Conversational intelligence and call transcript analysis are tools that help businesses better understand and improve how people communicate. Conversational intelligence uses technologies like natural language processing (NLP) and sentiment analysis to not just focus on what’s said, but also how it’s said—looking at tone, context, and emotions.
Call transcript analysis is commonly used in customer service, sales, and healthcare to review conversations and spot trends, patterns, and areas for improvement. By using these tools, companies can enhance customer experiences, boost sales, and improve team performance, leading to more effective and meaningful interactions.
In this guide, we'll walk you through how to using LLMs in Refuel's platform to analyze large volumes of call transcripts, specifically focusing on identifying key topics and extracting relevant data points. This process is useful for businesses that need to derive insights from conversations, such as sales calls. Let's dive into the steps:
To get started, upload your anonymized call transcript dataset into Refuel. In this demo, we are using a dataset of transcripts of our own sales calls as an example.
Once your dataset is uploaded, the next step is to create a task. The steps are as such:
1. Task Name: Start by giving your task a name, such as “Transcript Analysis.”
2. Select a Model: Choose your preferred model for analysis. Refuel allows you to pick from several models, including Refuel models, OpenAI models, Anthropic models etc.
For this demo, we are creating two fields:
Fields are the questions you want the model to answer based on your dataset. For instance, you can ask:
Topic Identification: Identify the key topics discussed during the calls, such as data quality, security, pricing, or competitors.
Product Features: Extract the product features mentioned by customers or discussed during the call.
The model will analyze each transcript and output the relevant topics and product features.
Once the task is set up, Refuel will begin processing the transcripts. The output will show both the topics and customer questions, exactly as specified earlier:
You can view these outputs in a table format, where each record corresponds to a transcript. The topics and product features are displayed alongside confidence scores indicating how certain the model is in its results.
While the model works effectively, it’s important to review the outputs for accuracy. If the topics or product features listed are incorrect, you can:
Correct the Results: Edit the output by providing the correct answer and a brief explanation.
Give Feedback: Thumbs up or thumbs down can be used to provide feedback to Refuel. This helps the system learn and improve over time.
If you need more specific insights, you can create additional fields. For example, if you want to analyze data volume discussed in the sales calls, follow these steps:
1. Create a New Field: Name the field “Data Volumes.”
2. Define the Extraction: In the task settings, instruct the model to extract specific data volumes mentioned during the call.
3. Set Output Format: Specify that the output should be in JSON format.
For instance, the JSON output might look like:
Data Volume Discussed: True or False
Exact Volume: The number of records or data points mentioned
Once you've added the new field, Refuel will process the data and provide the required output. For instance, if a customer mentioned requesting 70 million records, the model would extract this number.
Finally, if you need to run this analysis on a large volume of data, you can choose to fine-tune / customize a smaller model for faster, lower latency and cheaper processing. You can do this in Refuel with only 3 clicks
After ensuring the output meets your expectations, you can deploy the task:
Deployment: Make the task live so it can handle more incoming transcripts.
Real-Time Analysis: Once deployed, the system will process new transcripts automatically, allowing you to get ongoing insights from your calls.
With the workflow live, you can start analyzing your call data. Refuel will continue to provide insights on topics, product features, data volumes, and any other custom questions you've set up. This process will help you make informed business decisions based on your sales calls.
If you have further questions or need assistance, feel free to reach out to the Refuel team!
