InsurTech pricing specialist Earnix has unveiled a new automated capability within its Model to Rating Structure Distillation lab.
InsurTech pricing specialist Earnix has unveiled a new automated capability within its Model to Rating Structure Distillation lab. The specialized feature bridges a long-standing structural gap in the insurance industry; the friction between highly accurate non-linear machine learning models and the rigid, transparent rating tables demanded by insurance regulators and internal risk governance teams.
By automatically translating complex algorithmic outputs into production-ready rating structures for its core Price-It platform, Earnix is enabling insurance actuaries to operationalize high-performance AI without compromising regulatory compliance.
Historically, the most predictive model an insurance firm built was often the one it could never legally put into production. This disconnect stems from contrasting structural priorities within underwriting:
• The Black Box Problem: Advanced machine learning frameworks (such as XGBoost or neural networks) are exceptionally skilled at uncovering complex, non-linear risk factors in consumer data. However, regulators require pricing logic to be fully transparent, explainable and easily auditable.
Traditionally actuaries had to manually reverse-engineer advanced algorithmic outputs into simple, additive linear rating tables. This manual distillation process was slow, highly subjective, and usually forced teams to sacrifice significant predictive accuracy for the sake of simplicity.
Earnix’s new framework addresses this by generating multiple candidate rating structures that closely mirror the behavioral patterns of the original black-box model. Rather than relying on unchecked automation, the process is heavily guided by human-defined business and legal parameters.
Pricing teams can set strict, unyielding guardrails before running the distillation process including monotonicity constraints, pre-determined offsets, risk weights, and specific interaction limits.
See also: GIM Series A Funding Round $20m Raised for Agentic AI Investing
The lab generates a spectrum of candidate models. Simple options utilize inherently transparent, interpretable additive frameworks like Auto-GLM (Smart Grouping) and Explainable Boosting Machines (EBMs). For cases requiring higher predictive nuance, more expressive options leverage CatBoost as a residual learner on top of a base model, using a post-processing LASSO step to strip away low-value data splits.
By replacing subjective manual conversions with objective mathematically backed alternatives, insurers can compare various candidate frameworks on a singular dashboard. This lets actuary desks select the optimal balance between high predictive performance and strict regulatory filing readiness depending on regional compliance demands.
