I build systems that assess and score people. That is most of what I have done for fifteen years, across data and AI. So I want to tell you about the time I was on the other side of one, as the person a system could not process.
It taught me something that should matter a great deal to anyone running an insurance operation.
I started a new job at a large company. Part of the package was supplemental coverage, broader than the state minimum, the kind of benefit that is genuinely good to have. But when it was time to enroll online, none of the options described my situation. HR arranged for a representative to help me, but his paper forms had no option for my situation either. He was completely boxed in by the same categories the software used.
Weeks later, the outcome was this: they would cover everything, except anything related to my hand. The one part of me that fell outside their categories was entirely carved out of the coverage.
I had not been denied insurance. I had been quietly issued a policy with a hole in it, shaped exactly like the thing their system could not describe.
Here is what strikes me most, looking back. Nobody in this story did anything wrong. HR tried to help. The representative sat with me for hours. The system was not broken, it did not crash or error out. It worked exactly as designed. The problem was upstream of all of them, in a single assumption baked into the form, the paper, and the process behind both: that every applicant would fit one of the categories that already existed. I did not, and the entire apparatus had no way to hold that fact. It could only round me to the nearest available profile and proceed.
This is the structural blind spot in a lot of what our industry builds, and it is worth naming plainly because it is invisible from the inside. Your intake forms, your rating categories, your automated underwriting, all of it is built around a picture of a typical applicant. It serves the people who match that picture very well. The further someone sits from it, the worse it serves them, and past a certain distance, it cannot serve them at all.
Bringing a human into the loop feels like the fix, but it usually isn't, because the human is working from the same categories. The form just becomes a person reading the form to you.
So I would leave insurance leaders with three questions.
First, look hard at what your categories cannot describe, not just what they can. Every dropdown and every rating field is a decision about which people are allowed to exist in your system. The applicants who don't fit are not edge cases to be cleaned up later. They are people your process is quietly turning away, and often they are turned away from precisely the coverage they needed most.
Second, be honest about what a human in the loop actually fixes. If your representatives are working from the same fixed categories as your software, adding them does not add flexibility. It adds a friendlier face to the same dead end. Real flexibility means giving someone the authority to record a situation the system did not anticipate, rather than forcing it into a box that changes the outcome.
Third, and this is the one I would push hardest: do you have any way to see the people who tried to enter and couldn't?
Think about my case. I wanted the coverage. I liked the company. I was ready to be a paying customer. And I nearly fell out entirely, not because you assessed me as too risky, which would at least show up somewhere in your data, but because I could not get through the door to be assessed at all.
A declined applicant is a record. Someone who abandons an intake they physically cannot complete is nothing. No row, no flag, no metric. If you have no measure of who failed to get in, then you are losing customers you will never know you lost, and no amount of studying the ones who made it will ever reveal them.
I got lucky enough to keep pushing until something resolved, imperfect as it was. Most people don't. They just quietly go away, and your system records the silence as success.
