López de Hoyos Street 64, 1G. 28002 Madrid, Spain
Digital Sales Agent with AI for Astara
Astara (Automotive Sector)
Generative AI, Multi-Agent Architecture & Conversational Solutions Development
Digital transformation in the automotive sector requires solutions that go beyond traditional chatbots. When we faced the challenge of creating the digital sales agent for Astara, we knew we needed to build something that truly understood the vehicle purchasing process, which is complex and has many stages. The result is a multi-agent system capable of managing the entire sales cycle: from helping a customer discover which vehicle suits them best, configuring it with colors and upholstery, generating personalized quotes, managing financing and trade-in, coordinating test drives, to closing online reservations. All this in a conversational and natural way, without the user realizing they are interacting with multiple specialized agents working in coordination.
Challenges
- Orchestrate multiple complex sales flows (vehicle configuration, financing, trade-in, test drives, reservations) in a fluid and coherent conversational experience.
- Synchronize data in real time between Salesforce, Cosmos DB and multiple external sources, keeping vehicle information, prices and availability always updated.
- Manage multichannel conversations (web, WhatsApp) with context persistence, allowing users to resume conversations at different times and devices.
- Support multiple brands (Mitsubishi, Hyundai) and countries (Spain, Peru, Switzerland) with differentiated configurations and sales processes, maintaining a single code base.
Key Success Factors
Multi-Agent Architecture with LangGraph
We implemented a specialized agent architecture using LangGraph that allows managing complex flows in a modular way. Each agent has a specific role: Product Assistant for vehicle configuration, Financial Assistant for financing simulations, Offer Assistant for quotes, Contact Dealer for coordinating with dealers, among others.
Complete integration with Salesforce ecosystem
We built a bidirectional synchronization with Salesforce that automatically extracts vehicle information, versions, colors, upholstery, accessories and points of sale. Data is processed, normalized and enriched before being stored in Cosmos DB.
RAG System with Azure AI Search
We implemented a Retrieval-Augmented Generation system that allows the agent to access technical documentation, vehicle specifications, manuals and FAQs. We use Azure AI Search to index and retrieve relevant information in real time.
Distributed state and context management
We designed a state management system that uses Redis for active sessions and Cosmos DB for long-term persistence. Each conversation maintains its complete context, including configured vehicle, user preferences, generated offers and purchase process status.
Methodology
Scalable serverless architecture
Native multichannel integration
Automated data pipeline
Complete observability and analytics
Scalable serverless architecture
Native multichannel integration
Automated data pipeline
Complete observability and analytics
Results
"The digital sales agent has completely transformed the way Astara interacts with its potential customers. Users can configure complete vehicles, obtain personalized quotes with financing options, and coordinate test drives without leaving the conversation. The system manages thousands of simultaneous conversations in multiple languages and countries, keeping all information synchronized with Salesforce in real time. The multi-agent architecture allows adding new capabilities without affecting existing ones, and the modular design facilitates the incorporation of new brands and markets. The analytics system provides complete visibility of the sales funnel, from the first interaction to the final reservation, allowing continuous optimization of conversation flows and increasing the conversion rate."