Technical

Predictive model

A model that predicts outcomes for insurance decisions.

How the laws define this

Both sources establish that a predictive model relies on data to generate forecasts, risk assessments, or predictions intended to guide insurance-related decisions and practices. They diverge primarily in how specifically they describe the model's technical design, data sources, and oversight rules. Colorado Senate Bill 21-169 (Colo. Rev. Stat. § 10-3-1104.9) treats a predictive model broadly as using data for scores or predictions, explicitly distinguishing it from general algorithms for governance, testing, and oversight purposes. In contrast, the definition under concerning protecting consumers from unfair discrimination in insurance practices specifically restricts the concept to a statistical or machine learning tool utilizing external consumer data to forecast outcomes such as claims likelihood, subjecting these tools to direct scrutiny for potential discriminatory impacts.

Synthesised from the 2 statutory definitions below. Each one is quoted in full, with its source.

What it triggers

In Colorado, United States under Colorado SB 21-169, predictive models are prohibited if they cause unfair discrimination.source
In Colorado, United States under Colorado SB 21-169, insurers must apply governance, testing, and oversight including risk frameworks and bias tests to predictive models.source
Under CONCERNING PROTECTING CONSUMERS FROM UNFAIR DISCRIMINATION IN INSURANCE PRACTICES, predictive models are subject to scrutiny for potential discriminatory impacts.

Definitions (2)

A model that uses data to generate predictions, scores, or risk assessments intended to inform insurance-related decisions and actions. The statute treats predictive models as distinct from general algorithms for purposes of governance, testing, and oversight of insurer use.

Restrict Insurers' Use of External Consumer Data (Protecting Consumers from Unfair Discrimination in Insurance Practices) — Colorado Senate Bill 21-169 (Colo. Rev. Stat. § 10-3-1104.9). ([leg.colorado.gov](https://leg.colorado.gov/bills/sb21-169?utm_source=openai))Definition 1 of 2

A predictive model is a statistical or machine learning tool that uses external consumer data and information sources to forecast future events or outcomes relevant to insurance practices, such as risk assessment or claims likelihood. These models are subject to scrutiny under the Act for potential discriminatory impacts.

CONCERNING PROTECTING CONSUMERS FROM UNFAIR DISCRIMINATION IN INSURANCE PRACTICES.Definition 2 of 2

Not to be confused with

Algorithms and Predictive Models

Algorithms and predictive models refer to broader systems utilizing external consumer data for insurance practices, while a predictive model specifically generates predictions, scores, or risk assessments.

Model

A model is a trained algorithmic artifact that performs tasks using learned parameters, whereas a predictive model is specialized for generating predictions to inform insurance decisions.

AI model

An AI model is the mathematical engine within an AI system, while a predictive model is a statistical or machine learning tool focused on forecasting insurance-related outcomes.

Questions and answers

What obligations apply to predictive models in Colorado insurance?

Colorado SB 21-169 prohibits insurers from using predictive models that cause unfair discrimination. It requires governance, testing, and oversight including risk frameworks and bias tests. The Colorado Division of Insurance enforces these rules.

Are predictive models scrutinized for discrimination in insurance?

Yes, under CONCERNING PROTECTING CONSUMERS FROM UNFAIR DISCRIMINATION IN INSURANCE PRACTICES, predictive models using external consumer data are subject to scrutiny for potential discriminatory impacts.

How does a predictive model differ from general algorithms in regulation?

Colorado SB 21-169 treats predictive models as distinct from general algorithms, subjecting them to specific governance, testing, and oversight requirements for insurer use.