Enterprise Sales Copilot: Enabling Real-Time AI Support with Automatic Information Retrieval in Live Sales Calls

Abstract

During live sales calls, customers frequently ask detailed product questions that require representatives to manually search internal databases and CRM systems. This process typically takes 25-65 seconds per query, creating awkward pauses that hurt customer experience and reduce sales efficiency. We present SalesCopilot, a real-time AI-powered assistant that eliminates this bottleneck by automatically detecting customer questions, retrieving relevant information from the product database, and displaying concise answers on the representative's dashboard in seconds. The system integrates streaming speech-to-text transcription, large language model (LLM)-based question detection, and retrieval-augmented generation (RAG) over a structured product database into a unified real-time pipeline. We demonstrate SalesCopilot on an insurance sales scenario with 50 products spanning 10 categories (2,490 FAQs, 290 coverage details, and 162 pricing tiers). In our benchmark evaluation, SalesCopilot achieves a measured mean response time of 2.8 seconds with 100% question detection rate, representing a 14xspeedup compared to manual CRM search in an internal study. The system is domain-agnostic and can be adapted to any enterprise sales domain by replacing the product database.

Figures (7)

• Figure 1: System architecture of SalesCopilot. The pipeline consists of three layers: the client (React-based dashboard with audio capture), the server pipeline (STT, conversation management, question detection, retrieval, and answer generation), and external services (Deepgram, LLM, SQLite, and ElevenLabs). Components highlighted in amber represent the core AI-driven modules that distinguish SalesCopilot from standard workflows.
• Figure 2: The SalesCopilot dashboard during a live sales call. The left panel shows the real-time conversation transcript with color-coded speaker labels (blue for sales rep, green for customer). The right panel displays AI-generated suggestion cards that appear automatically when customer questions are detected, each showing the extracted question, a concise answer with specific product data, the database source, and a confidence score.
• Figure 3: Response time comparison between manual search (internal CRM) and measured SalesCopilot latency across six question categories. Error bars indicate standard deviation. Green annotations show the speedup factor. SalesCopilot achieves 9--23$\times$ faster responses depending on question complexity.
• Figure 4: Distribution of response times for manual search (left, from internal CRM study) and SalesCopilot (right, measured). Note the different y-axis scales. Manual search exhibits high variance (5--100+ seconds) while SalesCopilot consistently responds within 2--4 seconds.
• Figure 5: Measured per-stage latency breakdown by question category. Answer generation (red) is the dominant cost, followed by knowledge retrieval (amber) and question detection (blue). LLM inference accounts for $\sim$71% of total latency.