Personal auto underwriting with computer models

Hi. How much personal auto underwriting is done with computer models nowadays?

If I apply for a personal auto policy with your company, is my data input into a computer model that produces a score, and if that score is within your range and if I have no unusual flags, would a positive decision be made with minimal individual underwriter attention?

Also, how many personal auto policies might an individual underwriter review in a day?


Not sure how much of a response you’re going to get given that such processes are pretty proprietary information.

However, given that operating lean seems to be a significant thing for most P&C companies, I’d be shocked if a company actually isn’t using some form of “AI” or computer models to facilitate their underwriting process as a means to trim down their underwriting staff and to keep “automatic” stuff (like your scores are so favorable that there’s really no reason to not write you (barring a hit on a suspended license from the DMV)) from showing up in the human’s in-box.

I think it’s safe to say that most companies are also going to get 3rd party reports (like credit scores, reports from an insurance clearing house, etc.) and use them along with whatever they ask you about explicitly. I’ve seen several smaller insurance companies that will ask you a few basic questions during the quoting process to give you a price; I have to assume that they’re getting more info from these 3rd party reports (if they want to stay profitable).

i don’t work in PA. I assume there is barely any human participation in a personal auto application review. you are what you drive, where you drive, your recent accidents and tickets, etc.


Here is one of the conundrums . . . is “where you drive” anything close to “where you garage the vehicle”?

where you garage is a proxy unless you have the fob/spy and they have that data. imperfect as that is.

It’s been over 20 years since I last worked with personal lines. My current employer doesn’t do personal lines in the US, because we realized many years ago that we couldn’t compete with larger carriers whose analytics, automation, and economies of scale meant they could potentially operate at a much lower expense ratio (and they were willing to accept leaner profit margins), and we just couldn’t realistically compete.

Back at a prior employer, where I last worked with personal lines, we were in the process of moving towards mostly-automated underwriting. A scoring algorithm was used to either tier a risk (at that time, at least, there was a polite fiction that tiering was “underwriting” rather than “rating”), or to refer an application/renewal to a human for manual review (and probable declination/nonrenewal).

In addition, review of traditional basic underwriting questions had been automated – answering the “wrong” way could trigger an automated declination or a referral to a human for review, without use of a scoring algorithm.

At that time, that company was also just starting to implement the use of models to predict the expected value of renewal reports, to be used in deciding whether to order them (if expected value of report < cost of report, don’t order). I have to believe that those models that I built for that purpose have come a long way since that time.

I strongly suspect that today, unless you’re looking at assigned risk business, or niche insurers, most personal auto underwriting is almost fully automated, relying on scoring, or at least decision trees built in a manner similar to a scoring algorithm.


Some of us are still fighting the good fight over here! :slight_smile:

More power to you. I much prefer working on products where it’s possible to target combined ratios in the low 90’s or better. :slight_smile:

1 Like

Exactly 95. EVERY year. Bother me not with your nonsense about hurricanes and randomness.

Without randomness, bonus day would be a lot less exciting.


Thanks for everyone’s comments. Any thoughts on my second question?

How many personal auto policies might an individual underwriter review in a day?

I think the second question is difficult because the definition of “underwriter” for personal lines has evolved, and may vary by carrier.

My first job out of college, over 30 years ago, was at a tiny regional company. While I was naĂŻve in getting a job there (in my defense, the EL job market sucked that year), the training I was put through there was an excellent introduction to the business. As part of that training, I spent a week shadowing/working as a line underwriter for personal auto.

This was pre-automation. We touched every application coming in – mailed or faxed. If everything was in order, an application could be reviewed in 90-120 seconds, before being handed off to whomever had the duty that day to go sit at one of the terminals to ender data into the minicomputer. (I did that for a morning too. It sucked.) If there were problems, depending on the problem the application would either be put in the outbox to be transferred to a more senior underwriter. or we’d need to call the producer to try to resolve.

Today, for standard carriers…line underwriters aren’t touching every application. There are people in that role who are touching business that doesn’t fit the boxes for automated processing. They do handle a large number of such records. The precise details will depend on how good the automated processing is (and thus how long it takes to resolve the exceptions), and on whether they have other duties.

I’ll inject here again the caveat that I do not currently work with personal auto in the US. What I’m writing is based on older experience (I did personal auto in a past life; my employer used to do personal lines in the US and we do personal lines outside the US), or what I’ve heard as part of my until-recent job (risk management) where we gleaned information about what other companies do from the folks we hire.

That caveat aside, my wishy-washiness in answering is because the definition of “underwriter” in personal lines has evolved / is evolving, and isn’t quite as uniform as it once was. The folks who once would have been described as “line underwriters” aren’t always called “underwriters” anymore. They might be described as technical/analyst staff in an underwriting/product department. They may have limited underwriting authority built into their job descriptions, instead of explicit letters of underwriting authority,

When we talk about “underwriters” for personal lines today, I think more of people who operate where “producer management” and “product management” intersect. They don’t tocuh individual policies, unless there’s a semi-unique exception involved, or unless they’re looking for examples to support their producer/product management duties.

I suppose I should also add the caveat that these comments are geared towards mainstream personal auto. There are niches/specialties that will be exceptions. For example, I think Hagerty has “underwriters” who are touching individual risks, because the nature of their book is one where only a portion of their market is really suited for automated underwriting.

And to reiterate an earlier point: Underwriting rules often fall into the “proprietary knowledge” category.

So those who might be close to this sort of information (as I am) may not be in a position to say much.

Or more accurately, shouldn’t say much.

1 Like