The suggested reply simulator is a preview console that simulates the live message exchange scenario between a customer and a customer service representative (CSR), in which the CSR uses suggested replies to respond to the customer. The CSR admin follows a Launch-Test-Train pattern to train suggested replies for seeded customer utterances using this preview console. The augmented agent displays suggested replies that match the customer input in the live interaction. The CSR can then send the appropriate automated reply from the system's suggestions. For example, if the system's suggestions are not relevant to the customer's input, the CSR can type in a response and send it to the customer. Training the augmented agent by matching the customer utterances to the corresponding phrase text in App Studio reduces the CSR's effort during the live chat by suggesting the most probable replies during the interaction. The CSR admin trains the model by using the preview console in App Studio to suggest more appropriate responses to the CSR during the live chat.
The following figure shows the suggested reply simulator in App Studio.
- When the CSR admin adds customer utterances to a specific common phrase, a number appears beside the Records field, which acts as a counter that keeps incrementing as the CSR admin adds customer utterances to a common phrase.
- To update and train the model for mapping utterances to a specific phrase, you click Build model. Once the model is updated, the system displays the message Model updated. To test whether customer utterances map to a specific phrase, you then enter the customer utterance in the Customer says field, and then click Test.
Every time the CSR admin clicks the Use this reply button, the percentage (which is the propensity) beside the common phrase increases. This way, the CSR admin trains the model to give more accurate responses for the CSR to suggest to the customer during the live chat.
Showing and hiding analysis
When a customer enters an utterance, the system suggests one or more suggested replies associated with the common phrase appearing in the customer utterance. The suggested reply simulator provides an option to analyze the associated common phrase's training data, topic, and propensity, when the system displays suggested replies for a customer utterance.
To hide the analysis, click More actions, and then click Hide analysis.
Testing untrained customer utterances
Untrained customer utterances are new customer utterances about which the model is unaware. The CSR admin must train the model to map those untrained customer inputs with phrases or suggested replies.
When a new customer utterance is not present in the model, the system displays the message No suggested reply.
The CSR admin can then select a new reply by clicking Select a new reply in the Suggested reply simulator section of the..... The system fetches and displays a list of common phrases and their phrase text. The CSR admin can then select the more suitable phrase for the new customer utterance and click Submit.
Upon submitting, the CSR admin can see a number beside the Records field, which means that the CSR admin must update the model by clicking on the Build model button. Once the CSR admin updates the model, the system displays Model updated.
When the model is updated, the system increments the customer utterances count for a common phrase under the Training data field.