Over the past year, nearly every conversation we’ve had with carriers, managing general agents (MGAs), and insurtechs has eventually turned to the same topic: AI. Everyone’s building something — a scoring model, an underwriting tool, an agency-matching engine. Almost nobody is asking the question that actually determines whether any of it works: What data is this thing trained on?
Table of Contents
Think of agency data like bread. The day you get it, it’s fresh — soft, useful, exactly what you need. The longer it sits around, the harder it gets, until eventually it’s not good for anything but the trash. Most agency databases never get taken out of the bag. They just sit there getting stale while somebody keeps sending emails to it and wondering why nothing’s landing.
We’ve been in the agency data business for a long time, and the honest answer, more often than not, is “whatever we happened to have lying around.” That was already a problem when it just meant a wasted mail campaign. It’s a much bigger problem now that the same data is getting fed into models making decisions automatically, at scale.
The Numbers Are Worse Than Most People Think
Bad data isn’t a minor inefficiency. Gartner puts the average cost of poor data quality at $12.9 million a year, per organization, across industries. More than a quarter of organizations estimate they lose over $5 million annually because of poor data quality, according to research cited by IBM. A 2025 IBM Institute for Business Value study also found that 43% of chief operations officers identify data quality issues as their most significant data priority.
Contact data decays, too. HubSpot’s Database Decay Simulation, citing MarketingSherpa research, puts B2B data decay at roughly 2.1% per month, or about 22.5% annually. HubSpot points to contacts changing companies or email addresses and leads opting out as reasons marketing databases lose usable contacts over time.
None of this is exotic. It’s just how time works. But most distribution and marketing spend still runs on lists that are one, two, sometimes three years past their expiration date.
Insurance Distribution Has Its Own Version of This Problem
Agency rosters age in ways specific to this industry, and it’s worth naming them plainly.
- Mergers and acquisitions (M&A) and consolidation. Insurance distribution M&A ran at roughly 695 deals in 2025, according to OPTIS Partners, with private-capital-backed buyers behind the large majority of them. Somewhere between 25,000 and 30,000 agencies remain nationally, and a meaningful share will eventually get bought. An agency you mapped as “independent” two quarters ago may not be independent anymore.
- Producer movement. The relationship you built with a specific person may not exist the moment that person changes agencies — which happens more often than most CRMs ever catch.
- Book-of-business shifts. An agency’s appetite and line mix can change enough that a good fit from 18 months ago is a bad fit today.
Any one of these can invalidate a targeting list. Together, they make “we bought a database two years ago and haven’t touched it since” one of the more expensive habits in this industry.
AI Doesn’t Fix Bad Data. It Amplifies It.
Here’s the part that should actually worry people. A stale marketing list wastes a campaign — annoying, but bounded. You lose the budget, you move on. Feed that same stale, incomplete data into an AI model, and it doesn’t just waste a campaign. It teaches the model the wrong pattern. That pattern then gets applied automatically, across every recommendation, every score, every decision the model touches, with nobody in the loop to catch it.
IBM has said plainly that poor data quality is one of the most common reasons AI initiatives fail outright, regardless of how good the underlying model is. Garbage in, garbage out isn’t a new idea. What’s new is the speed and scale at which the garbage now gets acted on.
What Actually Predicts Fit
We track nine fields on every agency in our database, and we’d argue these are the fields that matter — whether you’re running a marketing campaign or training a model:
- Annual Written Premium Volume
- Percentage of Commercial Premium
- Revenue
- Employee Size
- Agency Management System
- Carriers
- Targeted Industries
- Special Affiliations
- Ownership Status
That last one deserves a second look, given everything above. Ownership status — independent, PE-owned, or publicly owned — is how you actually track the M&A problem instead of just complaining about it. It tells you whether the relationship you think you have is still the relationship that exists.
Fixing This Isn’t Complicated. It’s Just Ignored.
Keeping data current is like shoveling your driveway while it’s still snowing. You don’t do it once and call it handled — if you stop, it piles right back up, and by the time you notice, you’re not shoveling anymore; you’re excavating.
Verify before you build a campaign on a list, not after. Filter by real data — premium volume, appetite, ownership — instead of geography alone. Re-score your agencies at least annually, since a fifth of what you know is wrong within 12 months whether you like it or not.
If you’re building AI into your distribution or underwriting strategy, do the unglamorous work first: Audit the data feeding the model before you trust what comes out of it. It’s a lot cheaper to fix bad data once than to defend a bad automated decision made at scale, over and over, before anyone notices.
See Larry Neilson speak at ITC in Vegas on October 1, 2026.
About Neilson Marketing Services
Since 1988, Neilson Marketing has been helping carriers, MGAs, wholesalers, and insurtechs build smarter distribution strategy — backed by real agency data, not guesswork. Contact us today at (866) 816-1849 to put our data, expertise, and platform to work for you.
