SOLUTIONS · AUTOMOTIVE · DEALER DATA
Map every dealer's op codes, repair-order lines and part descriptions to your standard categories
Every dealership writes its service data its own way. One store's 01FZ is another's LOF-SYN, a repair-order line says "CUST STATES BRAKES GRINDING FRONT", and a parts line reads "PAD KIT FRT". Before the data is worth anything to an OEM, an analyst or a marketing program, someone has to map it to one standard set of categories, dealer by dealer, line by line. Most teams do it with lookup tables, rules and analysts, and the long tail never ends. StayCharted learns the mapping from the lines your team has already done, maps the new ones with a confidence on every answer, and routes uncertain answers to an analyst.
- Learned from your own mappings. Years of mapped op codes and lines are the training data. No rules to write.
- A confidence on every line. Choose a review cutoff and check performance before sending answers into your data product.
- Improve with reviewed corrections. What your analysts fix in the Review Queue trains the next version.
Where it fits
| Data | What arrives | What the model returns |
|---|---|---|
| Op codes | A dealer's op code and its description: 01FZ LOF 5W20 SYN BLEND |
Your standard service category: Oil and filter change |
| Repair-order lines | Complaint, cause and correction text written by advisors and technicians | Service category, system or component |
| Part descriptions | Dealer and supplier wording: PAD KIT FRT, STRTR MTR REMAN |
Your parts taxonomy: Brakes > Pads; Electrical > Starter motors |
| Pay type and work type | The line's text and labor details | Customer pay, warranty, internal, recall, maintenance or repair |
| Warranty and goodwill claims | Claim narratives | The component or failure type your quality team uses |
A model can read several columns at once: the op code, its description, the complaint line and the vehicle's make, for example. Use whatever your analysts look at when they map a line by hand.
How it works
- Export lines your team has already mapped. One row per op code or repair-order line, with the columns you map from and the category you gave it. Remove customer names, addresses, phone numbers and VINs first; the model doesn't need them. StayCharted also finds and masks email addresses, phone numbers and card numbers before training, but does not reliably remove names, addresses or VINs; review the export before uploading.
- Train and check the report. StayCharted tests the model on lines it never saw and shows how often it's right for each category, and which categories it confuses. Set the confidence cutoff where the answers above it are right often enough to go straight through.
- Map new lines. Fill a file of a new dealer's op codes, or call the API from your pipeline for each record. Every line comes back with its category, a confidence and whether it needs review.
- Review the rest. Lines below the cutoff wait in the Review Queue. Your analysts' answers become training examples, and you publish a new version when its report is better.
Test on a representative sample from a new dealer or market before relying on automatic routing. Review a sample of confident answers as well as the cases below your cutoff.
For data companies
- Volume: consolidators map millions of lines a month. Enterprise limits are set with you, and the Partner API lets you build StayCharted into your own data platform for API-managed customers, with separate models for each client or program.
- Your data agreements: StayCharted processes only the data you send it, under our DPA. This workflow uses your exports or API submissions; StayCharted does not obtain dealer records from a DMS on your behalf. On Business, models can be trained in your own Amazon Bedrock or Microsoft Foundry account.
- One model per taxonomy: keep separate models for op codes, parts and warranty, or per OEM program, each with its own report.
Common questions
We already have lookup tables. Why a model?
Lookup tables handle the op codes you've seen. The model handles the ones you haven't: a new dealer, a new abbreviation, a free-text line. Keep the table for exact matches and send the rest to the model.
How many mapped lines do we need?
A few dozen per category is a start, and more for categories that look alike. The report shows which categories need more.
Can it match part numbers or decode VINs?
No. Exact part-number matching, supersessions and VIN decoding are reference-data jobs. StayCharted classifies the text around them: the description, the complaint, the correction. Decoded vehicle details can be extra columns the model reads.
Does it work in other languages?
Use a multilingual AI model and representative labeled examples for the languages you need. Validate performance separately for each language and market before combining them in one workflow.
Which plans?
Essentials and Business include the API and the Review Queue. For consolidator volumes and the Partner API, talk to us.
Put it into practice
Vehicle quality and complaint classification →