SecondChance

Animal shelters make time-sensitive decisions with limited staff, foster homes, outreach capacity, and hours in the day.

The people doing the work understand why some animals wait longer, but reviewing every case at the right moment is difficult. Public intake and outcome records contain useful patterns, yet those records need careful interpretation before they can support a decision.

A useful platform had to make those patterns actionable while keeping data freshness, uncertainty, and model limitations visible. A higher-risk result needed to become a reason for more support, never less.

Build a transparent decision-support product that connects shelter data with practical action.

SecondChance combines current City of Austin public shelter records with calibrated machine-learning estimates based on information available when an animal arrives. It estimates adoption within 30 days and the likelihood of a stay exceeding 60 days, helping users identify cases that may benefit from earlier attention.

The dashboard, support queue, record browser, outreach planner, and model documentation bring that information into one consistent experience. Observed outcomes, predictions, and planning assumptions remain clearly distinguished.

1. 01Dashboard for inferred open stays, adoption timing, intake and outcome trends, and model health
2. 02Searchable support queue with species, age, risk, and waiting-time filters, plus table and card views
3. 03Animal-record browser and individual detail views with source context and data-quality indicators
4. 04Weekly outreach planner with channel selection, staff-time estimates, CSV export, and a printable view
5. 05Impact and methodology pages explaining intended use, calibration, model performance, limitations, and safeguards

• Current public intake and outcome feeds are distinguished from the historical archive following Austin's May 2025 source-system change.
• Inferred open stays are not presented as confirmation that an animal is available for adoption.
• The current model estimates adoption within 30 days and stays beyond 60 days using intake-time information.
• Model documentation reports testing on a later held-out period, calibration, subgroup performance, and model lineage.
• The outreach planner distinguishes observed evidence from assumptions; projected intervention benefits are not presented as proven results.
• Predictions support photography, profile improvements, foster recruitment, outreach, and care review. They must never justify reduced care, intake refusal, or euthanasia.
• SecondChance began as a collaborative UNC Charlotte predictive analytics project, with Davis Higgins responsible for the website and presentation.
• The platform is a research prototype built from public data and is not affiliated with Austin Animal Center.