The Smart Response model supports English, Spanish, French, Portuguese, Arabic, Dutch, Italian, German, Swedish, Danish, Norwegian, Urdu, and Bengali.

The Smart Responses model uses an in-house LLM to suggest three possible responses to the latest customer message. It takes the current case conversation context and suggests relevant responses.

The global model is trained using a vast amount of synthetic customer care conversations that mimic real-world scenarios. Through this training, the model learns how agents respond to incoming customer messages. Additionally, the model is aligned to always follow certain guardrails to avoid any inappropriate response suggestions.

Yes, both the channels are catered to by different sets of models, thereby providing response suggestions that are more specific to a particular channel type.

For English: 2 weeks

For French, German, Portuguese, and Spanish: 3 weeks

Check the ​​Eligibility to Enable Smart Responses​​.​

In such a situation, the partner needs to wait until the number of cases reaches the 1K count.

The best practice is to test the model with a set of users for a few weeks in a test environment. Once found satisfactory, it can be rolled out to a set of agents in production, followed by a full rollout to all users.

The success criteria for Smart Responses are:​

​1. Reduction in Average handling time (AHT)​

​2. Adoption of Smart Responses (Number of Smart responses used by agents/Number of predicted Smart Responses).​

​Yes, Smart Responses are permission controlled.​

​No, this is not possible.​

No, there is no desired cadence for model improvement for Smart Responses, as there is minimal scope for feedback incorporation.

No, currently this is not supported.

​Yes, after adding one Smart Response, the agents can add more Smart Responses in the reply box as required.​

​Yes, after selecting the Smart Response, you can edit it as necessary before sending it to the customer.​

You can check the reporting on Smart Responses by visiting this link.

​​​​After an agent clicks and uses a smart response to publish, the reporting is generated in the system.​

​Yes, it can be done. Or, a placeholder %agent% can be added for agents to add their names if required.​

You can choose the following options:​

​1. You agents can submit the feedback using the option available with Smart Responses. Once a good number of feedback are in place or a visible pattern of correction can be identified, we can discuss the improvements.​

​2. You can share the changes you want to get incorporated.​

​3. You can also share the outbound message and the Smart Response predicted for us to analyze the relevancy of Smart Responses being predicted. After discussing the feasibility of these changes, we will make the changes.​