Conversation Analytics Software

Stefania Solivardi

How conversation analytics software works

Conversation analytics software turns conversations into data in four stages. It captures the interaction: a call recording, a live audio stream, a chat log, an email thread. It converts voice into text through speech-to-text transcription, with speaker separation so that the customer’s words and the agent’s words stay apart. It analyzes the text. Then it indexes the results so they can be searched, filtered and aggregated across thousands of conversations.

The analysis stage is where tools differ, and it usually combines three techniques. Keyword spotting flags every conversation that contains a word or phrase defined in advance, such as “cancel my contract” or a required legal disclosure. Topic detection groups conversations by what they are about, without anyone listing the topics first. Sentiment analysis scores the tone of each conversation or segment as positive, negative or neutral.

Conversation analytics reads the content of an interaction: what was said. That separates it from contact center analytics software, which reads the records that systems write about the interaction: when it arrived, how long it waited, which queue it went through, how long it lasted. The two answer different questions. Records show that handle time rose on Tuesday; content shows that on Tuesday half the callers were asking about the same invoice error.

Conversation analytics vs. speech analytics: what is the difference?

Speech analytics is the older category, built for the phone. It transcribes voice calls and analyzes the transcripts: keywords, topics, sentiment, silence, talk-over. Everything it sees comes through a phone line.

Conversation analytics applies the same analysis to every channel where customers talk to the business. A chat transcript, an email thread and a call transcript are all text once the call is transcribed, and conversation analytics treats them as one body of conversations rather than three separate archives.

The practical consequence depends on how customers actually reach you. A customer who opens a chat about a failed payment, gives up, and calls an hour later produces two conversations about one problem. Speech analytics sees the call. Conversation analytics can see both, and count them as one issue instead of two. For an operation where almost all contact happens by phone, the difference is small and speech analytics covers most of the volume. For an operation that has moved a large share of contact to chat, analyzing only calls means analyzing a shrinking and increasingly unrepresentative part of what customers say.

Vendors use the two terms loosely, and some use them interchangeably. The criterion that holds is channel scope: ask which channels the analysis actually reads.

What conversation analytics software is used for

Conversation analytics use cases split into two groups, and the dividing question is whether you already know what you are looking for.

Searching for something defined in advance. You write the list; the software finds every match.

  • Checking that agents read a required disclosure, on every call instead of a sample, which is the core of contact center compliance monitoring
  • Finding every conversation that mentions a competitor, a cancellation or a specific product
  • Pulling all the conversations with one customer when a dispute or a complaint arrives
  • Filtering negative conversations for quality review, instead of listening at random

Discovering something nobody defined. The software tells you what is there.

  • Ranking the reasons customers get in touch by volume, week by week
  • Spotting a new topic that appears after a price change, a release or an outage
  • Identifying the requests that repeat most, as the first candidates for contact center automation through a conversational IVR or an AI voice agent

The two groups have different costs. Search needs a list that someone keeps up to date, and it only ever answers questions you already had. Discovery needs a model and enough volume for patterns to emerge, and it is the only way to find the question you did not know to ask.

Dig deeper into Conversation Analytics Software.
Discover Imagicle's solution.