Most tow software calls everything “AI.” We don’t. Deterministic rules-based engines handle lien deadlines, fee math, and dispatch matching — same input, same output, every time. ML models handle the work that genuinely requires judgment: reading a VIN off a glare-streaked photo, predicting an auction hammer price, skip-tracing an owner.
If a vehicle has been in the yard for 30 days in California, the lien-sale notice goes out. Period. There is no model, no confidence score, no “the AI thinks.” This is the work that should never need a human decision — or a probabilistic one.
Notice timing, mailing requirements, sale-eligibility windows, and required forms encoded for all 50 states. Cron-driven; runs every morning against every vehicle.
Weighted scoring with hours-of-service guardrails. Heavy-duty calls never go to a flatbed. Won't dispatch into an HOS violation. Re-balances mid-shift as ETAs drift.
Daily, hourly, indoor, outdoor, after-hours, holiday rate rules. State-by-state max-cap enforcement. Idempotent ledger writes mean no double-charging on retries.
Cron-driven date evaluators across vehicle and staff records. Flags expiring CDL endorsements, lapsed insurance, overdue DOT inspections, and equipment recalls.
Real-time clock against the priority SLA matrix. Fires 75% warnings and post-breach escalations to supervisors. Police-rotation order tracked separately.
Configurable grace periods, flat or %-based fees, jurisdiction-specific caps. Inspects existing items to prevent double-charging. Same logic, every cycle.
Reading a VIN off a glare-streaked dashboard photo. Predicting an auction hammer price. Skip-tracing an owner across DMV records. These are problems where deterministic rules fail and ML genuinely helps. Here's exactly which models do what, and what their confidence thresholds are.
Reads VINs off dashboards, door jambs, and engine bays. Reads plates from any angle. Handles glare, rain, low light, and partial-view shots taken from a driver's phone.
Labels each photo with pre-existing vs. tow-related damage, severity, and location — with a defensible audit trail. Surfaces damage at intake, before release, so disputes don't happen.
Predicts hammer price by year, make, model, condition, and region. Suggests reserves. Flags vehicles likely to redeem before sale — saving wasted notice cost on inventory that will never auction.
Cross-references plate, DMV records, insurance hits, and prior-address data to locate elusive owners. AI drafts certified mailers in the right legal format for each state — a human reviews every one before it goes out.
Ask: “How many vehicles are 30+ days in the yard?” or “Revenue by tow class last quarter.” Get a narrative + chart + sortable table. Chat with your live operational data.
Reads CAD descriptions and suggests tow class, equipment, and priority before a dispatcher touches it. High-confidence calls auto-dispatch. Low-confidence calls queue for review.
Sophisticated buyers can tell the difference between a model and a SQL query. We split the platform this way because conflating the two undermines the parts that genuinely are AI. When we say "the lien engine knows your state's rules," that's deterministic rules — and that's what you want it to be. When we say "the AI extracted the VIN from this photo," that's actually a vision model.
Rules where rules belong. AI where AI earns it. Both honestly labeled.
Explore the modules