A decision-support and automation platform that replaces a manual, twice-weekly pricing review with continuous, confidence-scored recommendations.
All built on a federated gradient-boosted decision-tree model, with the human approval gate sitting after stage four and before any write to Titan DMS.
Conformal prediction (90% coverage) produces the interval; it widens automatically when data is thin.
The score predicts whether a vehicle sells within the tenant's target day window at or near the recommended price. It does not claim the action beat an alternative; that counterfactual is never observed.
Keeping personal data outside the platform boundary lowers breach risk, simplifies tenant contracts, and strengthens the case for cross-tenant learning.
Postcode is the only location signal retained: enough to tell local from interstate, not tied to any person.
Success is measured on days-to-sell against baseline, front-end-gross retention, manager override rate, and pricing-manager time saved.