Customer support transcripts

1Description

Multi-turn customer service conversations from chat, email and phone, with every turn attributed to the agent or the customer, the actions the agent took in back-office systems, intent and resolution labels and, where surveyed, the CSAT outcome. Personal details are masked. Contact-center software vendors, CX teams and builders of support assistants use it to train intent classifiers, agent-assist and summarization models, and to benchmark resolution flows.

Subsets

• Chat and email transcripts
• Call transcripts
• Intent and resolution labels
• Agent and customer turns
• CSAT outcome

2Schema

3Sample records

Sample record

{
  "conversation_id": "abcd-train-7431",
  "channel": "chat",
  "domain": "Online retail account and billing support",
  "intent": "Subscription inquiry",
  "sub_intent": "Dispute a bill",
  "customer_tier": "guest",
  "turn_count": 16,
  "agent_turns": 8,
  "customer_turns": 8,
  "turns": [
    {
      "speaker": "customer",
      "text": "hello?"
    },
    {
      "speaker": "agent",
      "text": "Hello, can I help you?"
    },
    {
      "speaker": "customer",
      "text": "I think I may have been charged for my subscription twice."
    },
    {
      "speaker": "agent",
      "text": "Sure, let's take a look. What is your full name?"
    },
    {
      "speaker": "customer",
      "text": "[customer name]"
    },
    {
      "speaker": "system",
      "text": "Account has been pulled up for [customer name]."
    },
    {
      "speaker": "agent",
      "text": "Next, I need your account ID and the order number please."
    },
    {
      "speaker": "customer",
      "text": "Order ID: [order id]"
    },
    {
      "speaker": "customer",
      "text": "Account ID: [account id]"
    },
    {
      "speaker": "system",
      "text": "Identity verification in progress ..."
    },
    {
      "speaker": "agent",
      "text": "and what membership level are you?"
    },
    {
      "speaker": "customer",
      "text": "Guest"
    },
    {
      "speaker": "system",
      "text": "Querying the system for an answer ..."
    },
    {
      "speaker": "agent",
      "text": "Alright. I looked over your account and there does appear to have been an error. What was the price you paid for the subscription?"
    },
    {
      "speaker": "customer",
      "text": "$40 is what I was expecting but it charged me that twice."
    },
    {
      "speaker": "system",
      "text": "A refund has been made for the amount of $40."
    },
    {
      "speaker": "agent",
      "text": "Alright, I have refunded the double charge to the credit card you have on file. Sorry for the double charge."
    },
    {
      "speaker": "agent",
      "text": "Is there anything else I can do for you?"
    },
    {
      "speaker": "customer",
      "text": "Thank you I appreciate it. That's all I needed."
    },
    {
      "speaker": "agent",
      "text": "No problem. You have a great day."
    }
  ],
  "agent_actions": [
    "Account has been pulled up for [customer name].",
    "Identity verification in progress ...",
    "Querying the system for an answer ...",
    "A refund has been made for the amount of $40."
  ],
  "resolution_action": "A refund has been made for the amount of $40.",
  "pii_masked": true,
  "source_url": "https://github.com/asappresearch/abcd"
}

4Uses

• Training intent and sub-intent classifiers
• Agent-assist, next-action and reply suggestion models
• Conversation summarization and after-call notes
• Benchmarking resolution paths and handle time
• Evaluating support chatbots on realistic dialogues

6Access

The full Customer support transcripts dataset (MLC-0802), records on request. Delivered as JSONL, CSV or Parquet under a commercial licence.

Request the full dataset