Mark Romano gripped the steering wheel and tried to keep his car from swerving into another commuter on the busy Illinois tollway. “God, please don’t let me hurt someone,” he prayed.
Dizzy again. These bouts of vertigo were barely noticeable at first, but something else was going on now. At night, he would lie in his bed, stare at the ceiling and watch everything twirl. In the morning, the spells came in waves during his commute to Allstate’s national headquarters in suburban Chicago.
Stress?
It was December 2007, and Romano was a senior manager at Allstate and its top expert in Colossus, a program that calculates how much a person might be paid for an injury claim. He was in charge of two projects to “tune” and “recalibrate” Colossus, work he knew could affect payments to thousands of people.
Colossus was part of a quiet revolution in the insurance industry.
Before the early 1990s, insurance was a decidedly human endeavor, especially when it came to setting rates and paying claims. To set premiums, insurers relied on computations from their actuaries — mathematical wizards armed with statistics and tables that assess various risks. When it came to paying claims, insurers often sent adjusters into the field, where they met face-to-face with people injured in car wrecks.
Today, insurers have an array of computer programs that guide the flow of trillions of dollars to and from customers around the world. These programs include sophisticated “catastrophe models” that use weather data and other factors to predict an insurance company’s losses in a disaster. “Scoring models” use credit histories and secret algorithms to estimate which customers are more likely to file claims. Colossus and similar programs help companies manage claims. Like a TurboTax program for medical injuries, adjusters plug in information about a person’s loss — from a damaged spine to a fractured finger. Colossus then cranks out a range of payments to cover the costs. Insurance industry critics and even many insiders call these programs “black boxes” because their formulas, data sets and operational policies are cloaked in secrecy.
Few people at Allstate knew more about Colossus than Romano. On organizational charts, he was Allstate’s Colossus “subject matter expert.” And in late 2007, at age 49, he was at the peak of his career, working in one of the nation’s largest insurance companies — and wondering whether he should leave it all behind.
Romano has thick black hair and wears thin glasses. His brown eyes widen when he wants to make a point. He had gone into the business to help people, but he knew that his work on Colossus would do the opposite.
During his hourlong commute to Allstate’s sprawling campus in Northbrook, his mind drifted to his daughter at the College of Charleston, to his son in private school, his wife’s multiple sclerosis, medical bills, his mortgage, the decades he put into his career. The dizzy spells grew worse. Doctors prescribed motion sickness medicine, relaxants and physical therapy. Then the headaches came — migraines as long and powerful as a Midwestern freight train, box cars of pain, one after another after another. Something had to give.
Birth of Colossus
Among computer programmers, the name Colossus has a rich history. In World War II, British code-breakers called their hulking new programmable machine Colossus and used it to decipher German teleprinter messages. In 1970, filmmakers released “Colossus: The Forbin Project,” a science fiction movie about an army supercomputer that tries to take over the world. (At one point, the computer tells its human creator, “You will come to regard me not only with respect and awe but with love.”)
The insurance industry’s version of Colossus was born in Australia. In the 1980s, a government-chartered insurer ran into financial trouble because of claim costs, which were growing at an annual rate of 14 percent. The insurer set its sights on its adjusters.
It’s the adjuster’s job to evaluate people’s losses and come up with ways to settle their claims. This often meant assessing what people did in their careers and how an injury might affect their future income and overall enjoyment of life. Longtime adjusters talk about the challenge of sizing up people when they’re suffering, and the knowledge adjusters need, from medicine to car repairs, to calculate a fair settlement.
The inherent complexity in putting numbers on injuries also meant that adjusters often came up with different amounts for similar types of claims. In Australia, payments varied by more than 80 percent. So to reduce these disparities and lower overall costs, the Australian insurer worked with a software company on a novel idea: embed the experience and knowledge of their best adjusters in a computer program.
The programmers studied how top adjusters made decisions and then created software to mimic their work. This program became known as Colossus and required answers to as many as 700 questions, ranging from the severity of injuries to how people experienced the loss of enjoyment in life. Injuries were broken down into 600 different codes. The program analyzed legal settlements and jury verdicts, combined this information with data entered by the adjusters, and generated what were supposed to be fair settlements.
A few people questioned whether computer programs were up to this complex task. An Australian law professor wrote that the development of Colossus was “just one instance of an important challenge of the information age: how to ensure that computer-based decision making is fair and non-discriminatory.” But Colossus was a huge success. Within a few years, payments for similar claims were more consistent and the costs of those claims had stopped rising.