**Data Sync Frequency: **If we're looking at using MSC's database read replicas instead of webhooks (based on Sandeep's chat with Rutger) how often should we pull the new conversation data? Will the replicas be able to manage the extra load they are going to get from the QA app? Do you have a preferred frequency in mind or should we let the account admin to set these according to their needs?
**Historical Data: **When someone sets up a new account, should we analyze their past conversations too? If yes, what's the best way to get the historical data from MSC's database?
**Quality Analysis Parameters: **What specific things should we be analyzing in conversations? And should some factors be more important than others? For example, should grammar and spelling mistakes carry the same weight when the LLM is doing its analysis?
**Analytics & Reporting: **What kind of performance dashboards are you looking at? Are we thinking along the lines of agent performance over time, maybe showing how someone's doing across different channels over the past few months.
Re-calculation of the metrics: Do we need to re-calculate the metrics of the already analyzed conversations, when a new parameter is introduced?
Default Parameters: are there any default parameters for quality analysis if the user has not set one?
When fetching for new conversation from the Conversation Table, should we rely on LastUpdatedDateTime to get the new conversations?
**Data Retention: **What's our plan for keeping conversation data? Should we follow MSC's data retention policy, or do something different? Also, what happens to the QA metrics when conversations get deleted - do we keep those or delete them? → Yes (Rutger has confirmed)
**Data Grouping: **Should users be able to slice and dice the metrics by agent, channel, or time periods? Any thoughts on showing historical comparisons too? → Yes (Rutger has confirmed)
**Subscription Setup: **When customers buy the Quality Assurance app subscription, does that create a separate order line item in their Orders object? → Yes (Rutger has confirmed)
Correctness of the Answer: How does LLM decide the correctness of an answer related to the customer? should the LLM be provided with prior knowledge? or Is there a feedback system in the MSC which LLM can make use of?