Understanding the full spectrum of conversation intelligence data

TL;DR. Conversation intelligence tools offer unprecedented insights into business conversations. From discussion topics to sentiment analysis to conversation analytics, conversation APIs give you the insights you need to build more useful conversation applications.

Today’s conversation intelligence tools make it possible to gain unprecedented insights into human to human conversations. With modern conversation APIs, you can gain deep insights into everyday conversations — identifying action items, discussion topics, questions, follow-up tasks, and more.

But that’s only the beginning. Conversation intelligence tools also provide granular analytics on business interactions, including word-level transcriptions and timestamps. In this post, we’ll look at the wide range of data that you can gain from conversation intelligence — and how you can use those insights in your conversation-driven applications.

Understanding human to human conversations

Conversation intelligence provides a genuine understanding of human to human conversations by analyzing everything that speakers say and placing it in the context of what has already been said in the conversation — whether you’re analyzing conversations in real time or making asynchronous requests.

Here are some of the ways conversation intelligence provides insights that help us better understand human to human conversations.

Contextual insights

Insights on conversations extracted from conferences, sales meetings, and support calls, conversation intelligence make it possible to know which conversations are helping the company achieve success. Conversation intelligence is capable of identifying and extracting a variety of contextual insights.

Topics. The most important keywords and phrases which are the key drivers of the conversation.

Action items. Specific outcomes in the conversation that require participants to take a specific follow-up action like sending out an agenda for a follow-up meeting.

Questions. Explicit requests for information that come up during the conversation, whether answered or not.

Tracking domain-specific insights

Contextual insights may often times be domain agnostic and looks at the context of conversations in a general sense. For applications and business looking to track and identify insights associated with unique use cases, Custom Trackers will be a more accurate solution. Custom Trackers pinpoint insights that are specific and unique to your business and users, not only based on explicit keywords, but also contextually similar phrases and sentences.

Suppose your company prides itself on incredible customer service. You might use a conversation API to analyze customer support calls. Custom Trackers can help you find out which agents are handling calls successfully and which ones need more training, by tracking contextual patterns and keywords that may indicate customer dissatisfaction and churn risks.

By training your API on common phrases that reflect frustration, like “let me speak to your supervisor,” you could easily screen for conversations that were running off the rails. And by training your API with the kind of phrases customers like to hear from support agents, like “let me see what I can do” or “thank you for your patience,” you can also create a tracker to determine which agents are consistently delivering positive experiences.

The possibilities are limitless:

Tracking the topics that are discussed in live webinars on your company website, so that visitors can quickly find the webinars that are of interest to them
Analyzing sales conversations to pinpoint when agents are moving interactions toward the close (and when they aren’t)
Filtering meeting transcripts to surface discussions where a particular project was discussed

Speaker sentiments

While conversation intelligence excels at understanding what people are saying, sentiment analysis empowers you to look beneath the surface of what people are saying to detect underlying affective states. With sentiment analysis, you can analyze interactions to determine whether people are expressing positive, neutral, or negative sentiments. It’s a powerful way to stay on top of employee and customer attitudes.

For example, let’s say your company is rolling out a new premium version of your solution that offers more advanced features at a higher price point. As lots of new customers talk with your sales team about signing up for your new offer, you’d initially notice an overwhelmingly positive reaction. But if customers find they aren’t getting good value for their money and start to downgrade or cancel their plans, you’d notice a lot more negative sentiment in your sales conversations.

Sentiment analysis lets you identify broad trends in business conversations and empowers you to adapt accordingly.

Conversation analytics

Conversation intelligence gives you much more than the power to understand what people are saying. It also gives you access to extensive conversational data including speaker time stamps, word-level transcriptions, speaker ratio, and much more.

Consider sales calls. You can get additional insights into your sales team’s performance by looking at the number of words per minute your sales reps are saying. If they’re saying twice as many words per minute as your prospects, you know that they’re spending most of their sales calls talking, instead of really listening to the customer. And by looking at the amount of overlap between when the prospect and when the salesperson are talking, you can identify sales reps who have a habit of interrupting or talking over their customers.

The challenges of understanding human to human conversations

Natural human to human conversations are unstructured, as everyone speaks differently and uses different words and sentence structures to express the same ideas. Every conversation typically has an underlying context that gives meaning to everything that’s said over the course of the conversation.

But within the conversation, participants often bounce around from topic to topic and go off on seemingly unconnected tangents. Within a sales call, participants might start off with get-to-know-you pleasantries, then start discussing the problems a prospect is looking to solve, then move to talking about other solutions a prospect has looked at, come back to additional challenges the prospect forgot to mention initially, and so on. The discussion has a single overarching topic — a potential sale — but a large number of subtopics come up as the conversation unfolds.

Human beings understand the context of a conversation and can interpret what they hear accordingly. But it’s very challenging to build tools that can understand these conversations in context. That’s why human to machine conversations, like talking to a voice assistant or a customer service phone tree, must be very rigidly structured in order for the machine to make sense of what’s said.

In these interactions, the machine typically serves up one of a range of highly specific, predefined prompts. And if the human’s response does not meet the machine’s expected criteria, the machine will prompt the human again. If the machine asks the human, “what’s your ZIP code?” the response will be rejected if it’s not a five digit number.

Human to machine interactions are designed to prompt human responses that machines can easily interpret. But when analyzing human to human interactions, machines need a great deal of contextual information in order to generate real conversation insights.

The structure of conversation data

The key to these insights is the structure of conversation data. Symbl.ai’s APIs produce conversation insights using a JSON structure — a simple, text-based format designed for storing structured data. With JSON, you get in-depth information on everything that happens in the conversation.

Here’s what JSON looks like in practice:

In this example, the input is a speaker saying, “Hello world.” The JSON output provides extensive data on the conversation sample, including the speaker name, a transcription of what was said, a confidence estimate for the transcription, and word-level data including timestamps indicating when each word was said.

Structured data offers powerful capabilities. You can compare expected conversation topics with the topics speakers are actually talking about by examining word-level data, surface specific conversations using the message ID, filter conversations based on the topics they address, and much more.

Human to human conversations are complicated. But with Symbl.ai’s conversation intelligence APIs, you can get a sophisticated contextual understanding of human to human conversations that yields action items, discussion topics, and speaker sentiments in sentence and topic level — whether you’re analyzing voice or text conversations. And since Symbl.ai’s APIs can be easily integrated with systems like Twilio, JIRA, and HubSpot, you can always get the exact insights you need asynchronously or in real-time depending on your use case.