Senior Predictive Liability Analytics Lead

Jobtailor; California (MO); On-site; USD 140,000 - 220,000; Full time

Job summary

BSRM is seeking a Senior Predictive Liability Analytics Lead to build next‑gen short‑term models of policyholder behavior, starting with annuity surrenders, withdrawals, and utilization. You will mentor by example and design models with GLMs/GAMs, survival methods, tree ensembles, and deep learning.

You'll leverage unstructured data with NLP/transformer features, pilot generative AI, and translate outputs for projection engines while collaborating with valuation/ALM teams and driving governance.

Qualifications

• 7+ years building production predictive models in insurance/long-duration liability.
• Experience with behavior modeling (surrender/utilization/lapse) and integration with projection engines.
• Strong Python/SQL coding with NLP/LLM and unstructured data experience.
• Clear, concise communicator with good documentation habits and governance.

Responsibilities

• Lead hands-on predictive liability analytics for short-term policyholder behavior.
• Build models for annuity surrenders, withdrawals, and utilization.
• Collaborate with FP&A, ALM, pricing, valuation and capital teams.

Skills

Python

ML/AI

SQL

R

Pandas/NumPy

scikit-learn

XGBoost/LightGBM

NLP/LLM

Databricks/Spark

PyTorch/TensorFlow

Communication

Model deployment

Education

Master’s/PhD in Statistics, Data Science/ML, Applied Math, CS, or Actuarial Science

Tools

Databricks

Snowflake

AWS/Azure

Tableau/Power BI

Job description

Responsibilities

• Join BSRM as our hands‑on Senior Predictive Liability Analytics Lead.
• You’ll build next‑gen short‑term models of policyholder behavior starting with annuity surrenders, withdrawals, and utilization.
• Collaborate with other stakeholders such as financial planning, ALM, pricing, valuation and capital as appropriate.
• This is a technical leadership role (not initially a people‑manager or strategy role).
• You’ll do the math, write the code, build models, and mentor by example.
• Design predictive and semi‑structural models with a short‑term focus (performance focused on fitting the next 18‑36 months) on long‑horizon behavior (surrenders, partial withdrawals, lapse, rider utilization) using GLMs/GAMs, survival & hazard models (Cox, discrete‑time, competing risks), tree ensembles (Boost/LightGBM/CatBoost), and deep learning choosing the most appropriate tool depending on the business problem.
• Leverage unstructured data (contract text, correspondence, customer relationship notes, call transcripts) via NLP/transformer embeddings, RAG pipelines, and LLM‑assisted document parsing to create novel behavioral features within guardrails.
• Pilot generative‑AI (foundation models) for feature extraction/summarization; use genetic/evolutionary algorithms for feature selection, architecture search, or synthetic cohort generation when appropriate.
• Make models scenario‑aware: incorporate drivers like credited rate, market rate spreads, moneyness, surrender charge state, distribution channel effects; calibrate elasticity to economic conditions documented in industry studies.
• Translate model outputs into curves/driver functions consumable by projection engines (e.g., Moody’s AXIS, Aon Pathwise, Prophet, RAFM, or internal models); generate reproducible, versioned results tables.
• Share models with valuation/projection/ALM teams so behavior sensitivities can be considered alongside assumptions that normally flow through cash‑flow projections, LDTI assumption updates, RBC/CTE stresses, and hedge effectiveness studies.
• Partner with valuation and pricing to reconcile actual vs. expected and attribute earnings/variance to behavior; document the “model story” and explainability for governance.
• Build training/scoring pipelines in Python/SQL on Databricks/Spark/Snowflake/AWS; track experiments with MLflow/DVC, version in Git, package with containers, and serve via batch/API.
• Stand up dashboards for calibration, drift, stability, and bias; set retraining schedules, fallback models, rollback criteria, and automated alerts.

Requirements

• Master’s/PhD in Statistics, Data Science/ML, Applied Math, Computer Science, or Actuarial Science; FSA/ASA a plus (or equivalent domain depth).
• Certifications in ML/AI (nice to have).
• 7+ years building production predictive models; insurance/annuity or long‑duration liability exposure preferred.
• Practical wins in behavior modeling (surrender/utilization/lapse) and integration.Comfortable spanning structured + unstructured data and bridging to projection engines.
• Clear, concise communicator; strong documentation habits; bias to ship and iterate.
• Mentors by example; sets standards for code quality, reproducibility, and testing.
• Balances accuracy, interpretability, and operational simplicity under governance.
• Collaborate within a highly matrixed organization.
• Python (pandas, NumPy, scikit‑learn, XGBoost/LightGBM, PyTorch/TensorFlow), SQL, R.
• NLP/LLM: transformers/embeddings, RAG, prompt engineering; genetic/evolutionary search for features/hyper‑params.
• Databricks/Spark, Snowflake, AWS/Azure; MLflow, model registries, CI/CD; Tableau/Power BI for monitoring & storytelling.
• Working familiarity with Actuarial platform (AXIS/Prophet/RAFM/etc.) integration patterns (assumption tables, mapping layers).