Comparison report in SpeechSense

The Comparison report allows you to display several parameters in a single chart and see how they correlate. For example, you can create a report that shows how the flaws in the agent’s speech, e.g., hesitation, poor articulation, or silence, affect the decision not to buy a service or product. You can use system tags as parameters for such a report: Refused to buy, Agent’s hesitation, Stop words, Filler words, etc.

How to build a reportHow to build a report

To generate a Comparison report, specify these settings:

• Evaluation parameters: Report will show changes in these parameters’ values.
• Filters: Apply to dialogs in the report.

With the settings configured, you can now build a report. It will show the values of several evaluation parameters in a chart and table.

Evaluation parametersEvaluation parameters

Evaluation parameters are those parameters whose value changes you can view in the report. Only numerical parameters are considered. If you select a tag, its value will be the number of times the tag was assigned to dialogs.

The report considers the total, average, minimum, or maximum value of the evaluation parameter for the selected period. For example, such values may be useful in the following cases:

• Total: How many times the customer asked to call a supervisor during the conversation.
• Average: Average dialog duration.
• Minimum or maximum: To determine the agent with the least or most violations.

In the report, you can select one of these evaluation parameter types:

• Speech statistics (only for audio): Agent and customer speech quality criteria, e.g., speech rate, mutual interruptions, etc.

• Tags: Classifiers applied to conversation audio recognition results or text chat messages. SpeechSense detects certain keywords, phrases, or intonations in a dialog, classifies the dialog, and marks it with a tag.

  SpeechSense has preconfigured tags. These can give you a clue as to whether there was an informal greeting or goodbye, whether the agent thanked the customer for waiting, whether it was the customer's repeat call to support, etc. You can learn more about tags here.

Filtering in the reportFiltering in the report

You can use filtering to select the dialogs to include in your report. To do this, use the following settings:

• Period: Time period of the report.

• Filters: The following filter types are available:

  • Agent: Agent data.
  • Customer: Customer data.
  • Bot (only for chats): Bot data.
  • Speech statistics (only for audio): Agent and customer speech quality criteria, e.g., speech rate, mutual interruptions, etc.
  • General metadata: Data about the conversation audio (collected via PBX) or text chat. Metadata is uploaded to SpeechSense together with the conversation audio or text chat and contains its key characteristics, e.g., date, topic, and dialog language.
  • Tags: Classifiers applied to conversation audio recognition results or text chat messages.
  • YandexGPT analysis: Agent’s performance criteria and customer’s behavioral characteristics during the dialog, such as whether the agent was polite, whether the customer acted in a rude manner, etc.

  You can also use neuroparameters in filters.

• Grouping: Choose how to group data in your report. You can only group by metadata fields, for example:

  • By agent, to analyze the performance of each one.
  • By product, to learn which products agents make fewer mistakes presenting in dialogs.

Data cross-sections depend on the dialog metadata. For example, if you want to filter or group data by product, make sure there is a relevant field in the metadata file. If you need a new set of metadata, prepare dialog recordings or chats with relevant metadata and upload these recordings or chats.

Visualizing and using the report dataVisualizing and using the report data

The report provides quantitative agent performance characteristics. You can view the report in the SpeechSense web interface in chart and table form or download it in CSV format.

The available Comparison report formats include:

• Chart: Allows you to visualize the evaluation parameter values at different points in time and their correlation. Each parameter has its own scale of measurement. To get detailed information about single parameter values, you can click the parameter on the legend, and the chart will display only this parameter. Clicking the legend again will bring all the parameters back to the chart.

  On the chart, you can set the report generation period as well as the data detail level. As a result, you can get values for different time intervals: from an hour to a quarter.

• Table: Shows the numerical value of the evaluation parameter for the specified period. The values are broken down by the customized data grouping.

• CSV file: Contains the same table as in the SpeechSense web interface. Use the CSV format to save the report locally.