Explainable boosting machines are showing up in insurance rate filings with increasing regularity, giving carriers a way to pair the accuracy of gradient boosted models with the transparency that state regulators require-and the actuaries behind the shift say firms that do not upgrade are mispricing risk as a result.
Peggy Brinkman, a principal actuary at Milliman, traces the trajectory from univariate methods through generalized linear models to gradient boosted machines and now to explainable boosting machines. "New modeling techniques can extract more value from the same data in terms of risk understanding," Brinkman said. "Advances in methodologies are as important as bringing in new data sources."
The compute power fueling the change
From roughly 1×10^15 floating point operations used to train notable machine learning models in 2010, compute capacity has grown by a factor of ten billion by 2023, according to a UK government scientific report on advanced AI. Since 2012, the amount of compute used to train leading models has roughly doubled every six months-a pace that outstrips Moore's Law. The RAND Corporation projects fourfold annual growth in frontier training compute through 2030.
"Computing power and tools' ability to take advantage of it remain the primary constraints on how much data can be processed," Brinkman said. "The industry pushes those boundaries with new technologies." Cloud and distributed environments now run calculations in hours that once took weeks, letting carriers process more granular data and update pricing more frequently.
Why explainable boosting machines matter now
Generalized linear models have anchored actuarial pricing for decades because they are transparent and auditable, qualities that matter when models must be filed and approved. Their shortcoming is expressiveness: they struggle to capture complex, non-linear interactions between risk factors. Gradient boosted machines correct this by building pricing models from ensembles of decision trees, each correcting the errors of the last. That yields far more accurate risk segmentation-particularly for perils like wildfire, where vegetation, slope, wind, and structural traits interact in non-linear ways-but the models are opaque, and regulators have been slow to approve them.
Explainable boosting machines resolve that tension. "Carriers can now pursue the modeling accuracy that competitive differentiation requires without sacrificing the transparency that regulatory compliance demands," Brinkman said. EBMs are increasingly visible in rate filings, and their spread marks a structural shift in what is actuarially possible inside a regulated pricing environment.
Data layers still coming online
Satellite and aerial imagery have been the most significant data additions of the past five years, enabling automated property assessments at scale. Credit data remains among the most predictive variables in auto pricing, and telematics continues to improve as programs mature. Internet of Things sensors-monitoring electrical systems, water leaks, structural stress-represent the next wave of property risk data, but Brinkman is direct about their readiness. "These applications are in a low-sample-size phase, not yet available at the scale required to build statistically credible pricing factors," she said. Satellite data moved through the same transition over several years, and IoT is expected to follow a similar path.
Regulatory approval shifts
Technology investments yield returns only when improved models can be filed and approved. Judson Boomhower of UC San Diego identifies that approval burden as a real constraint. "Real costs are involved in implementing sophisticated pricing systems," he said, "both from a backend technology and regulatory perspective."
Recent California reforms-making it easier to incorporate frontier catastrophe model outputs in rate cases and to include reinsurance costs in filings-signal a shift. Carriers operating in California's admitted market should review the Department of Insurance's current guidance on catastrophe model use. Whether the changes are enough to restore broad admitted market participation remains untested. The modeling race is accelerating, and the regulatory framework is beginning to catch up.
Why this matters for insurance professionals
Carriers that invest in modern cloud infrastructure and adopt explainable boosting machines can file pricing models that segment risk more precisely and avoid the adverse selection that comes from underpricing volatile perils. Those still relying on generalized linear models risk losing profitable business to competitors who see risk more clearly. The regulatory environment is opening, but the advantage will go to teams that move first to build the technical and actuarial capability needed to deploy these methods in rate filings.
Professionals who want to stay current on how AI is reshaping underwriting and claims can find resources at AI for Insurance.
Your membership also unlocks:
