Intro to Conversation Metrics

Contents:

What are Conversation Metrics

Audio and Video conversations analyzed by Revenue.io's Conversation AI provide insight into the key behaviors driving an interaction's outcome. Below you will find key concepts and metric categories found in Conversation AI along with the types of conversations they are available for. 

 

Active listening 

When someone is talking, it is common for the listener to interject to convey agreement ("yes," "absolutely," etc.) or understanding ("got it," "uhuh," etc.). This type of interruption is detected as a "backchannel" and is a sign of engaged conversation on a call. To detect a backchannel interruption, we look at the length of the interruption, the location of the interruption in a streak, and the specific words used in the interruption. This metric is only available on calls with two speakers. This metric is presented as a count of active listening events and includes a value which is our best estimate of "what percentage of the time the other speaker was talking did you spend actively listening?"

Available on - Dual Channel Phone Calls

See Also: How to view Active Listening for a Conversation

 

Interruptions 

Interruptions are exactly what you expect them to be. If a speaker is mid-sentence when another speaker begins talking, this is classified as an interruption. Backchannels are not classified as interruptions. Interruptions are a great window into your agent's behavior on calls. If the agent interrupts at a high rate, this may imply they are not focused on listening on a call and are eager to interject and talk. If a prospect is interrupting at a high rate, this indicates the agent is not creating space for the prospect to share what is important to them.

Available on - Dual Channel Phone Calls

 

Streaks 

Streaks represent a continuous stretch of uninterrupted talking from one participant on a call. Streaks start when a call participant starts talking and can end in one of 3 ways: A silence incident, a natural change in speaker, or an interruption. Backchannel interruptions do not end a streak.

Calls with a single speaker (single-sided recording) will have streaks for both speakers computed. We have analyzed millions of calls to determine what portion of a typical call is silent. That silence falls into two categories: silence associated with long pauses and silence associated with a speaker change. We take this information, along with the length of a call and the streaks of the recorded party, to estimate the streaks of the party who is not recorded.

Available on - Single Channel Phone Calls, Dual Channel Phone Calls, and Video Calls.

 

Talk Ratios 

Talk ratios represent the amount of talk time each speaker on a call spends talking. This means that talk ratios should always sum to 100%, and silence incidents are removed from the talk ratio.

Calls with a single speaker (single-sided recording) will have talk ratios estimated for the conversation. We have analyzed millions of calls to determine what portion of a typical call is silent. This information is used to estimate what portion of the single-sided call is filled with the other participant speaking and what portion of the call is filled with the sounds of silence.

Available on - Single Channel Phone Calls, Dual Channel Phone Calls, and Video Calls.

 

Sentiments 

Sentiments represent the emotion of a sentence, -1 is very negative ("I despise the competition"), 0 is neutral ("that is fine"), and 1 is very positive ("I love these features"). Sentiments are computed on every sentence of a call and are displayed as an average value of the sentiment of an overall call.

Calls with a single speaker (single-sided recording) will only compute sentiments for the party that is being recorded.

Available on - Single Channel Phone Calls, Dual Channel Phone Calls, and Video Calls.

 

Talk Rates 

Talk rates represent the number of words spoken per minute by a speaker. This metric ignores silence in its calculation, meaning that you are looking at: (number_of_words_spoken)/(talk_time). A long silence after a rapid sentence will not impact the talk rate metric. Talk rates are computed on every sentence of a call and are rolled into an average talk rate in ConversationAI.

Calls with a single speaker (single-sided recording) will only compute talk rates for the party that is being recorded.

Available on - Single Channel Phone Calls, Dual Channel Phone Calls, and Video Calls.

 

Vocabulary

Vocabulary represents the number of unique words used per minute by a speaker. This metric computes the number of unique words in each streak and then divides the number of unique words by the talk time in that streak. The final value represented in the application shows the average vocabulary across the entire call.

Calls with a single speaker (single-sided recording) will only compute vocabulary for the party that is being recorded.

Available on - Single Channel Phone Calls, Dual Channel Phone Calls, and Video Calls.

 

Questions 

Questions represent every time a question is asked on a call. Revenue.io's conversation AI utilizes a deep learning grammar model along with a few hand-coded rules to identify questions in a conversation. Currently, this feature only works on English phone calls.

Calls with a single speaker (single-sided recording) will only detect questions for the party that is being recorded.

Available on - Single Channel Phone Calls, Dual Channel Phone Calls, and Video Calls.

 

Silence

Silence is a pause of 4 or more seconds between speakers talking. The 'between speakers talking' part means that no talking at the beginning or end of a call is not counted towards silence. Hold times are also not counted as silence. The silence percentage is the total amount of time on a call, with silence divided by the time between the first and last spoken word on the call. 

 

Keyword Detection 

Keyword detection flags points in the conversation where user-specified keywords are mentioned. Conversation AI utilizes phonetic keyword detection meaning that even if our transcription service does not transcribe a word correctly, we can still identify it as a keyword if it sounds similar enough to the keyword entered.

Calls with a single speaker (single-sided recording) will only detect keywords for the party that is being recorded.

In Conversation AI, users can enter important keywords to flag on calls. Furthermore, these keywords can be grouped by type for spotting trends on a large group of calls. For example, you can enter the name of every competitor as individual keywords and group them under "Competitors." This will give you a high-level view of how often competition generally comes up on calls while allowing you to drill down and see which competitors come up more or less often over time.

Available on - Single Channel Phone Calls, Dual Channel Phone Calls, and Video Calls.

See Also:

Was this article helpful?

/