How Insurance Algorithm Was Analyzed

How Insurance Algorithm Was Analyzed

State regulators and consumer advocacy groups have scrutinized Allstate Corporation’s use of big data and personalized pricing in the way it calculates how much the company charges.

State regulators and consumer advocacy groups have scrutinized Allstate Corporation’s use of big data and personalized pricing in the way it calculates how much the company charges its private auto insurance customers.

We tested whether Allstate’s personalized pricing algorithm treated customers differently based on non-risk factors by analyzing rare customer-level data we found in documents that were part of a 2013 rate filing submitted for approval and subsequently disapproved by Maryland regulators. This filing provides the most insight into Allstate’s retention model available to the public, with a level of detail that is typically shielded from public view by Allstate and other insurers.

Our analysis revealed that one of the most significant factors correlated with policyholders’ ultimate proposed price shift was how much they were already paying.

In Allstate’s filing, the company indicated that more than half of its customers were paying too much and others too little for car insurance based on current risk factors. The rate the company calculated the policyholders should be paying is called the “indicated premium” in its filing. (In this paper, we’ll call it the “ideal price.”)

But Allstate didn’t propose adjusting rates to those ideal prices. The company submitted a rating plan based on proprietary customer “retention model” algorithms to more slowly adjust their customers’ rates. Allstate called the rates they wanted to charge the “selected premium.” (In this paper, we’ll call it the “transition price.”)

Our analysis found Allstate saved its highest percentage and dollar rate increases for policyholders who were already paying high rates. Drivers whose premiums were more than $1,883.97 during the preceding six-month period and were due an increase faced transition rate increases of up to 20 percent.

Other customers to whom Allstate had assigned the same ideal price but were paying lower premiums at the time would not have gotten anywhere near as high a bump. Instead, Allstate capped their transition price increases at 5.02 percent.

In other words, it appears that Allstate’s algorithm built a “suckers list” that would simply charge the big spenders even higher rates.

We found that customers who would have received massive rate hikes under Allstate’s plan were disproportionately middle-aged. Customers between the ages of 41 and 62 were the most likely to receive a massive rate hike, likely related to the fact that this age group had the highest median current prices.

Those with massive rate hikes were also disproportionately male. They were also disproportionately living in communities that were more than 75 percent “nonwhite.”

In addition, while Allstate’s own data stated that more than half of its customers in Maryland were being overcharged, the company’s algorithm determined that none of them should be given a substantial discount. Customers aged 63 and older were disproportionately affected by the lack of meaningful discounts.

There are limitations to our analysis. The proposal was never put into use in Maryland. And we can’t state with certainty that Allstate customers in other states would be affected in exactly the same way because company officials say they create variants of its models for each state [1]. However, we did find filings in 10 states where Allstate said it uses retention models in its auto insurance pricing.


Car insurance is mandatory for drivers in every state but New Hampshire and Virginia [2] and is regulated at the state level. Car insurers in the United States are supposed to set customers’ rates based primarily on drivers’ risk of getting into an accident or suffering other losses that will cost the insurance company money. Most states forbid insurers from charging customers rates that aren’t tied to risk—a common phrase used is “unfairly discriminatory” [3] — meaning two customers with similar risk profiles shouldn’t be charged different prices.

Insurers regularly submit “rating plans” to regulators. According to the National Association of Insurance Commissioners, these plans include “a set of rules, risk classifications and sub-classifications, factors, discounts, surcharges, and fees applied to a base rate”. [4] These plans are used to calculate premiums.

In recent years, some insurers have introduced predictive analytics using increasing amounts of customer data in their rating plans. Unlike the techniques of the past, these methods are not straightforward. It can be unclear, to regulators and consumer advocates alike, how customer characteristics affect the pricing decisions made by insurers.

One controversial data-driven practice is called “price optimization,” which involves charging customers personalized prices that are based on factors other than risk. Among them is “retention” or how likely a customer would be to switch companies based on a price hike or to stay without a price drop.

Concerns about price optimization arose in 2013, when the software developer Earnix published a market survey of 73 executives and pricing professionals representing large insurers in the United States and Canada, showing that 45 percent were using price optimization and another 29 percent planned to join them in the near future. [5] [6]

In a 2014 letter to state insurance regulators, the Consumer Federation of America (CFA) accused Allstate of incorporating price optimization in some of its recently introduced rating plans through a factor called “Complementary Group Rating”. [7]

The CFA analyzed a Wisconsin Allstate filing and found that the company had divided policyholders into nearly 100,000 “micro-segments” based on the zip code, years of prior insurance, birthdate and gender of the oldest person on the policy. The consumer group accused Allstate of using “illegal techniques” to shift rates.

“The filing shows what we have alleged throughout the ongoing debate over Price Optimization: that insurers are unable to resist the temptation to use Price Optimization to break up risk classes and base prices on non-risk related factors such as price elasticity,” the consumer group said in the letter to state regulators. They urged the regulators not to accept any Allstate rate changes that used the factor and to force insurers to disclose whether they used price optimization techniques.

In 2015, the National Association of Insurance Commissioners (NAIC), a professional advisory group for state insurance regulators, published a white paper on price optimization. It found that while there was no agreed-upon definition of price optimization, many of the practices labeled as such could lead to the use of non-risk based factors, including charging drivers based on how much they are willing to pay, whether they shop around for other insurance options, and if they ask questions or file complaints.

The report stated that these practices could cause customers with similar risk profiles to be charged different rates for the same coverage. The report raised questions about whether price optimization would harm people of color and low-income drivers the most but did not provide an answer.

In the paper, the NAIC suggested that insurance commissioners publish guidance reiterating that rates should not be unfairly discriminatory, and specifying that discrimination includes “[r]etention adjustment at an individual level”[8] In the past five years, at least 18 states and Washington, D.C., have issued public statements prohibiting “price optimization.”

Nevertheless, Allstate was undeterred in its push toward setting rates based on retention. Eight months after the NAIC white paper was adopted, Allstate claimed that it was using retention models in 23 states[9] It typically inserts these retention models as part of its Complementary Group Rating (CGR)—or a successor factor that it calls Table Assignment Number (TAN) Group Rating. In the past decade, Allstate has proposed rate plans that use CGR or TAN in at least 39 states. 

At least three states have rejected these proposals. Georgia explicitly disapproved a plan submitted by Allstate that used TAN, stating: “The Department does not allow the use of price optimization.” Maryland regulators said the use of CGR “results in rates that are unfairly discriminatory.” Florida regulators disapproved an Allstate filing in 2014, writing in their rejection letter that Allstate’s plan to set an individual’s premium based on his or her “modeled reaction to rate changes” was “unfairly discriminatory.” Allstate withdrew proposals in some other states, including Louisiana and Rhode Island after regulators asked pointed questions.

Our review of public records shows that in at least 10 states, Allstate’s current auto insurance pricing schemes include TAN or CGR with a retention model.

insurance cars the-markup allstate's-algorithm data analysis

Bootstrap 5 Complete Course with Examples

Bootstrap 5 Tutorial - Bootstrap 5 Crash Course for Beginners

Nest.JS Tutorial for Beginners

Hello Vue 3: A First Look at Vue 3 and the Composition API

Building a simple Applications with Vue 3

Deno Crash Course: Explore Deno and Create a full REST API with Deno

How to Build a Real-time Chat App with Deno and WebSockets

Convert HTML to Markdown Online

HTML entity encoder decoder Online

How To Blend Data in Google Data Studio For Better Data Analysis

Google Data Studio helps us understand the meaning behind data, enabling us to build beautiful visualizations and dashboards that transform data into stories.

Exploratory Data Analysis is a significant part of Data Science

Data science is omnipresent to advanced statistical and machine learning methods. For whatever length of time that there is data to analyse, the need to investigate is obvious.

Tableau Data Analysis Tips and Tricks

Tableau Data Analysis Tips and Tricks. Master the one of the most powerful data analytics tool with some handy shortcut and tricks.

Analysis, Price Modeling and Prediction: AirBnB Data for Seattle.

Analysis, Price Modeling and Prediction: AirBnB Data for Seattle. A detailed overview of AirBnB’s Seattle data analysis using Data Engineering & Machine Learning techniques.

What Is Data Analysis?

DISCLAIMER: absolutely subjective point of view, for the official definition check out vocabularies or Wikipedia. And come on, you wouldn’t read an entire article just to get the definition.