Back to Projects
demoDatadataml

Geospatial Revenue Analytics

End-to-end analytics pipeline converting raw QuickBooks exports into geospatial and time-series analysis, deriving sales territories and revenue insights.

2025-09 — 2025-12
View Source

Key Highlights

  • 65% high-confidence customer geolocation coverage
  • Derived 6 sales territories capturing 75% of revenue
  • End-to-end pipeline from QuickBooks exports to ML analysis

Overview

A comprehensive data analytics project that transforms raw financial data into actionable business intelligence through geospatial clustering and time-series analysis.

Problem

Raw QuickBooks exports contain valuable business data but require significant processing to extract insights about customer distribution, revenue patterns, and optimal sales territories.

Solution

Built an end-to-end analytics pipeline that normalizes raw data, recovers customer locations, identifies sales territories, and analyzes seasonal revenue patterns.

My Contributions

  • Built complete data pipeline converting QuickBooks exports to normalized relational tables
  • Implemented multi-stage geolocation resolution achieving 65% high-confidence coverage
  • Validated clustering approach using Hopkins statistic
  • Derived 6 operational sales territories capturing 75% of total revenue
  • Performed seasonal decomposition to identify demand cycles

    • Technical Details

    Python with Pandas handles data wrangling and transformation. GeoPandas enables spatial analysis and visualization. Scikit-learn provides clustering algorithms. SQL manages the normalized relational structure.

    Dataset & Evaluation

    Dataset: Real QuickBooks transaction exports with customer and invoice data

    Evaluation: Hopkins statistic for cluster tendency, silhouette scores for cluster quality

    Results: 6 distinct territories, one region accounting for ~75% of revenue

    Challenges & Tradeoffs

    Challenge: Many customer records lacked explicit geolocation data.

    Solution: Implemented multi-stage resolution using address parsing, zip code lookups, and city-level approximations, achieving 65% high-confidence coverage.