bot-analytics-pipeline

Bot Analytics Dashboard

AI-powered analytics pipeline for customer service chatbot monitoring and optimization

An end-to-end system that transforms raw customer service call data into actionable management insights. Built as a modular pipeline of 5 specialized AI agents, each handling a distinct analysis layer — from topic classification and response quality evaluation to a fully interactive executive dashboard.

Live Dashboard

🔗 View Live Dashboard—

The Pipeline

The system processes raw service call data through 5 sequential stages:

Stage 1 — Topic Classification Agent

Ingests raw call reports (Excel/CSV) and classifies every customer inquiry into predefined service categories. Outputs topic distribution with counts, percentages, and full call details per topic.

Stage 2 — QA Quality Agent

Analyzes the same report from a quality perspective. Evaluates each bot response as valid, invalid, or indeterminate based on defined criteria. Produces a conservative success rate and flags problematic responses.

Stage 3 — Merge & Insights Agent

Cross-references both analyses by call ID into a single managerial document. Adds envelope data (total service volume, bot usage rate), generates training insights, and produces an executive summary with KPI metrics.

Stage 4 — Service Branch Dashboard

Generates a separate dashboard from weekly service call data, showing branch-level performance: call volume per branch, close rates, customer counts, software version distribution, and module installations.

Stage 5 — Executive Dashboard Generator

Produces the final interactive HTML dashboard, combining all metrics into a single view: KPI cards with historical trends, topic distribution charts, funnel analysis, quality breakdown, and drill-down into every problematic conversation — with a direct link to the branch dashboard.

Dashboard Features

9 KPI cards with click-to-view weekly trend charts (pure SVG, no external libraries)
Topic distribution — bar charts and detail tables for all calls and for problematic calls separately
Funnel diagram — visual flow from bot users → agent escalation → top issue topics
Interactive conversation explorer — click any question to reveal the bot’s actual response
Historical comparison — week-over-week trend tracking across all metrics
Branch dashboard link — integrated access to service performance by branch
Fully self-contained — single HTML file, works offline, no server required

Tech Stack

Component	Technology
Analysis Agents	Python, Claude AI (custom skills)
Data Processing	openpyxl, pandas
Dashboard	HTML5, CSS3, Vanilla JavaScript
Charts	Pure inline SVG (no external chart libraries)
Hosting	GitHub Pages
Data Storage	Embedded in HTML + Excel history file

Architecture

Raw Call Report (xlsx)          Weekly Service Report (xlsx)
        │                                │
        ▼                                ▼
┌─────────────────┐            ┌──────────────────┐
│ Topic Classifier │            │ Branch Dashboard  │
│   (Agent 1)      │            │   (Agent 4)       │
└────────┬────────┘            └────────┬─────────┘
         │                              │
         ▼                              │
┌─────────────────┐                     │
│  QA Analyzer     │                     │
│   (Agent 2)      │                     │
└────────┬────────┘                     │
         │                              │
         ▼                              │
┌─────────────────┐                     │
│  Merge Agent     │                     │
│   (Agent 3)      │                     │
└────────┬────────┘                     │
         │                              │
         ▼                              ▼
┌──────────────────────────────────────────┐
│         Executive Dashboard              │
│            (Agent 5)                     │
│  ┌────────────────────────────────────┐  │
│  │ KPIs │ Charts │ Funnel │ History  │  │
│  └────────────────────────────────────┘  │
└──────────────────────────────────────────┘

Key Design Decisions

No external chart libraries — Charts are rendered as pure inline SVG. The dashboard is often opened from local file paths where CDN scripts fail silently. Pure SVG guarantees consistent rendering everywhere.

Single-file output — Each dashboard is a self-contained HTML file with all data, styles, and scripts embedded. No build step, no dependencies, no server.

Modular agents — Each analysis stage is an independent skill that can be run, tested, and improved separately. The pipeline is composable: run any subset of stages as needed.

Historical tracking — KPI values are automatically logged to an Excel history file after each run, enabling week-over-week trend analysis without manual data entry.

Screenshots

KPI cards with historical trends

Topic distribution charts

Funnel diagram

Interactive conversation explorer

Historical trends comparison

Branch service dashboard

Status

In active production use. Updated weekly with live data.

Author

Amit Rubin ERP & AI Operations Lead | QA, Automation & Product at Hashavshevet (Wizsoft)