Insurance has spent decades getting better at collecting information. The harder problem is knowing when that information stopped being true.
A homeowner replaces a roof. A teenager gets a license. A business buys another hundred and fifty thousand dollars of equipment. Revenue goes from two million to four. Somebody finishes a basement, starts renting the lake house on weekends, or puts up a detached garage.
None of that starts an insurance conversation on its own. That is the problem.
The record sitting in the system may be exactly what the client said two years ago. Entered correctly. Confirmed at the time. Nobody made a mistake.
It just is not true anymore.
Insurance information has a shelf life
The industry sorts information into accurate and inaccurate. That is one category short. There is a third that matters just as much: information that was accurate when it was collected and has since aged out of reality.
A roof recorded as eighteen years old three years ago was correct. It is also now a twenty-one year old roof. A business accurately insured at $1.5 million in sales did not suffer a data entry error when it grew to $3 million. The business changed. The record did not.
Montana wrote this idea into law. An insurer using credit information cannot take an adverse action against a consumer unless the report behind it was pulled within ninety days, and cannot use credit information at all unless it refreshes at least every thirty-six months. Nothing in that statute concerns accuracy. A four-year-old credit report can be a perfectly accurate description of a person four years ago. A legislature looked at that and decided being accurate about the past is not the same as being right.
Insurance is full of information like this, because insurance describes things that move. Homes, businesses, families, vehicles, values, replacement costs, drivers, ownership. All of it changes. The record usually does not.
New business asks questions. Renewal asks whether anything changed.
That difference sounds small. It is not.
During new business everyone expects discovery. Questions get asked, applications get completed, property gets reviewed, drivers get identified, limits get discussed. At renewal, most of that gets carried forward, which is the only sensible way to run the work. Rebuilding an entire profile every twelve months would be expensive and irritating and would mostly confirm what was already there.
But carrying information forward smuggles in an assumption. If nobody said it changed, it must still be right.
The claims data suggests otherwise. In Swiss Re Corporate Solutions figures published for the Big “I” errors and omissions program, inaccurate information provided to the carrier has been a consistently top three cause of agency claims. One detail in the 2020 numbers is worth sitting with: renewal errors rose while new business errors fell.
New business is where information gets gathered. Renewal is where it gets assumed.
Clients are not hiding anything when they fail to mention a change. Most people do not spend their day sorting life events by insurance relevance.
They know they renovated the kitchen. They do not know it moved the replacement cost. They know their daughter moved home and drives one of the cars. They do not know anyone needs to be told. They know the company bought more equipment, and they assume the policy somehow already covers it.
That is not negligence on either side. It is a communication problem created by time.
More data does not fix it by itself
The obvious answer is to collect more. But more is not better if nobody knows which answer to trust.
Three systems can hold three different roof ages. The client said one thing. The public record says another. The inspection says a third. The management system has a fourth, entered by somebody who is no longer there. Which one wins?
A system that can only store another answer has not helped. It needs to know where an answer came from, when it was last confirmed, and whether something credible now disagrees with it. That is a different problem from collecting data. It is keeping data current.
Regulators arrived at this before most vendors did. The NAIC model bulletin on artificial intelligence systems, adopted in December 2023, asks insurers to document data currency and lineage alongside quality and integrity. New York’s circular letter asks for a description of the data lifecycle. Currency is how old it is. Lineage is where it came from. Neither has anything to do with the algorithm.
The question worth asking
Not: is the client’s annual revenue on file. But: when was that revenue last confirmed.
Not: is there a roof age on the property. But: where did that roof age come from, and has anything happened since.
Asked that way, a policy stops looking like a set of fields and starts looking like a set of facts with different rates of decay. A VIN almost never changes. Annual revenue can double in a year. Drivers come and go. Replacement costs move every quarter. Treating all of it as though it ages at the same speed is what puts a twenty-one year old roof on a policy that says eighteen.
Where this goes next
The industry has spent years moving information faster. Connect the systems. Eliminate duplicate entry. Prefill the application. Download the policy. Automate the renewal. Every one of those was worth doing.
But moving old information faster does not make it current.
The harder job is working out whether what is already in the system still describes the real world. Noticing change. Comparing sources. Remembering when something was last verified. Knowing which fields go stale quickly and which do not. And putting the ones that matter in front of a person while there is still time to do something about it.
Because the dangerous record is not always the one entered incorrectly. Sometimes it is the one entered perfectly.
Three years ago.
Disclosure: GTA Technology builds and sells insurance technology, which means it has a direct interest in how this question gets answered. The sources below are listed so the argument can be checked independently of who is making it.
Sources
Ninety day and thirty-six month recency requirements for credit information used in underwriting and rating. Montana Code Annotated 33-18-605(f) and (g). mca.legmt.gov
Inaccurate information provided to the carrier as a consistently top three cause of agency errors and omissions claims, and the 2020 finding that renewal errors increased while new business errors declined. Swiss Re Corporate Solutions claims data published in IA Magazine, 1 March 2021, and 8 January 2024. iamagazine.com
Data currency and lineage as required elements of a written artificial intelligence systems program. NAIC Model Bulletin on the Use of Artificial Intelligence Systems by Insurers, adopted 4 December 2023, Section 3. content.naic.org
Description of data lifecycle management process. New York Department of Financial Services Insurance Circular Letter No. 7, 11 July 2024. dfs.ny.gov
