Delivery Delay Predictor — Operations Dashboard

Priority Alert List

Open orders ranked by delay risk, scored before dispatch by a Vertex-tuned model. Act on the highest first.

Order ID	Courier	Zone	Service	Daily Vol	Risk	Key Driver

AI Recommendations Gemini ✓ Verified

Each recommendation is checked for accuracy and usefulness (see Model Performance).

Courier Performance

Delivery Zone Analysis

Operational Bottleneck

Live from BigQuery (2,409 orders) · Risk scores from a Vertex-tuned, calibrated XGBoost + SHAP model · Recommendations verified for accuracy and usefulness · BigQuery + Vertex AI + Gemini + Cloud Run · © 2026

Model Performance

What the refinement changes for the operations manager: a shorter, more trustworthy shortlist of at-risk orders. The figures are validated on held-out orders so a reviewer can verify them.

v1 vs refined (held-out test)

Tuned on Vertex AI, probabilities calibrated. Higher is better. Reproducible from cleaned.csv.

Validated metrics

The refined model improves on every measure: more precise, higher recall, better ranking, better calibrated.

Recommendation Quality ✓ Verified

Every Gemini recommendation is scored by an evaluation (LLM-as-judge) before the manager sees it.

Grounded in the actual metrics, with no invented numbers
Actionable for an operations manager
Relevant to the current orders on screen

Model metrics validated on a held-out 25% of the 2,409 orders, from the model tuned on Vertex AI, and reproducible from cleaned.csv. Recommendation quality measured over 18 recommendations across 6 scenarios.