Challenge
Customers personalising a vehicle can choose from countless equipment options - with many implicit dependencies. In sales this is handled with spreadsheets today: look up equipment codes, look up prices, check dependencies, maintain the lists by hand. The request was for a classic app to support this - with a search field and filters, no AI.
Our proposal
Instead of yet another interface with a search field and filters: a chat. Advisors type the codes or their question and get the answer - including price and dependencies. We worked this out as a case study and built it as a working prototype:
- Ingestion of the product documentation and price lists (PDF and spreadsheets), including OCR for information inside images.
- Knowledge base built from the product information, from which the assistant draws its answers (retrieval-augmented generation: find the relevant passages first, then answer from them).
- Assistant configuration for questions about codes, possible configurations and dependencies per model.
- Cloud infrastructure to run the prototype.
Result
- Lookups in 5 to 10 seconds instead of 3 to 5 minutes: that is how long a lookup takes via the spreadsheets the usual way, and how fast the prototype answers.
- No searching through long PDFs and tables: typing the codes or the question is enough.
- A tangible prototype as the basis for deciding between a classic app and a chat.
Case study with a working prototype, not in operation. Times: the usual way via spreadsheets and the prototype’s response time, not measured in a study.
* Prototype: built and working, not in operation
