Contextual Model

Updated 

The Contextual Model, also known as the LLM Model, analyzes entire case conversations to provide insights and relevant messages for defined quality parameters. The scoring logic is configured to assign scores based on the model output.

Compared to the Classification Model, the Contextual Model offers greater accuracy by analyzing the entire context of the case rather than focusing solely on message or phrase-level analysis.

By comprehensively analyzing entire case conversations, the LLM Model enhances user experience by delivering more accurate and contextually relevant insights.

Supported Languages

The model supports English, Spanish, and German languages for both digital and voice cases.

Example of Contextual Model Scoring on Sprinklr Platform

In the image provided, the Quality Management (QM) model has identified one of the agent messages in the conversation as depicting Courtesy. The model provides an insight statement explaining the reason the agent needs to be courteous, and the relevant messages are highlighted accordingly. As Courtesy is considered a positive parameter, the quality score assigned for this specific parameter is 100.

Creation of Contextual Model

Perform the following steps to create the contextual type agent quality model.

  1. Click the New Tab icon. Within the Persona Apps, select the Quality Management persona.

  2. Click on the Settings icon in the bottom left corner.

    The Governance screen is displayed.

  3. Click the Agent Quality Model.

    The Agent Quality Model screen is displayed.

    ​​

  4. Click Add Model appearing on the top right corner of the screen.

    The Select Agent Quality Model screen is displayed.

    ​​

  5. Select Contextual Quality Model from the Agent Quality Model drop down menu.

    The Overview and Model Details screen is displayed.

  6. Enter the Name and Description and under the Overview section.

  7. ​Click the Share Agent Quality Model With field, Share screen is displayed.

    1. Enter the name or email of the agents you want to give Editor or View Only permissions.

    2. Click Share.

    3. Close the Share screen.

  8. Under the Model Details section, enter the Model Question that will be used by Sprinklr AI+ for Automated Quality Management.

  9. Enter the Model Description that will be used by the Sprinklr AI+ for Automated Quality Management.

  10. Click the Include Call Recording Toggle button to enable call recording.

  11. Enter the Possible Answers as Yes or No.

  12. Click Save.

Refer to the below table for parameter description of Contextual Quality Model.

Parameter

Description

Name

Name of the agent quality model. This field is mandatory.

Description

Additional information of the agent quality model. This field is mandatory.

Share Agent Quality Model With

Set permissions for the agents for the Quality Model. The agents can be given either Editor level permissions or View Only permission.

Model Question

Ability to define and use placeholder in the question using {$..}. This field is mandatory.

Minimum Length Possible is of 10 characters and the Error Message that is displayed for this is: “The question should must be at least 10 characters long”.

Maximum Length Possible is of 5000 characters and the Error Message that is displayed for this is: “The question must not exceed 5000 characters”.

Model Description

Description of the Model and additional details for the Model. This field is mandatory.

Minimum Length Possible is of 10 characters and the Error Message that is displayed for this is: “The question description must be at least 10 characters long”.

Maximum Length Possible is of 10000 characters and the Error Message that is displayed for this is: “The question description must not exceed 10000 characters”.

Include Call Recording

Allows the system to use both audio and transcripts for evaluating quality parameters. This enables analysis of tonality-based aspects such as agent rudeness, frustration, or lack of energy. By default, this option is turned off.

Note: Audio based scoring is only supported via OpenAI, Azure openAI, Google Vertrex and xAI and is not supported via Sprinklr in-house, Amazon Bedrock and Anthropic Claude.

Possible Answers

For the Possible Answers only Yes and No value is present. You cannot add or delete any possible answers. You can only enable or disable the Hide Insights toggle button. By default it is disabled. The Hide Insights decides if the ML Insights for a model should be displayed.