Chat transcript quality review

Review support chats for resolution and agent tone without reading every transcript.

Run this on your file Download the template CSV

Columns to upload

Name the columns in your CSV like this. Extra columns are fine: every column of the row is sent with the questions.

Column	What it holds
`conversation_id`	Conversation ID
`agent`	Agent name
`transcript`	Full chat transcript

Questions

Each question adds a column to your file. The answer is one of the listed values, or Unsure when the model's confidence is below the cutoff: 70% for pick-one and scale questions, 80% for yes or no.

New column	Question	Possible answers
`resolved`Yes or no	Was the customer's problem solved by the end of the chat?	yes The customer accepts the fix or says thanks, or the agent completed the fix no The problem is still open at the end
`agent_tone`Scale	How did the agent come across?	rude or dismissive neutral friendly and clear Lowest to highest
`escalated`Yes or no	Did the agent hand the case to another team, or say someone will contact the customer later?	yes The agent handed the case on or promised a later contact no The agent handled it in the chat
`customer_repeated`Yes or no	Did the customer have to explain the same thing more than once?	yes The customer repeated themselves no The customer explained once

Example row

Column	Value
`conversation_id`	C-7781
`agent`	Ana
`transcript`	Customer: my box says delivered but it's not here. Agent: Sorry about that! I've filed a claim with FedEx and sent a replacement, arriving Friday. Customer: great, thanks.

Answers from a test run on this row

Column	Answer	Confidence
`resolved`	yes	92%
`agent_tone`	friendly and clear	100%
`escalated`	no	93%
`customer_repeated`	no	96%

More support use cases

Use case	What it does	New columns
Support ticket triage	Route a backlog of support tickets by category and urgency, and flag the angry customers and refund requests.	`category` `urgency` `angry` `refund`
Customer satisfaction comments	Find out whether low satisfaction scores are about the agent, the product or the policy.	`about` `sentiment` `agent_praised`
Bug report labeling	Label incoming bug reports by product area and severity, and send back the ones with no steps to reproduce.	`area` `severity` `repro_steps` `money_or_data`
Help center gaps	Match customer questions to your help articles and list the questions no article answers.	`article_answers` `self_serve` `question_type`
Cancellation reasons	Code why subscribers cancel and find the ones a pause or a discount could keep.	`reason` `pause_would_help` `price_sensitive`