SNEHA DUTTA

Root Cause Analysis into deterioration of critical business metrics to reduce delivery cost

• Objective: To provide strategic solutions to help improve critical business metrics and reduce delivery cost by identifying the causes for the metrics performing badly.
• Performed in-depth analysis and led a cross-team collaborative war room to conduct extensive hypothesis testing and identify the root cause for the metric deterioration. Handled end-to-end process and drove the project to successful implementation.
• Proposed new assignment logic that helped save ~Rs. 1.2 crore/week on delivery cost. Won the Swiggstar award for this project.
• Technique/Software: SQL, Excel, Python, Power BI

Revise payout structure of delivery executives

• Objective: To identify and address any egregious errors in the DE payout system and propose implementable business solutions to improve the unit economics per order.
• Collaborated with teams across Business, Product and DE Governance to design data requirements. Extensively used statistical techniques to propose data driven insights that helped save ~Rs. 4.7 /order and ~10% decrease in delivery partner attrition.
• Technique/Software: Statistical hypothesis testing, advanced design of experiments, Pareto analysis, exhaustive variable identification, data visualization through Power BI dashboard to determine optimal course of action.

Identify top customers and provide differential customer experience

• Objective: To provide preferential treatment to high value customers by assigning top agents for their concerns and handling their issues on priority.
• Identified top customers and top customer care agents through statistical data analysis, leading to statistically significant improvement in all CX metrics like customer retention, NPS and CSAT.
• Technique: Principal Component Analysis, K-means clustering, Gaussian Mixture Model Clustering technique.
• Software: SQL, Python, Excel, Zeppelin, AirFlow

Building Analytics features and models on the production platform

• Objective: To help Risk Management Team make critical decisions regarding refunds and decrease false positive customer fraud tagging.
• Built 182 different features on the production platform, to enable data driven decisions on refunds and help re-train the fraud detection model.
• Lead SME in building features and models on the production platform, helping production onboarding across Analytics division.
• Created an alerting system by setting up quality gate alerts using Qubole notebook, which reduced the chance of model failures and helped debug failures.
• Technique/Software: SQL, Zeppelin notebook and Airflow to schedule automated query runs, staging/production platform to build and deploy models, Postman API for model testing.

Identify MOs for frauds and abuses by restaurants, delivery partners or customers

• Objective: To detect and curb MOs for cross-spectrum abuse/fraud.
• Analyzed and identified multiple abuse/fraud MOs, and executed suitable SOP changes, leading to ~Rs. 35 lakhs/week savings, and 5% increase in precision of customer fraud model.
• Technique used: Case-oriented deep dives, Automated real-time alerts to on-ground city-wise operations heads, Power BI dashboard to evaluate abuse trends and find new MOs.

Project allocation lead

• Managed a team of 3 through project assignment, bandwidth tracking, and tracking progress through measurable metrics, to effectively deliver on stakeholder requests and independent business innovative data-driven insights.

Machine Health Prediction Model

• Built real time models using real time data which is captured through API's in JSON format to predict abusive behavior of machines in the upcoming time frame.
• Technique Used: Boruta package for feature selection, Logistic Regression and Random Forest for building propensity model
• Software Used: R

• Data science driven solutions to big data problems in mechanical engineering scenarios. Predictive maintenance algorithms to identify abnormality in machine health. High frequency data analysis to identify deviations from expected behavior.