{
    "action": "final",
    "content": "## Dataset Overview\n\nThis dataset captures **call center performance metrics** from the Callyeah platform, tracking how incoming calls are handled across three channels: AI automation, human agents, and voicemail.\n\n## What a Single Row Represents\n\nThe data contains **two distinct row types** mixed in one table:\n\n### 1. Daily Call Summaries (Primary Grain)\nMost rows represent **one day's aggregate call statistics**, with columns:\n- **Day**: The date (e.g., \"01-Jul\", \"02-Jul\") or day of week (\"Mon\", \"Tue\")\n- **Total Calls**: Volume received that day\n- **AI Handled**: Calls resolved by automation\n- **Human Handled**: Calls escalated to agents\n- **Voicemail**: Calls sent to voicemail\n- **AI Resolution %**: Success rate of AI handling\n- **Human %**: Percentage requiring human intervention\n\nFor example, the row for July 1st shows:\n- 321 total calls\n- 245 handled by AI (76.40% resolution rate)\n- 63 escalated to humans\n- 13 went to voicemail\n\n### 2. Summary/Header Rows\nThe dataset also contains **non-data rows** that appear to be:\n- **Weekly rollups** (rows with Day = \"Thu\", \"Fri\", \"TOTAL\")\n- **Section headers** (empty rows, rows with \"=== DAILY TREND — LAST 30 DAYS ===\")\n- **Column label rows** (\"Date | Total Calls | AI Resolved...\")\n\nThese are artifacts from importing a formatted report rather than clean transactional data.\n\n## Data Quality Note\n\nAll numeric columns are stored as **VARCHAR (text)** rather than proper numeric types, and the unnamed columns (_unnamed_7, _unnamed_8, _unnamed_9) suggest the source file had additional columns that weren't properly labeled during ingestion. The 216 rows span what appears to be a 30-day reporting period (June 23 - July 23, 2026 based on the filename), though the exact date range requires parsing the Day column more carefully to separate actual dates from day-of-week labels and summary rows."
}