When designing the Workflows to be used with Awaken CoPilot interactions, there are a few core points to consider: an Awaken Audio - Start Control is required to initiate the audio stream from the agent's machine to the transcription service, CoPilot - Intent Actions Control(s) are required to add and react to intents within the transcribed conversation, and particular Workflow and Page designs should be observed in order to successfully utilise CoPilot.

 

 

 

 

The CoPilot widget is displayed in the bottom-right corner of the agent's screen, and when expanded it overlays a frame at the right-hand side of the screen that comprises either a substantial fraction or nearly the entirety of the height of the Page. As such, care should be taken with Workflow design to ensure that content isn't hidden beneath these areas; a simple resolution might be to include an empty Sidepanel Field on the right-hand side with a specified width, so as to "reserve" this space for the CoPilot widget.

 

The agent will be able to collapse and expand the CoPilot widget to see obscured content in the majority of cases, but this may make for an inefficient agent experience. Similarly, if the Workflow Page is designed so as to exclude scrolling for the agent, then any content that is hidden by the CoPilot widget icon will never be visible.

 

 

It is possible to easily add intents to the CoPilot widget by the use of the CoPilot - Intent Actions Control, with the intents having a defined condition under which they are considered to have been completed, and optionally a Field that will be triggered in a manner similar to Update Links or .click() if the intent is completed.

 

These intents can be made visible to the agent in the CoPilot widget, or remain hidden and just used to trigger or track behaviour.

 

An intent can only be added a single time, and adding identically-named intents will not overwrite or reset an existing intent.

 

A single utterance can complete multiple matching intents, but each intent can only be completed by the content of a single utterance.

 

 

It may be desirable to know within the Workflow when intents have been completed in order to drive related processes, or to take some record for reporting purposes. While it's not possible for the Workflow itself to directly monitor or record the state of intents from a CoPilot - Intent Actions Field, it is possible to use its ability to trigger actions in order to record this data as required. For example, if the CoPilot - Intent Actions Field triggers a JavaScript - Button Field, then that Field can set the value of other Fields and trigger any subsequently-required action itself as part of a chained sequence.

 

 

When an intent has been added to the CoPilot widget via a CoPilot - Intent Actions Field, then any actions that it has configured can only be triggered if the Workflow hasn't navigated away from the Page where the intent was added, even if it subsequently returned to that Page before the intent was completed. This means that Workflow design needs to prevent navigation around these times should there be any critical behaviour within any specified actions; this may required temporarily disabling the navigation toolbar buttons, preventing any button-type Fields from proceeding without confirmation that the action(s) have completed, or simple agent guidance.

 

 

It may be the case that you want to check whether the agent or caller have discussed something that isn't desired, or in a way that isn't desired. For example, you may want to detect whether the caller has objected about the price or terms of a service, or caution the agent if they have started using more negative language.

 

These behaviours can still be detected as intents, but they are intents that you do not wish to encourage the agent to try and complete; in such a case, marking them as a hidden intent prevents the agent from seeing it in the intent list (and therefore possibly unwittingly "chasing" the intent), but still allows them to trigger an action if the intent is matched. This could be used to trigger a Toast Popup Field to guide the agent on how to remedy the encountered situation, or any other process as appropriate for the matched intent and desired outcome.

 

 

While CoPilot may most often be used as a widget to the side of a Workflow that helps assist the agent through the current process, it is also possible to use the CoPilot widget in "panel mode" such that it fills the entirety of the Intelligent Agent window for processes where the agent may not need to directly or manually interact with the Workflow itself and instead wants to use the CoPilot capability side-by-side with other system(s).

 

This transition is automatically triggered when the window frame that the CoPilot widget is in is reduced to a width of 500px or fewer, and works well with various panelised integrations such as the Genesys Cloud interaction panel usage when in its narrowest configuration.

 

It should be noted that despite the Workflow not being visible to the agent, it is still present, and can be manipulated as normal. This means that CoPilot - Intent Actions Fields can still be used to record information based on completed intents, trigger Page navigation, or trigger any Helper Functions. As such, the Workflow can be designed such that when an intent is completed it then automatically and transparently progresses through a process that then may cause an additional set to intents to be registered by navigating to a new Page.

 

 

It is possible to change between different AI models and deployments within a single Workflow by simply specifying a different Connector within a CoPilot - Intent Actions Field. The Connector is a setting within the Field configuration, and the Connector that has been encountered most recently in the Workflow by the agent will be used by the CoPilot widget's AI tab and intent processing.

 

For most cases, a single Connector will be used across the entire Workflow. However, it is also possible to use a range of different Connectors through the course of a Workflow, taking advantage of the strengths of different models at different times, or swapping between different deployments of the same model (for instance to change the "temperature" of the generation, switching between more literal or creative responses and interpretations at different phases of the Workflow).

 

 

As the CoPilot intent detection relies upon analysing each utterance individually, thought should be applied in cases where Workflow design may break information into separate utterances and how that compares to the configuration of any applicable intents. For example, if the Workflow directs the agent to ask a caller's forename, confirm the spelling, and then ask for their surname, then an intent of "Has the agent asked for the caller's full name?" cannot succeed. In such a case, it may be a case of adjusting the Workflow to instead ask for the caller's full name and then confirm spelling afterwards, or instead splitting the intent into separate forename and surname intents.