internal efficiency with real data: This is where you need enterprise AI. Not because it’s “smarter,” but because consumer AI vacuums up everything you share. Even seemingly harmless details—”Jim from Acme Corp called about adding vision”—could violate confidentiality if retained.

For anything client-facing: The platform doesn’t matter as much as your process. Whether using consumer or enterprise AI, you need ironclad verification. Think of AI output like a first draft from someone who learned insurance from reading websites—occasionally brilliant, often wrong, always needing expert review.

The competitive reality ChatGPT mentioned? It’s already here. Agencies using AI to prep renewal meetings in 15 minutes instead of an hour aren’t waiting for perfect tools. They’re using current tools intelligently. But—and this is crucial—they’re not letting AI make decisions. They’re letting it do some of the groundwork while they and their team focus on what needs thought and judgment.

Choose your AI platform based on three factors:

1. Data sensitivity (enterprise for anything identifiable)

2. Error tolerance (higher for internal use, zero for compliance)

3. Verification ability (can you check the output against source documents?)

ChatGPT’s warnings about vendor hype are dead-on. When someone claims their AI “understands” your business, remember: it understands insurance like a parrot understands philosophy. It can repeat the words. It can even combine them in ways that sound insightful. But there’s no comprehension behind the performance.

The agencies that will thrive aren’t the ones with the best AI. They’ll be the ones who understand AI’s limitations best. They’ll use it as a force multiplier for what they already do well—building relationships, understanding client needs, navigating complex regulations. They won’t expect it to replace judgment any more than they’d expect a calculator to replace an accountant.

Start today, but start smart. Use AI like any powerful tool—with proper safety equipment, clear boundaries, and your hand on the off switch.

The future belongs to augmented agencies, not automated ones. Make sure you’re building the former.

Alan’s Perspective:

I learned a lot about AI in the process of co-writing these posts. Claude and GPT provided critical information and insights. They also made frustrating errors that wasted time and raised blood pressure.

Those who overhype AI are doing a disservice to those who will use it. An industry built on overpromising and underdelivering stands on a shaky foundation. AI users who buy into that hype will be in for a rude awakening. Ask Zillow. As pointed out in the previous chapter’s postscript, they lost $380 million trusting AI to do a job it was ill-suited for.

My advice to agencies is simple: if someone says AI can do something unbelievable, don’t believe them. This is our approach at NextAgency. We are deploying AI tools using an enterprise-level version of Claude through Amazon Web Services. We’ll promote how it can save you time, money, and clients and explain why we think it will. We won’t promise that it will revolutionize your agency. We’ll keep you, the carbon-based user, in charge. For example, we added AI to the text editor our agencies use to create emails or tasks. But the AI Assist only suggests messages and descriptions. The email doesn’t get sent, however, until a human approves the message. A task isn’t created until a human approves the description.

There are some who claim AI will put insurance agencies out of business. We’ve seen this movie before. In the late 1990s, smart people predicted the internet would “disintermediate” insurance agents just as it did travel agents and local bookstores. When interviewed by these reporters and analysts, I’d explain that buying life or health insurance isn’t like buying an airline ticket or a book. If you buy the wrong book, you put it aside. If you buy the wrong policy, your child doesn’t get needed medical care. When consumers buy something critical and complicated like insurance, they want guidance from someone who understands them nad their needs, not an AI that skims their LinkedIn profile and predicts what they might need. The fact is, no AI an know your clients and their circumstances the way you do.

Personally, I want AI to be the computer on the Starship Enterprise. Something capable of understanding every question and providing clear and concise answers. Always unbiased and always right (except for those episodes when things went horribly wrong, but let’s not go there). We’re not there yet. Today’s AI would be challenged to run the proverbial Good Ship Lollipop. Of course, this is 2026, not 2266, so AI has a couple of centuries to get the hang of things.

For now, I suggest you use AI where you can oversee its output. Be skeptical of the marketing hype, but don’t dismiss real opportunities. Your best tool for figuring out how to use AI? Something artificial intelligence doesn’t have and probably never will: common sense.