Workforce Bridge Program — Intern Pipeline Analysis

People analytics project built on synthetic, BLS/EEOC-modeled public data.
A full-stack workforce analysis covering pipeline attrition, employer quality scoring, cost modeling, volume trends, concentration risk, early exit prediction, and fiscal year budget reconstruction.

---

Project Overview

This project analyzes a simulated intern placement pipeline across five campuses and multiple employer partners spanning Spring 2022 through Summer 2026 (~1,223 placements). All data is synthetic and modeled after publicly available BLS and EEOC distributions. No real student, employer, or institutional data is used.

Analytical sections:

Part	Focus
1	Outcome classification and pipeline funnel
1B	Full-term (9-month) completion rate analysis
2	Employer quality scoring (composite, 0–100)
3	Cost per completed placement by employer
4	Term-to-term volume trend with rolling average
5	Role concentration risk (HHI-style index)
6	Early exit prediction model (3 ML classifiers)
6B	Fiscal year budget reconstruction
7	Policy recommendations
8	Deep-dive on high-volume employer partners

---

Key Findings

• Employer quality score correlates with planned completion rate at Pearson r = 0.968
• Early exit prediction best AUC: 0.590 across Logistic Regression, Random Forest, and Gradient Boosting (5-fold stratified CV) — interpretable baseline, not production-grade
• At-risk threshold: employer partners with > 30% student exit rate are flagged for review
• HHI concentration: role-employer pairings scored as LOW (< 1,500), MODERATE (1,500–2,500), or HIGH (> 2,500)
• Cost model uses $18.00/hour flat rate, 17 avg hours/week, 270-day max placement duration, plus $75k coordinator overhead and 8% admin load

---

Figures Generated

The notebook produces 11 figures saved to the working directory:

File	Description
fig1_outcome_distribution.png	Overall outcome distribution
fig2_exit_timing.png	Attrition timing histogram + CDF
fig3_outcome_by_campus.png	Completion rate by campus (stacked bar)
fig3b_full_term_completion.png	Full-term completion distribution
fig3c_full_term_by_employer.png	Full-term rate by employer
fig4_employer_quality.png	Employer quality scores
fig5_cost_per_completion.png	Cost per completion by employer
fig6_term_volume_trend.png	Term-to-term volume trend
fig7_top_roles.png	Top 7 roles by placement volume
fig8_role_concentration.png	Role concentration index
fig10_fiscal_year_budget.png	Fiscal year budget requirement

---

Data

The notebook reads from a local Excel file (WBP_Anonymized.xlsx) with three sheets:

• Total Placements
• ARCHIVED-Ended Assignments
• PLACED-No Call and No Shows

This file is not included in the repository — it is synthetic data stored separately. To run the notebook, mount your Google Drive and place the file at:

/content/drive/MyDrive/workforce_bridge/WBP_Anonymized.xlsx

---

Tech Stack

Tool	Use
Python 3.10	Core language
pandas / numpy	Data manipulation
matplotlib / seaborn	Visualization
scikit-learn	ML models (Logistic Regression, Random Forest, Gradient Boosting)
scipy	Pearson and point-biserial correlations
openpyxl	Excel file parsing
Google Colab	Notebook runtime
Google Drive	Data file mount

---

Setup

pip install -r requirements.txt

Then open WBP_Pipeline_Analysis.ipynb in Google Colab or Jupyter and mount your Drive when prompted.

---

Employer Quality Score

The composite quality score (0–100) is weighted as:

• 50% — completion rate
• 30% — retention rate
• 20% — re-engagement rate

Only employer partners with 5 or more placements are included in scoring.

---

Early Exit Prediction Model

• Target: early_exit = 1 if student exited within 60 days
• Features: campus (label-encoded), employer quality tier, season, hours
• Evaluation: 5-fold stratified cross-validation, ROC-AUC
• Note: AUC results reflect limited feature availability in a synthetic dataset. This section demonstrates methodology, not a deployable model.

---

Notes

• All employer names are anonymized (e.g., "Generic County IT", "Generic County Public Health")
• Structured placement programs (Accenture Federal Services, Quantum Institute Fellows, Inc) are excluded from attrition timing analysis due to known program design differences
• All metrics are computed from synthetic data and should not be interpreted as real institutional outcomes

---

Author

Alexis Prieto
MS Computer Science — Texas A&M University San Antonio (December 2025)
SHRM-CP | People Analytics | HR Systems and Automation