Call Intelligence
It all begins with a phone call. Call Intelligence is the process of capturing, analyzing, and acting on data from phone calls. Call Intelligence is the foundation of Call Tracking, Call Analytics, and Call Attribution.
The analytics and insights come from fine-grained data about the phone calls themselves. This data is captured by Call Tracking software and stored in a database. The data is then analyzed to provide insights into the performance of marketing campaigns, sales teams, and customer service.
CallStudio aims to maximise the range of fine-grained data that can be captured from phone calls using AI and ML. We fine-tune our models to different analytics tasks on different languages and dialects.
Supported measures
In the following table you can see an overview of the measures and whether they can be performed on in real-time or post-call.
| Measure | Description | Real-time | Post-call | Per-role |
|---|---|---|---|---|
| Call duration | Duration of the call | ✅ | ✅ | ✅ |
| Sentiment Score | Score sentiment between 0 and 1 | ✅ | ✅ | ✅ |
| Sentiment Label | Either positive, neutral, or negative | ✅ | ✅ | ✅ |
| Emotion | Angry, happy, or neutral | ✅ | ✅ | ✅ |
| Text summary | A full summary of the text | ✅ | ||
| Topics | Topics discussed in the conversation | ✅ | ✅ | |
| Issue type | The issue type of the call (e.g., billing) | ✅ | ||
| Resolution status | The resolution status of the call (i.e., resolved, unresolved, unknown) | ✅ | ||
| Interruption duration | Duration or % of interruptions in the call | ✅ | ✅ | ✅ |
| Agent/Client talking ratio | Ratio of agent/client talk time in the call | ✅ | ✅ | ✅ |
| Silence duration | Duration or % of silence in the call | ✅ | ✅ | ✅ |
| Talk duration | Duration or % of talk in the call | ✅ | ✅ | ✅ |
| Words / second | Average number of words per second in the call indicating talking speed | ✅ | ✅ | ✅ |
| Language | The language of the call | ✅ | ✅ | ✅ |
| Empathy score | Score empathy between 0 and 1 | ✅ | ✅ | ✅ |
| Loyalty indicators | Phrases indicating loyalty (e.g., "I will be back") | ✅ | ✅ | ✅ |
| Transcript | A full transcript of the call | ✅ | ✅ | ✅ |
Choosing the right measures
The more measures you include, the more likely to get a better understanding of your calls. However, you should be aware that some measures are more computationally expensive than others. For example, text summarization is more expensive than sentiment analysis. Therefore, you should choose the measures that are most important to you and your business. Many of these tasks include usage of LLMs that are computationally demanding. Hence, you must evaluate the trade-off between the number of measures and the latency of the system & cost. The range of measures can be adjusted at any time and so the decision is not final.
Custom measures
Our models are trained on a large dataset of phone calls. However, you might want to train your own models on your own and extend the range of measures that can be performed on your calls. CallStudio can be easily modified to suit your analytics needs. This is quite common as sometimes the models are introduced to new languages or dialects or special conditions.