Provide Data Science and Data Engineering services to clients in Financial, Insurance, and Utilities industries. Operated in MS SQL Server DB, Databricks, and AWS environments using Notebooks,Scikit-Learn, Pandas, Polars, Keras, NumPy, SciPy and built custom Python libraries.Reduced model builds from 1-2 weeks to <1 day with standardized SQL queries and white-boxed Artificial Intelligence methods: data wrangling, feature engineering/reduction, missing value imputation, outlier… Show more Provide Data Science and Data Engineering services to clients in Financial, Insurance, and Utilities industries. Operated in MS SQL Server DB, Databricks, and AWS environments using Notebooks,Scikit-Learn, Pandas, Polars, Keras, NumPy, SciPy and built custom Python libraries.Reduced model builds from 1-2 weeks to <1 day with standardized SQL queries and white-boxed Artificial Intelligence methods: data wrangling, feature engineering/reduction, missing value imputation, outlier analysis, collinearity, algorithm choices, model performance metrics.Data Engineering consulting to clients on custom ML pipelines in Databricks by optimizing SQL queries, choosing when to use Spark vs. Pandas DataFrames, utilizing data persistence, and software development techniques for Data Science – runtime reductions from 31%-86%.Built response models on datasets ranging from 250k to 40MM via XGB or Logistic Regression using limited a set of Acxiom elements (~220) with F1 scores ranging between 67.2-91.4.Co-created Direct Mail strategy to generate 28.% higher Inbound call rate and $9.2MM in incremental assets, optimizing mix between for audience segment creatives.Predicted Inbound Calls at the campaign level using GLM (99.2% accuracy ±2.5%), then at the cell level (~550 cells) using GEE (87.4% accuracy ±2.5%) using cohort/grouped data – guided where to invest in marketing matrix design for optimal response.Built Times Mailed (TM) table, initially as a Polars dataframe (78.1% time reduction from Pandas) then as columnar DB table – adopted as Best Practice and built into Production. Used by Data Science to build predicted TM into models as a highly predictive feature.Scraped Econometric data (M2, CPI, GDP, et al) to include in analyses/modeling to explain and control for exogenous factors.Authored and maintained all internal shared Python code libraries to expedite model building, exploratory data analysis, wrangling, statistical testing (A/B, DOE, MVT), and power analyses. Show less

Set team culture/vision/mission, project direction, technical oversight, and professional development for 2 Data Scientists, 2 Data Engineers, and 1-2 IT Program. Engineer and self-support infrastructure and Data Science environment.Researched techniques to equip Marketing with models to optimize each business process in the value chain (Response, Conversion, Application, Time-to-Contact). Authored the semantic layer from Oracle database to process milestones, used Acxiom data to model… Show more Set team culture/vision/mission, project direction, technical oversight, and professional development for 2 Data Scientists, 2 Data Engineers, and 1-2 IT Program. Engineer and self-support infrastructure and Data Science environment.Researched techniques to equip Marketing with models to optimize each business process in the value chain (Response, Conversion, Application, Time-to-Contact). Authored the semantic layer from Oracle database to process milestones, used Acxiom data to model using eXtreme Gradient Boosting, then posted scores to Salesforce.com daily via Apache Airflow.Increased retention of best leads, forecast accuracy, and Best Practice testing with Graph model that quantified each Sales action as increasing/decreasing the Probability to Apply.Built challenger models for Actuarial Experience using base table features and clustering additional covariates for Mortality table (Life), cluster Daily Benefit Amounts for table use, cluster A/Es by Claim Incidence/Benefit Utilization/Claim Termination, and model Situs (LTC).Gave Underwriting the ability to test question efficacy over time by scraping applications to build an application question database and used Natural Language Processing techniques to standardize 18MM prescription entries from notes fields into modeling-ready features.Reduced time-to-model to 1 business day via by authoring feature engineering, multivariate significance, outlier, and initial model performance (20 models, 9 metrics) pipelines via Kedro.Decreased model building and analysis time by 12% through engineering a multi-asset/multi-core cluster to distribute Apache Hadoop, Spark, and Flink. Used to build to discover architectures for Deep Learning models to explain and predict business events for partners.Decreased new model build determination by 2 weeks via creating model scorecard analysis (every model in Production), emailing test results and graphics weekly via Apache Airflow. Show less

Identified opportunities to build value into the organization. Utilized open-source software for all solutions in-house and built a Data Science team to grow new sources of analytical value. Built Fraud detection system that saved 37% more in economic avoidance and reduced time- to-investigate by 19% and then historical processes by using multiple techniques (anomaly/outlier detection, deep learning/variational autoencoders, time series, decision tree, CHAID, logistic regression) to… Show more Identified opportunities to build value into the organization. Utilized open-source software for all solutions in-house and built a Data Science team to grow new sources of analytical value. Built Fraud detection system that saved 37% more in economic avoidance and reduced time- to-investigate by 19% and then historical processes by using multiple techniques (anomaly/outlier detection, deep learning/variational autoencoders, time series, decision tree, CHAID, logistic regression) to feature engineer and model ongoing LTC Claims Fraud propensity.Reduced new LTC application question set by 31% using Decision Trees to model future customer state (Current vs. Claimed/Exhausted/Cancelled/Death) with ~90% accuracy.Built and beta-released application Fraud model into Underwriting system, on 8 billion record XML database via Hadoop. Built Semi-Supervised Logistic Regression model to grow the modeling sample. Partnered with Underwriting to feature engineer, build process flow, create system verbiage, and pass all Compliance and IT/Architecture reviews. Show less

Enhanced Marketing efforts and extended reach of Sales (Wholesalers) by modeling Agent / Advisor behavior. Linked best treatments to agencies with the greatest likelihood to sell consumers Genworth wanted as customers. Democratized analytics across the organization.Scored the entire U.S. population using proprietary Consumer Segmentation scoring models. Used by senior leadership to inform strategy with preferred customer segments. Built into corporate systems for access by Actuarial… Show more Enhanced Marketing efforts and extended reach of Sales (Wholesalers) by modeling Agent / Advisor behavior. Linked best treatments to agencies with the greatest likelihood to sell consumers Genworth wanted as customers. Democratized analytics across the organization.Scored the entire U.S. population using proprietary Consumer Segmentation scoring models. Used by senior leadership to inform strategy with preferred customer segments. Built into corporate systems for access by Actuarial, Operations, Finance, Sales, Marketing, et al. Met 100% goals of attendees, leads and agents for new LTC agent recruitment. Created target set, utilized Discovery data and internal databases to build likelihood models, scored prospects, and optimized calculations subject to minimizing all costs.Increased LTC Group Sales enrollment unit-rate by 48.3% and premium by $8.7MM, applying proprietary U. S. Consumer segmentation system for direct mail campaigns over a 2-year period.Increased Wholesaler effectiveness by ranking appointments via score of Binary Logistic model that quantified agent likelihood to use eApp proprietary application software. Show less

Conducted wing-to-wing analysis (Database, Mail, Response, Agent Assignment, Application Submission, Policy Placement) of LTC direct mail program on 41MM record dataset in SAS via Unix. Analysis findings included utilizing a segmentation approach to identify potential annual savings ranging from $3.1MM - $5.4MM.Built time series models to forecast monthly agent web usage rates to schedule site enhancements/upgrades.Built Logistic Regression model to quantify propensity of using… Show more Conducted wing-to-wing analysis (Database, Mail, Response, Agent Assignment, Application Submission, Policy Placement) of LTC direct mail program on 41MM record dataset in SAS via Unix. Analysis findings included utilizing a segmentation approach to identify potential annual savings ranging from $3.1MM - $5.4MM.Built time series models to forecast monthly agent web usage rates to schedule site enhancements/upgrades.Built Logistic Regression model to quantify propensity of using proprietary application software. Utilized dimension reduction (principal components and binary factor analysis), outlier analysis (cluster analysis using various structures), missing value imputation (multiple imputation, random draws from distributions) and collinearity reduction. Utilized by Sales to prioritize wholesaler visits (minimize trips, maximize time w/agent). Influenced insurance general agencies to obtain information on their agents to score using proprietary Insurance Advisor segmentation system. Coordinated data transfer, scored agents using models, built and complied presentations, presented findings, and consulted on using segmentation to drive the business.Utilized proprietary U. S. Consumer segmentation system for LTC Group Sales direct mail campaigns, which lead to a 48.3% increase in enrollment rates and premium of$8.7MM. Show less

Received Management Award for work on team that built industry-wide Insurance Advisor segmentation system and built Multinomial Logistic Regression models to predict segment membership with Acxiom data.Received Management Award for work on Genworth’s annual ‘Cost of Care’ study. Designed sampling system to QA against manual input errors and built compound annual growth rate cost calculations by locale. Conducted Next Sale Propensity or “Stickiness” analysis of Insurance… Show more Received Management Award for work on team that built industry-wide Insurance Advisor segmentation system and built Multinomial Logistic Regression models to predict segment membership with Acxiom data.Received Management Award for work on Genworth’s annual ‘Cost of Care’ study. Designed sampling system to QA against manual input errors and built compound annual growth rate cost calculations by locale. Conducted Next Sale Propensity or “Stickiness” analysis of Insurance Agents/Advisors utilizing proprietary application software. Recommended as part of quarterly Executive Strategic Review.Developed multivariate matching sampling algorithm to simultaneously control for multiple metrics (e.g., sales dollars, sales units) to select test/control groups for marketing campaigns. Show less

Received Management Award for creating system to obtain, cleanse and merge data, sample test/control cells, sum agent contacts, and weekly performance reports with executive summaries and trend analysis.Received Management Award for building company’s 1st Insurance Agent/Advisor marketing campaign database process, integrating/standardizing/transforming data from multiple disparate internal systems.Designed company’s 1st large-scale measurable Marketing campaign to Insurance… Show more Received Management Award for creating system to obtain, cleanse and merge data, sample test/control cells, sum agent contacts, and weekly performance reports with executive summaries and trend analysis.Received Management Award for building company’s 1st Insurance Agent/Advisor marketing campaign database process, integrating/standardizing/transforming data from multiple disparate internal systems.Designed company’s 1st large-scale measurable Marketing campaign to Insurance Agents/Advisors utilizing test/control methodology. Resulted in $24MM lift in sales of Variable Annuities.Built algorithms to join datasets without merge keys (used for Marketing, Sales, Licensing, et al). Show less

Developed strategies, targeted audiences and drove behavior for clients in Automotive, Banking, CPG, Insurance, Print Media, Retail, Spirits, Telecommunications/VOIP and Travel industries.Built likelihood/propensity models for a national advocacy group with strong brand and numerous affinity partners. Built Multinomial Logistic Regression model to predict optimal offer assignment for 40MM+ members.Built survival analysis (time-to-event) model on Retail Housewares client purchase… Show more Developed strategies, targeted audiences and drove behavior for clients in Automotive, Banking, CPG, Insurance, Print Media, Retail, Spirits, Telecommunications/VOIP and Travel industries.Built likelihood/propensity models for a national advocacy group with strong brand and numerous affinity partners. Built Multinomial Logistic Regression model to predict optimal offer assignment for 40MM+ members.Built survival analysis (time-to-event) model on Retail Housewares client purchase data to develop marketing strategy around timing of “Next Purchase” teaser offers to frequent, occasional, and infrequent purchaser segments.Performed Cluster Analysis—resulting in 6 actionable segments—to focus marketing messaging and positioning efforts to tailor offers that resonated with target consumers for a Travel Industry client.Built response and activation models using Acxiom data for top-tier Cellular provider, then devised multi-stage sampling plans for marketing test/control cells from 40+ vertical lists. Show less

Jointly designed mail strategies with Marketing and Risk for Home Depot Private Label Consumer credit card programs. Utilized a Mine/Test/Analyze/Model/Implement process to meet corporate EBIT and ROA hurdles.Obtained historical performance data with IT and assisted building the direct mail P&Ls with Finance to project campaign performance.Part of team that built response model using tri-bureau data (Equifax, Experian, Trans Union) via Logistic Regression in SAS for use in name… Show more Jointly designed mail strategies with Marketing and Risk for Home Depot Private Label Consumer credit card programs. Utilized a Mine/Test/Analyze/Model/Implement process to meet corporate EBIT and ROA hurdles.Obtained historical performance data with IT and assisted building the direct mail P&Ls with Finance to project campaign performance.Part of team that built response model using tri-bureau data (Equifax, Experian, Trans Union) via Logistic Regression in SAS for use in name selection in the Consumer pre-approved direct mail channel.Reviewed direct mail specification documents and provided feedback on creative, messaging and positioning. Show less

Built bank's first response model, which equipped the bank to take control over direct mail (from a third-party leads provider) and enabled a profitable sale.Model was built using Experian credit bureau data in SAS using Binary Logistic Regression; coarse classing (binning) of select features accomplished via Weight of Evidence.Designed / tested strategies to mail below cutoff scores profitably and expand mailable universe.Conducted statistical tests of campaign… Show more Built bank's first response model, which equipped the bank to take control over direct mail (from a third-party leads provider) and enabled a profitable sale.Model was built using Experian credit bureau data in SAS using Binary Logistic Regression; coarse classing (binning) of select features accomplished via Weight of Evidence.Designed / tested strategies to mail below cutoff scores profitably and expand mailable universe.Conducted statistical tests of campaign, strategy and creative performance and presented 5-year NPV of each program monthly to all Chief Executives.Redesigned P&L from simple accounting tool to named-range-driven dynamic forecasting model.Supported bureau conversion from Trans Union to Experian.Evaluated new business economics with partners such as Tiger Direct, Brandsmart, and Dell. Show less

Built response, conversion, net conversion and premium models for Progressive Auto Insurance joint venture direct mail program. Utilized Trans Union credit bureau and Acxiom data as well as SAS (PC and Unix) and CHAID softwares.* Techniques: Binary Logistic Regression, Ordinary Least Squares, Cluster Analysis, Decision Trees.Conducted statistical significance tests on creatives, process changes and modeling strategies. Created dynamic spreadsheet model that incorporated… Show more Built response, conversion, net conversion and premium models for Progressive Auto Insurance joint venture direct mail program. Utilized Trans Union credit bureau and Acxiom data as well as SAS (PC and Unix) and CHAID softwares.* Techniques: Binary Logistic Regression, Ordinary Least Squares, Cluster Analysis, Decision Trees.Conducted statistical significance tests on creatives, process changes and modeling strategies. Created dynamic spreadsheet model that incorporated prior mail volumes with response, conversion and premium metrics. Monthly campaign performance was forecasted by state accounting for mailable volume. Performance was unbeaten by updated challenger methodologies until entire direct mail process was overhauled four-and-a-half years later.Built all campaign performance analytics, linking SAS -> Excel -> Powerpoint. Reduced prior creation time by 70%, then re-engineered the process to include dynamic forecasted performance for a total processing time reduction of 80%. Show less

Served as lead statistical programmer on OCC & FCRA reviews; interviewed by examiners.Lead statistical programmer on two HELOC securitizations (each over $1B).Lead statistical programmer on RFPs for appraisal, flood and title vendors.Built forecast models for delinquency using historical performance data via roll-rate and vintage methodologies.Set Credit Policy by conducting background studies to set DTI, LTV limits for Home Equity products.Established… Show more Served as lead statistical programmer on OCC & FCRA reviews; interviewed by examiners.Lead statistical programmer on two HELOC securitizations (each over $1B).Lead statistical programmer on RFPs for appraisal, flood and title vendors.Built forecast models for delinquency using historical performance data via roll-rate and vintage methodologies.Set Credit Policy by conducting background studies to set DTI, LTV limits for Home Equity products.Established reporting process via extensive Lotus 1-2-3 macros for entire Retail Credit Risk Portfolio for all banks in 16 states.Utilized SAS, Lotus 1-2-3, MS Excel / Powerpoint / Word, NOMAD, RCO. Show less