AI-Driven Call Analytics

Inspiration

Organizations receive thousands of customer calls every day, yet the valuable insights hidden within these conversations often go unused. Support agents spend time documenting notes manually, managers struggle to identify recurring concerns, and critical follow-up actions get delayed or missed entirely.

We wanted to change that. Our goal was to build a solution that transforms unstructured voice interactions into structured, actionable business intelligence automatically, at scale, and with human oversight built in.

What We Built

Acme Call Center Analytics is an AI-driven platform that turns every customer call into a strategic business asset. It transcribes conversations, extracts sentiment and intent, detects fraud and compliance risks, seeds follow-up workflows, and surfaces everything through a live dashboard and a conversational AI agent all natively on UiPath.

How We Built It

We built this solution across six components, using Claude Code as our Coding assistant throughout helping us fast-track development, debug complex issues, and gain better context on how each part of the system works together.

Coded Agent — the intelligence layer of the pipeline. It leverages two open-source models to transcribe raw call audio and perform the initial analysis, extracting sentiment, emotions, intent, fraud signals, compliance indicators, and trigger words, outputting a structured JSON result ready for downstream processing.
Maestro BPMN — acts as the central orchestrator. It is triggered when a call audio file and its metadata are uploaded, then calls the Coded Agent to retrieve the initial analysis. The output is passed through a validation and enrichment step that cleans the data and structures it into 43 fields before routing each call across four intelligent pathways: Fraud, Compliance, Escalation, and Retention. From there, it automatically seeds follow-up actions and drives the supervisor approval workflow.
UiPath Data Fabric — serves as the system of record. All structured data produced by the JSON validator is persisted across three entities (CallRecord, CallFollowup, AiUsage), with a supervisor-driven human-in-the-loop approval workflow governing follow-up actions.
Conversational Agent (Aria) — built on UiPath Agent Builder, Aria answers plain-English questions over live Data Fabric data in real time (e.g. "Which agent had the most escalations this week?"), giving managers instant access to insights without navigating reports.
UiPath Coded App — built using React, this 17-route analytics dashboard surfaces 8 KPI tiles and 6 drill-down modules, giving operations and leadership a real-time view of call performance, compliance, sentiment, and escalation trends. Aria is integrated directly into the app, enabling conversational querying alongside the visual analytics.
UiPath Action App — provides supervisors with a dedicated interface to review AI-generated follow-up actions for each call and either approve or reject them, ensuring human oversight remains at the centre of every decision before actions are executed.

What We Learned

Successful voice analytics extends far beyond accurate transcription. The real value lies in converting conversations into decisions and actions. We gained hands-on experience orchestrating multiple AI components within UiPath, designing reliable Maestro BPMN workflows, enriching unstructured outputs into structured schemas, and balancing full automation with human trust through supervisor approval workflows.

Challenges We Faced

Data enrichment complexity — mapping raw AI output into 43 structured Data Fabric fields required careful schema design and a multi-step validation pipeline.
Accurate intent and sentiment detection — reducing false positives across diverse call scenarios required iterative prompt refinement and testing against real transcripts.
Human-in-the-loop design — balancing automation speed with supervisor oversight for follow-up approvals without creating operational bottlenecks.
End-to-end integration — connecting the Coded Agent, Maestro BPMN, Data Fabric, and conversational agent into a single cohesive workflow while keeping each component independently testable.

Impact

The platform enables organizations to:

Reduce inbound call volumes by surfacing intent-driven self-service opportunities
Improve customer experience with faster resolutions and proactive, automated follow-ups
Increase operational efficiency by eliminating manual note-taking and after-call work
Enforce compliance at scale with automated flagging and audit trails on every call
Empower leadership with real-time KPIs, drill-down analytics, and natural-language querying via Aria

Every customer conversation becomes an opportunity for insight, efficiency, and improved decision-making transforming voice data from a passive record into a live strategic asset.