The situation
A Fortune Global 500 automotive manufacturer collects real-time data from a connected-vehicle fleet of 1.8M vehicles across North America, each continuously streaming telematics and diagnostic data. While the data is being captured, it isn’t intelligently combined and evaluated to detect low-speed collisions.
Without that signal, the dealer network has no way to reach a customer in the small window after a fender bender when they’re deciding where to take the car for repair. Most of the repair business goes to third-party shops, and the manufacturer never even knows an event has occurred. The lost service and parts revenue reaches nine figures per year.
The challenge
Detecting a low-speed collision from telematics data is harder than it sounds. Fender benders don’t trigger airbags or produce distinctive impact signatures. Instead, they look like routine events the data records every day: hard stops, potholes, parking maneuvers, curb strikes.
Identifying actual collisions requires multiple signals read together because no single indicator does the work. The detection also has to be accurate enough for the dealer to trust and fast enough for them to act on before the customer takes the car elsewhere.
The approach
Revel Labs started by defining a clear business outcome: detect low-speed collision events with enough precision and speed to put dealer service in front of customers before they contact a third-party shop.
Building a detector meant working through the inputs. Telematics streams, diagnostic signals, and historical service data carry only partial information on their own. Read together, however, they allow collisions to be distinguished from everyday noise. The team focused on the multi-signal combinations that surfaced collision events with enough confidence to act on.
The solution
Revel Labs designed a system in two layers.
The first is a detection layer: telematics and diagnostic data flowing through Kafka and BigQuery, with multi-signal pattern models targeting greater than 95% accuracy on low-speed collision events.
The second is an action layer that wires detection alerts into the dealer service flow. Each alert has to clear two bars: reliable enough for the dealer to act on and rapid enough to reach the customer before they decide where to take the car for repair.
Looker dashboards surface the event details service teams need: time, location, severity, and supporting sensor data. Guardrails keep alerts consistent, traceable to source data, and usable by service teams without technical interpretation.
The outcome
Once built and deployed, the system stands to provide visibility into low-speed collision events for the first time. Service teams can identify outreach opportunities in near-real time and reach customers with a dealer service offer before they take the car to a third party.
The goal is to capture the more than $100M in annual service and parts revenue that is failing to reach the dealer network. Beyond the dollars, the system has the potential to build a tighter feedback loop between connected-vehicle data and dealer service.
The ongoing opportunity
By structuring and interpreting vehicle data in new ways, the system could extend to capabilities including:
- Predictive maintenance
- Service routing and customer outreach
- Parts demand forecasting
- Improved inventory planning
What started as a single unread signal can become the foundation for turning many more into dealer service.