A European automotive group runs Ticket Analyzer in production at its global aftersales operation. 22M+ diagnostic-information-system tickets ingested on a daily Athena → S3 → embedding → LLM pipeline. Each ticket is summarized, labeled with semantic-intent tags, and embedded as a similarity vector. A custom application at the dealer technician's workstation queries the open ticket and returns the closest prior resolutions, ranked by similarity.
Mechanics fix issues 25% cheaper. The importer / OEM saves more than €15M per year in unnecessary parts swaps and repeat workshop visits.
What the workshop-triage deployment is, and what it isn't
The customer's aftersales program is the umbrella for the data flowing between dealerships, workshops, and the OEM's central engineering on every diagnostic event the worldwide service network records. This is the third use case under that umbrella: workshop ticket triage, where a technician at any dealer in the global network needs to find the closest prior resolution to whatever DTC (diagnostic trouble code) pattern the vehicle in front of them is producing.
Scale is what distinguishes this deployment from other Ticket Analyzer surfaces. 22M+ tickets ingested, growing toward 50M+. The technician's question becomes a similarity query: find resolved tickets with the closest signature, filtered by vehicle configuration and DTC family.
Three Ticket Analyzer surfaces, three audiences
The customer runs three Ticket Analyzer surfaces on the same product codebase, serving three different audiences with three different value paths:
| Surface | Audience | Volume |
|---|---|---|
| Editor-side harness analytics (editor side) | ~50 harness editors at the customer's HQ | Editor-time leverage, 1,000× downstream |
| North American warranty-claims surface | Warranty reviewers at a sister business unit | 256K claims processed |
| Workshop ticket triage | Dealer technicians globally | 22M+ tickets, toward 50M+ |
Same product family. Same Ticket Analyzer codebase. Three different deployment patterns. Workshop triage is the only one that reaches the technician at the workshop directly. The other two stop earlier in the data flow.
What changes for the workshop technician
Before this deployment, the workshop technician's diagnostic mode relied on personal memory of similar prior tickets, OEM-issued DTC documentation, phone calls or escalations to senior technicians or central engineering, and walking the parts catalog against the symptom signature, manually. That loop is slow at the volume each workshop processes.
- Read the DTC against memory of similar prior tickets
- Walk OEM documentation against the symptom signature
- Phone or escalate to a senior technician or central engineering
- Wait for a resolution path; vehicle stays on the lift
- Manually check the parts catalog against the symptom
- Open the diagnostic application, paste the open-ticket DTC pattern
- Ticket Analyzer surfaces closest prior resolutions, ranked by similarity
- Summaries and semantic-intent labels already attached
- Filter by vehicle configuration and DTC family
- Evaluate which surfaced candidate fits this vehicle's actual symptoms
Same shift in question shape that recurs across the SPREAD portfolio: search becomes evaluation.
What's actually live
The production tenant is running. The diagnostic application is deployed at the dealer technician's workstation. The daily ingestion pipeline runs. Weekly sync between the customer's aftersales engineering and technical leads and the SPREAD engineering team.
The outcome
- Mechanics fix issues at the workshop 25% cheaper
- Importer / OEM saves more than €15M in unnecessary parts swaps and repeat visits
The technical deployment is well past delivery: the tenant has been running, the pipeline ingests daily, technicians at the dealerships are using the application.
Program shape
| Program | Workshop ticket triage, Ticket Analyzer at the dealer technician's workstation |
|---|---|
| Volume | 22M+ diagnostic tickets ingested, growing toward 50M+ |
| Pipeline | Daily Athena → S3 → embedding → LLM ranking |
| Outcome | 25% cheaper workshop fix · >€15M saved per year per importer / OEM |
| Workflow shift | Search the documentation universe → evaluate ranked candidates |
| Sister surfaces | Editor-side harness analytics · North American warranty-claims surface |
| Status | Live production tenant; weekly sync with the customer's aftersales engineering |