Cevriye Horn

Senior lead data scientist leading modeling efforts for allegations in the Conduct Analytics and Reporting team.

Allegation Identification Model:

• Built a machine learning model with 95% precision, overcoming challenges with large and complex allegation text data.
• Collaborated with technology partners to implement the model in JupyterHub, ensuring smooth daily operations and model performance evaluation.
• Modularized (standardized) the model for text cleaning, model building, and implementation stages.

Allegation Data Module and Binary Text Classification Model Module:

• Created clean function format modules for building allegation data quickly for analysis and future modeling opportunities.
• Developed a detailed binary text module for scenario analysis, incorporating various text cleaning and modeling techniques. This tool is used by other team members to develop new models on allegation data.

Python in JupyterHub:

• Tested the feasibility of implementing models in JupyterHub and prepared documentation for team members on accessing and using the platform.

Copilot Summarization Tool:

• Developed a process for summarizing allegation text via Copilot, which involved extensive data engineering and text cleaning.
• Prompt Engineering: Created a comprehensive prompt for the Case Investigation summarization task and collaborated with business partners to finalize the work. This work saves our business partners hundreds of hours by quickly analyzing the allegation data content.

Senior lead data scientist in the Customer Excellence Data & Analytics and Reporting (CEDA) team, responsible for improving customer experience through the application of advanced analytics: data exploration and model development (Natural Language Processing) using unstructured voice of customer data (speech and written communications), structured operational data, and customer interaction data.

Easy-to-resolve complaints identification model:

• Developed a model to identify complaints from the escalated population considered "easy to resolve," making them candidates for quick resolution or re-assignment as a service complaint. Applied feature engineering techniques via NLP and classification using logistic regression and gradient boosting.

Reg Specific Safety Net (Elder Financial Abuse) Model:

• Built a model to identify potential elder abuse occurrences in Voice of the Customer data (call transcripts).
• Applied feature engineering techniques via NLP, and explored classification using various statistical/ML techniques: GEE Logistic Regression, GBM, and Neural Networks via Keras/TensorFlow. This model, built in 2023, is still in use.
• Prepared an extensive Model Development Document (MDD), and obtained all necessary approvals from Corporate Risk.
• Analyzed model performance results, and created templates that are now used by many CEDA analysts.
• Extended the model's usage to Consumer, Small, and Business Banking Operations (CSBBO) with nearly 90% precision.
• Achieved significant impact in Fraud and Claims LOB, with potential savings of over $7 million for customers.

Complaint Reason Classification Model:

• Developed a model for classifying nearly 200 complaint reasons, utilizing strategies like binning similar classes, and stacking algorithms.
• Prepared a comprehensive deck explaining the methods that were used in a company-wide presentation.

Documentation: Prepared a Modeling Documentation (MDD) checklist, advising analysts on completing essential modeling steps, which accelerated model approval and implementation stages by addressing Model Governance requirements.

Consultation: Worked with technical and non-technical business partners to explain models and develop strategies to improve customer experience.

Modeling Lead for the Deposit product in the Marketing Insights & Analytics Team, responsible for leading targeting models to deepen Wells Fargo's existing customer relationships, and increase market share through the acquisition of new customers.

• Modeling Lead for the Deposit Product Group (DPG): Supported all modeling-related activities, including model development and validation, model implementation, model documentation, and model performance tracking; worked with Model Governance to obtain all necessary approvals from Corporate Model Risk and Legal & Compliance teams.
• Team Management: Managed Wells Fargo team members and external contractors (from Chile and India) supporting model development activities.
• Marketing Targeting Models: Developed several marketing targeting models for targeting existing Wells Fargo customers and pure prospects within the Deposit Product Group (DPG).
• Created a machine learning model identifying existing Wells Fargo non-deposit customers who are likely to bring over and retain new deposit balances in response to cash marketing offers.
• Digital Marketing Proof of Concept: Created a proof of concept model for the first digital marketing targeting effort, expanding audience optimization models to include paid digital touchpoints, balance and attrition behaviors, and sales channel specificity.
• Markov Chain Attribution Model: Applied the Markov Chain Attribution Model strategy to evaluate the effectiveness of different marketing channels by analyzing the probability of each touchpoint leading to a conversion. This approach provided detailed insights into the effectiveness of each marketing touchpoint, and supported strategic decision-making for optimizing marketing spend.
• Modeling Consultation: Provided consultation on modeling approaches, both inside and outside the department.
• Collaboration: Collaborated with marketing partners to establish mutual objectives, priorities, and measures of success.
• I actively engaged with other modelers to share ideas and led modeling training sessions in my team that I managed.
• Model Risk Management: Served as the "model owner" for all credit card targeting models, in addition to deposit targeting models. Among all modeling teams (not only within the Marketing Insights & Analytics department), I was the only person serving as both a model developer and a model owner (a role typically reserved for senior managers).

Customer Value Management (CVM) and Statistical Modeling:

• Customer Value Management (CVM): Built numerous statistical models with primary responsibility for Customer Value Management (CVM), including the annual refresh of personal lines retention, loss ratio, and premium models.
• Lifetime Value (LTV) Models: Explored customer relationships and transition states via Markov chain transition models, and evaluated the effectiveness of different product offerings leading to a conversion.
  Regularly presented model results to the finance department for strategic decision-making.
• Homeowners Insurance Pricing Models: Created homeowners and commercial lines segmentation models (generalized linear models) and underwriting models.
• Modeling Mentor: Served as a modeling consultant and mentor to other department members within the company.

Examples of projects completed:

• LTV Estimates Methodology: Developed a new methodology to calculate LTV estimates, enabling easier completion of the LTV process without the need to compute large transition matrices, and providing more flexibility in segmentation.
• Agency Cancel Models: Built agency cancel models using logistic regression and Markov transition models; examined the characteristics of cancelling agencies and their performance, and developed a model to estimate the number of agency cancellations.
• Territory Analysis: Performed extensive territory analysis by applying a spatial filtering method, creating location-dependent, but data-independent "spatial filters," and incorporating them into the GLM model as covariates.
• Fixed Expense Fees: Derived new fixed expense fees for Personal Lines Homeowners rating, with fixed expenses derived as a function of retention, allowing the implementation of different fees by segmentation.
• Rare Frequency Claims Classification: Built a machine learning model to classify rare frequency claims in the commercial property program.

• Completed pricing rate reviews for Personal Lines, Auto, and Homeowners programs.
• Assisted with the quarterly rate indication process, and was part of the by-peril rating implementation team for four years for the Personal Lines Homeowners program. Collaborated with Pricing, Product, and Research partners during the implementation.

Additional Accomplishments:

• Managed one pricing analyst (direct report).
• Created Excel templates for departmental use.
• Developed an automation process (via VBA code linking Excel to Word) for the Homeowner's By-Peril Program Agents' Manuals.
• Provided training to new teammates joining the Homeowner's By-Peril Pricing Team.
• Served as Captain of the Coaching and Feedback Values Team (June 2006 - January 2007).
• Served as Editor of the Department Newsletter from 2005 to 2010.
• Served as a Recognition Committee Member (2006–2010).
• Awarded by the Personal Lines Pricing Department for outstanding performance.
• Blue Pen Award (2 times).
• On Your Side VP (5 times).
• On Your Side Award (3 times).
• Standing Ovation (2 times).
• Most Valuable Participant Award (1 time).

• Performed statistical analysis of Wellmark Reports
  Performed claim frequency modeling and forecasting via Time Series Analysis
• Co-developed additional trend modeling for claim payments

• Taught introductory level statistics courses
• Conducted statistical research
• Awarded Graduate Study Scholarship in the U.S. by Turkish Higher Education Committee

• Translated ISO standards from English to Turkish

• Created new standards needed for Turkish norms