Conversational AI Cloud logging data is made available via:
The dashboards provide various aggregated overviews of some or all interaction types. Tables and/or charts on the dashboards may allow drilling down to other logged features of interactions, and the (aggregated) data underlying the visualization can be downloaded from the dashboards and used for further analysis if desired.
By default, reports in Conversational AI Cloud’s dashboards show data from interactions of active users in the production environment.
This is the same for downloads of interaction logs via the reporting API.
Reporting Features: Interactions
- Interaction logs only include information about articles, FAQClicks, FAQSearches, dialogs, feedback and LinkClick interactions by default. An interaction log containing Events is a different, more specialized export. Data retrievals are not available in the Interaction logs
Interaction logs that do include Events will also include Events from Inactive Sessions.
Some Dashboards show an Interaction %. This is intended to be an indication of how many visitors to your website or users of your app actively engage with Conversational AI Cloud.
Feedback and Linkclicks are logged as separate interactions, but they also have a reference to the Interaction that they originate with. Feedback in particular may occur multiple times on a single original interaction. The Dashboards, by default, count the originating interactions of Feedback and Link Clicks rather than the Feedback/LinkClick interactions themselves.
DataRetrieval interactions are not shown as interactions in most Dashboards. They are used to determine the Interaction % overall, and to determine how often FAQs were shown in Dashboards where that is a metric.
- One interaction can activate another. These interactions will have the same interactionId (a GUID) in the logs, but a different logId. In the Dashboards, aggregate interaction counts are determined by counting unique interactionIds. If no grouping is made on Interactiontype, these multiple interactions will therefore be counted as one. But with grouping on Interactiontype, if an article activates an Event, this interaction will be counted both as an article and as an event.
Reporting Features: Events
The logic of the Conversational AI Cloud platform supports activating Events because of other interactions. The activation of these Events can take place either before or after the original interaction has been fully processed.
- Events that are activated before the interaction has been fully processed can be used to 'catch' article questions containing very high-importance terms to ensure that these get a consistent response regardless of any other possible recognition and preventing the need to add much extra recognition in this area.
The interaction that caused an 'Exception' Event to activate will be logged as the original interaction(type), but it will not have its own response. Any log data that would result from Conversational AI Cloud processing the interaction are also not available for this interaction. Instead, the response is logged in the Event interaction and further logging data will be for the Event.
Events can also be activated after an interaction has been processed, to intercept the response of the original interaction and substitute the response of the Event.
- Escalation and Exception Events can be activated from any interaction of any type (even other Events), but are most commonly activated because of article (Q&A) interactions.
When a session with an unfinished T-Dialog times out, the Conversational AI Cloud engine generates an OnSessionExpired Event– in a new session.
Reporting Features: autoDialogs
In the Interaction log, an end of an autoDialog ID is actually a QAID not a node ID of a Dialog.
Aritcle IDs on other autoDialog interactions are shown as 'ad-' this is because they do not actually exist in Conversational AI Cloud but are generated at the point an autoDialog is triggered. autoDialogs are not stored anywhere, so you will not find them in the CMS.
Reporting Features: Dialogs
- Conversational AI Cloud response to a article (Q&A) interaction can be to start a dialog. Because we log every step in a dialog as its own interaction, this means that in the logs, you will see a article (Q&A) interaction with the 'dialog' answer, and then a *second* interaction, which is of the type dialog, with the same answer.
- Articles (Q&A), FAQClicks and Events can start dialogs.
One dialog can also start another. This will be logged as an end step in one dialog and a start step in another.
Reporting Features: T-Dialogs
- Just as with regular Dialogs, Conversational AI Cloud's response to an article (Q&A) interaction can be to start a T-Dialog. The response from DigitalCX will be on the T-Dialog.
- A T-Dialog can be activated by articles, FAQClicks, Events or regular Dialogs.
The End slot of a T-Dialog can activate an Event. This Event can/should be used to report the result of handling the transaction of the T-Dialog.
When a session with an unfinished T-Dialog times out, the Conversational AI Cloud engine generates an OnSessionExpired Event–in a new session.
Reporting Features: DataRetrievals
Any time FAQs are requested to be shown together with responses to articles/events/FAQClicks/dialogs where these FAQs are chosen from Categories that are similar to the category of the response given, this is an automatic DataRetrieval initiated by Conversational AI Cloud.
DataRetrievals are usually not shown in the Interaction logs. Most Dashboards do not explicitly display DataRetrievals either, but they are used to calculate Interaction% and to determine which FAQs were shown