Know how the conversation went.
You see who ended up unhappy while you can still win them back.
Customer satisfaction measurement usually means a survey that a small share of customers answer. Conversation scoring adds a second view: a model reads settled conversations and rates them on resolution, responsiveness, sentiment and quality. Together they show both what customers said and what actually happened in threads nobody rated.
- Send a rating request and capture the reply as a score
- Every settled conversation scored on four dimensions
- Topics classified so you can see what people ask about
- Trends over time, and the topics behind them
How it works.
- 01
Request a rating
Send a rating request on a finished conversation. It goes through the same consent and window checks as any other outbound message.
- 02
Capture the reply
The customer replies with a number from one to five. The reply is parsed and stored against the conversation.
- 03
Let the model score the rest
Conversations that have been quiet for a few minutes are read and scored on resolution, responsiveness, sentiment and quality, each on a one to five scale.
- 04
Read the trend
Reporting shows scores over time and the topics behind them, so a dip has somewhere to point.
Surveys, without the survey tool
The rating request is a message in the conversation the customer is already in, so the response rate is not fighting an email inbox or a separate link.
- Sent the same way as any other message, with consent checked
- Numeric replies parsed and stored on the conversation
- Scores available in reporting alongside everything else
Scoring the conversations nobody rated
Most conversations never get a survey answer. A model reads settled threads and scores them, which turns a thin sample into full coverage.
- Resolution, responsiveness, sentiment and quality, each one to five
- Only conversations that have been settled for several minutes
- Runs on a schedule, so nobody waits for it
- Stored per conversation so a score can be traced to a thread
Topics behind the numbers
Conversations are classified by topic on a schedule, so contact volume can be read by subject instead of only by channel.
- Topic classification across incoming conversations
- Volume by topic in reporting
- A way to spot the question your knowledge base should answer
What to do with a bad score
A score is only useful if it reaches a person. Low scores and negative sentiment sit in the same workspace as the tools to act on them.
- Search and filter to find the conversation behind a score
- Escalate with a macro that tags, notes and assigns
- Segment contacts by their satisfaction score for follow-up
What it does not do.
Model scoring reads the conversation text. It cannot see what happened on a phone call or in person.
Scores cut by channel, topic and time. There is no per person breakdown.
Scores are recorded for review and do not raise an alert on their own.
Common questions
How is the survey sent?
- As a message in the conversation, the same way as any other message you send, with consent and the reply window checked first.
What scale does the rating use?
- One to five. The customer's numeric reply is parsed from the conversation and stored against it.
What is scored automatically?
- Conversations that have gone quiet are read by a model and scored on resolution, responsiveness, sentiment and quality, each on a one to five scale.
Is scoring real time?
- No. It runs on a schedule and only looks at conversations that have been settled for a few minutes, so an active thread is not scored mid flow.
Can I segment customers by satisfaction?
- Yes. Satisfaction is available as a condition when building a segment, so you can follow up with unhappy customers as a group.
Does a low score alert anyone automatically?
- No. Scores are stored and reported, and an escalation macro or a reminder is how you act on one.
Are scores comparable across agents?
- Scores are stored per conversation and compared by channel, topic and time. Agents are not scored individually.
Does the customer see the model's score?
- No. Model scoring is internal. Only the rating you explicitly ask for involves the customer at all.
Every question in one place, answered once.