Smart responses use an in-house Large Language Model (LLM) to suggest responses to agents. The AI is not custom trained, making it best suited for the opening and closing stages of case conversations. At any given time, multiple smart responses are suggested to the user, who can choose the most suitable response for the ongoing conversation. This approach also supports basic fine-tuning using prompt engineering to enhance relevance.

Why Smart Responses Are Useful

1. Reduced SLA/AHT

Smart Responses eliminate the need to manually search or type repetitive replies when handling customer queries. Agents can simply hover over the suggestions displayed in the middle pane of the agent console and select the most appropriate option for their response.

2. Reduced Human Error

Smart Responses consider grammatical accuracy and the context of the conversation when generating replies, thereby minimizing human errors.

3. Provides Alternate Ways to Answer Customer Queries

Smart Responses suggest three possible answers, giving agents a variety of options for their messages and making interactions feel more organic.

​Capabilities of Smart Responses

1. Smart Suggestions

Customer care agents no longer need to search for the desired script. The Smart Response feature reads the ongoing conversation with the customer and offers agents the top three responses to reply with. Agents can then use the response directly or edit it as needed.

​

2. Accuracy Rate

Based on past responses by your agents, AI learns, writes, and recommends responses. The expected accuracy of the Smart Responses model is ~70%, meaning at least one of the suggested responses can be used by the agent as it is or with minor modifications.

​

3. Agent Name Insertion

Once an agent selects a smart response, their signature will automatically be inserted at the end of the message.

​

4. Customer Name Insertion

In the Smart Response suggestion, the customer's name will be inserted where applicable.

• Twitter: Handle Name

• Facebook: First Name of the Customer

​

5. Feedback Support

Users can provide feedback using the thumbs up/down buttons shown alongside the responses. This feedback can be used to further improve the model post manual validation through reporting.