Skip to content

csv.md

Latest commit

 

History

History
105 lines (82 loc) · 3.27 KB

csv.md

File metadata and controls

105 lines (82 loc) · 3.27 KB

Using CSV/JSON Files as a Data Source

The simplest way to get data into vox. Export from any tool, drop the files into a data/ directory, and run your analysis.

Supported Formats

JSON (Recommended)

The ideal format for vox. Structure your data as an array of conversation records:

[
  {
    "id": "conv-001",
    "type": "customer_call",
    "date": "2025-12-15",
    "customer": {
      "name": "Acme Corp",
      "segment": "enterprise",
      "industry": "SaaS",
      "employee_count": 500,
      "plan": "pro",
      "tenure_months": 18
    },
    "participants": [
      {"name": "Sarah Chen", "role": "VP Product", "type": "customer"},
      {"name": "Mike Torres", "role": "CSM", "type": "internal"}
    ],
    "transcript": [
      {"speaker": "Mike Torres", "text": "How has onboarding been going for the new team members?", "timestamp": "00:02:15"},
      {"speaker": "Sarah Chen", "text": "Honestly, it's been rough. Every new person takes about two weeks to get up to speed, and we lose momentum.", "timestamp": "00:02:28"}
    ],
    "tags": ["onboarding", "team-expansion"],
    "summary": "Optional summary of the conversation"
  }
]

CSV

For simpler data (support tickets, survey responses):

id,date,customer_name,segment,type,text,sentiment
1,2025-12-15,Acme Corp,enterprise,support_ticket,"Can't figure out how to add team members. Documentation doesn't help.",negative
2,2025-12-16,StartupCo,smb,survey_response,"Love the product but wish onboarding was faster.",mixed

Markdown

For meeting notes or interview transcripts, use markdown files with YAML frontmatter:

---
type: customer_interview
date: 2025-12-15
customer: Acme Corp
segment: enterprise
participants:
  - name: Sarah Chen
    role: VP Product
---

## Key Discussion Points

**On onboarding:**
Sarah: "Every new person takes about two weeks to get up to speed. We lose momentum every time someone joins."

**On reporting:**
Sarah: "I need to show my board that we're getting value. Right now I'm manually pulling numbers from three different places."

Setup

  1. Create a data/ directory in your project root
  2. Add your data files
  3. Run any vox command — the agent will discover and use the files

Exporting from Common Tools

From Gong (manual export)

  1. Go to Calls > Select calls > Export
  2. Choose JSON or CSV format
  3. Save to data/gong-export.json

From Intercom/Zendesk

  1. Export conversations or tickets as CSV
  2. Ensure the export includes full message text, not just subjects
  3. Save to data/support-tickets.csv

From Google Sheets

  1. File > Download > CSV
  2. Save to data/survey-responses.csv

From Dovetail/Condens

  1. Export tagged data as JSON or CSV
  2. Include tags, quotes, and participant info
  3. Save to data/research-data.json

Tips

  • Include metadata. The more context per record (segment, date, customer size, plan tier), the better the segmented analysis.
  • Don't clean too much. Keep the raw language. Vox needs to see how customers actually talk, not sanitized summaries.
  • Separate sources. Use different files for different data types (sales-calls.json, support-tickets.csv, interviews.json). This helps vox triangulate across sources.
  • Update regularly. Replace files with fresh exports to keep analysis current.