Risk Analytics Pipeline

The problem

Workforce and operational data lived in separate systems — HR, payroll, and business-unit tools — each with its own employee identifiers, department codes, and location formats. Executive reporting was slow to produce and hard to trust: the same employee could appear under three different ID formats depending on the source system, and nobody had a single validated view of risk exposure across the organization.

How it's built

The pipeline runs in three stages, each owned end to end — from raw source extraction through the model that leadership actually reads.

SQL & the database layer

Snowflake serves as the enterprise warehouse. The core integration query standardizes identifiers across source systems before any downstream reporting or modeling touches the data:

-- unify workforce records across HR, payroll, and operations
CREATE OR REPLACE TABLE risk.workforce_unified AS
SELECT
    h.employee_id,
    COALESCE(p.dept_code, h.dept_code)      AS department_code,
    COALESCE(o.location_id, h.location_id)  AS location_id,
    h.hire_date,
    p.overtime_hours,
    o.incident_count,
    o.turnover_flag
FROM hr.employee_roster h
LEFT JOIN payroll.hours_summary p
    ON h.employee_id = p.employee_id
LEFT JOIN ops.incident_log o
    ON h.employee_id = o.employee_id
   AND h.location_id  = o.location_id
WHERE h.status = 'active';

Python & the predictive model

from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score

X_train, X_test, y_train, y_test = train_test_split(
    features, labels, test_size=0.2, random_state=42
)

model = RandomForestClassifier(
    n_estimators=300, max_depth=12, random_state=42
)
model.fit(X_train, y_train)

preds = model.predict(X_test)
print(accuracy_score(y_test, preds))  # 0.95

Results

The pipeline replaced a manual, spreadsheet-driven process with a repeatable one — and gave leadership a forecasting signal that didn't exist before.