Project / Fall 2025

B2B Sales Pipeline Analytics

Year

Fall 2025

Type

Data Mining · ML · R

Skills

R

dplyr

K-Prototypes Clustering

Data Engineering

Feature Engineering

CRM Analytics

What I Built

Applied k-prototypes clustering to a mock B2B CRM to surface pipeline segments from messy, mixed-type data.

Applied k-prototypes clustering to a mock B2B CRM in a Data Mining course, surfacing five distinct pipeline segments from mixed-type data. The dataset was synthetic and the scale was modest, but the work taught me how much of real ML happens before the model runs.

One of my earlier data projects. Part of my Data Mining course. Applied three core techniques — classification, clustering, and association rule mining — to a domain I actually wanted to work in: B2B SaaS sales. Took a mock CRM dataset with four relational tables (leads, accounts, products, sales agents) and merged them into a single analysis-ready master table using R. Performed significant data engineering & cleaning. Ran k-prototypes clustering and mapped aggregate profiles for each cluster. The dataset was mock, the scale was modest, and this was a class project — but it was a good experience to wrangling *somewhat* realistic data. VISUALS & MODEL EVALS COMING SOON.

Sales Motion

Mapped the full B2B sales motion from prospecting through close

Pipeline Segmentation

5 distinct deal/customer segments surfaced from CRM behavior and firmographics

Business Use

Built to support lead prioritization, sales cadence, and cross-sell planning

POST-PROJECT REFLECTION

What I Took Away

What I learned. Ranges anywhere from technical skills, to general lessons.

My goal is to always learn from an experience, no matter how big or small.

Real-world CRM data (even a mock one) is messy. A naming mismatch between tables ("GTXPro" vs. "GTX Pro") would have silently broken the product join. 1,425 NAs in firmographic columns all came from the same independent accounts — dropping them would have removed a legitimate segment. 500 NAs in engage_date all corresponded to prospecting-stage deals — dropping them would have deleted signal, not noise. Each of those required a judgment call, not a formula. The actual model ran in minutes. The cleaning and feature engineering ran for days.

POST-PROJECT REFLECTION

What I'd Do Differently

Redos of project parts I was unsatisfied with, or simply things I would've done differently.

Redo 01

When I first did this project, it was part of a simple assignment for my data mining class. I was limited by both my ML knowledge at the time and the assignment requirements, and wasn't able to fully dive into what I would do if I were doing this for a real job. Recently looking back, I realized just what an interesting problem it was, so I chose to redo it.

Coming soon.