Timeline
10 weeks
Team
4 engineers + 1 PM
Client
E-commerce company
The Problem
Customer service representatives struggled to pull up customer data (order history, support tickets, account details) while actively speaking with customers. Every lookup pause damaged the conversation flow and frustrated callers.
Our Approach
Call Flow Analysis (Weeks 1 to 2)
Analysed 500+ recorded calls to map the context lookups CSRs performed most frequently and the pauses that hurt customer experience.
Real-Time Transcription & Intent (Weeks 3 to 5)
Built a streaming transcription pipeline with sub-second latency and an intent classifier that fires context fetches before the CSR even asks.
Context Surfacing UX (Weeks 6 to 8)
Designed the agent UI to surface information proactively without becoming noisy. Prioritization logic ranks what to show based on conversation context and customer history.
CRM/ERP Integration & Rollout (Weeks 9 to 10)
Connected the co-pilot to Salesforce and SAP. Ran supervised pilots with 20 CSRs before rolling out to the full 200-agent contact center.
The Solution
We built a silent AI co-pilot that listens to live conversations, interprets intent, and instantly surfaces relevant customer context directly to the CSR. It integrates with CRM and ERP systems to retrieve order status, account details, and support history, without the CSR ever leaving the call flow.
Why This Tech Stack
FastAPI with Uvicorn gave us the async performance needed for sub-second transcription round-trips. Azure OpenAI's streaming responses meant the UI updated as the customer spoke, not after.
The Outcome
CSRs handled calls faster and with more confidence. First-call resolution improved, average call duration dropped, and the cognitive load on agents reduced significantly. Customer satisfaction scores moved up quarter over quarter.
Key Metrics
+31%
First-call resolution
−22%
Average handle time
+18 points
CSAT uplift
3 wks → 5 days
Agent onboarding time