ABOUT SOTERIS
Soteris is a YC-backed AI company building the future of pricing and product management for the
insurance industry. Our mission is to infuse the $5 trillion P&C insurance industry with best-in-class,
proprietary, AI-driven data analytics. We’ve spent years building our own proprietary AI models on personal auto claims and exposure data to help insurers improve their loss ratios, with over 100 million submissions and $180 billion in premium scored to date.
Each year, roughly $750 billion in insurance policies are written in the United States. Our machine learning platform helps insurers evaluate policies at a granular level, moving beyond broad segmentation approaches that often lead to risks being over- or underpriced. Our novel modeling approach incorporates multiple model families and calibration methods to create a product which is best-in-class at ranking policy risks within a book of business.
We are a team of 10 and growing quickly. As our Actuarial Data Scientist, you’ll have the opportunity to
work with cutting-edge technology to iterate and improve on our process, as well as deliver production-ready models which directly impact our customers’ bottom line. This is a builder role: you will develop models, challenge assumptions, explain results, and help move validated work into production.
WHAT YOU'LL BE DOING
Model Development & Methodology
- Owning the development, evaluation, and iteration of policy-level pricing and risk models, including expected claim cost and loss-ratio models.
- Designing and reviewing cross-validation, parameter tuning, temporal holdouts, calibration, lift, uncertainty, explainability, and model-selection approaches.
- Building forward-looking underwriting and threshold strategies using information observable at the decision date, with a clear separation between deployable evidence and hindsight diagnostics.
- Using modern AI coding tools to accelerate research and implementation while applying the statistical and actuarial judgment needed to validate generated code and conclusions.
Actuarial Analysis & Economic Validation
- Developing and reviewing loss development, IBNR and IBNER, loss and premium trends, rate on-leveling, earned premium, ultimate loss, and expense assumptions used in model evaluation.
- Reconciling modeled data to customer control statistics before analysis and investigating differences in premium, losses, claim counts, fee income, and other economic drivers.
- Designing backtests and EBITDA analyses with assumptions locked before outcomes are evaluated, and clearly communicating uncertainty and limitations.
- Partnering with the CEO and Head of Engineering on model-methodology decisions and translating actuarial economics into practical underwriting actions.
Production Modeling & Platform
- Running reproducible experiments in Databricks and MLflow through our internal modeling orchestration pipeline and custom Python model-training library, including model YAML configuration and experiment tracking.
- Working with Engineering to productionize approved models in SageMaker and validate the scoring API, model artifacts, input contracts, and deployment checks.
- Defining model monitoring for predictive performance, calibration, score distribution, data drift, threshold behavior, explainability, traceability, and business impact after launch.
- Partnering with Data Engineering to ensure data pipelines provide complete, well-defined, point-in-time-correct model inputs.
Customer Onboarding & Cross-Functional Leadership
- Leading the modeling workstream for new customer implementations, from data interpretation and actuarial assumptions through backtest, results review, threshold guidance, and soft launch. Supporting customer presentations with model methodology, performance, profit implications, and limitations to customer actuaries, executives, product leaders, and technical teams.
- Creating durable model documentation, experiment notebooks, decision records, and review checklists that make the work repeatable by the broader team.
- Identifying reusable modeling patterns across customers while recognizing where differences in product, state, data system, and economics require customer-specific judgment.
OUR CURRENT STACK
- Python, SQL, and PySpark through the Databricks platform
- LightGBM and related statistical and machine learning methods for insurance risk modeling
- MLflow for experiment tracking, model lineage, validation artifacts, and deployment handoff AWS, including S3, SageMaker, Lambda, and production API infrastructure
- Internal, configuration-driven tooling for model orchestration, cross-validation, tuning, scoring, and model checks
- GitHub and modern generative AI development tools, including Claude, ChatGPT or similar models
ABOUT YOU
You must have the following:
- 5+ years of experience developing and validating predictive models in insurance, pricing, risk, or another data-intensive environment.
- Strong Python and SQL skills, with the ability to independently investigate data, build and review models, and produce reproducible analysis.
- Strong command of statistical learning concepts, including cross-validation, regularization, parameter tuning, bias and variance, calibration, explainability, and comparison of competing model approaches.
- Experience designing leakage-resistant validation strategies, including temporal holdouts and point-in-time feature controls, that provide a realistic assessment of performance on future data.
- Ability to diagnose model-performance and stability issues, determine whether problems originate in the data, methodology, implementation, or evaluation design, and recommend defensible next steps.
- Demonstrated experience developing and iterating on supervised machine-learning models, from target and feature construction through model selection, evaluation, and interpretation.
- Familiarity with core P&C insurance concepts, including loss development, trend, rate on-leveling, credibility, and expense provisions.
- Excellent communication skills and the confidence to explain technical tradeoffs to actuaries, engineers, underwriters, and executives.
- Comfort operating with significant ownership and ambiguity in an early-stage company where research, implementation, and customer work frequently overlap.
- The judgment to use AI tools effectively without outsourcing methodological review or accepting plausible-looking results without verification.
You’d be a great fit if you also have:
- A strong actuarial foundation with hands-on machine learning experience. ACAS, FCAS, or meaningful progress toward an actuarial credential is strongly preferred, but equivalent practical expertise is also welcome.
- Experience developing or reviewing insurance pricing, underwriting, frequency, severity, pure premium, claim cost, or loss-ratio models using GLMs, gradient boosting, or similar approaches.
- Experience supporting state Department of Insurance filings or regulatory reviews involving predictive models, rating algorithms, actuarial assumptions, model documentation, or responses to regulator questions.
- Experience working in an early-stage insurtech or another environment requiring substantial individual ownership.