I was also curious about why the exported data from an Email Performance Report would over-report key metrics for the identical report on the Marketo UI...Kenny's response makes a lot of sense and, I think, explains things a little better. it would be awesome if we had detailed logic about how this works so we could replicate the logic in the extract/export and then have confidence we could match what's shown on the UI...
But from the general logic, I think we can infer a few more detailed logical filters. Here's how I would summarize it:
First of all, it seems that the Email Performance Report UI differs from the exported version of the report in that the UI reports the most current/relevant information related the the LeadID for that email.
So for example, if there was a soft bounce but then subsequently the email was delivered, it seems that the UI will report 1 soft bounce and 0 delivery at first but then CHANGE the metric to 0 soft bounce and 1 delivery. While the export seems to track (via time-stamp) both activities, thus the count would higher showing 1 soft bounce and 1 delivery for the same LeadID for that email.
I think there must a similar logic for sends to the same LeadID and Email Address where only one is counted. There may be multiple sends because of soft bounces of course!! So whereas the export may show 3 sends because there were 2 soft bounces at first and 0 delivered, eventually the UI would simply display 1 send, 1 delivered, 0 soft bounce, 0 hard bounces whereas the export would show 3 sends at three different times, 2 soft bounces at two different times, and one delivered with its own time-stamp.
I wonder too if the export under-reports sends in some cases, for example, if the recipient's email client has images turned off, the tracking pixel that tells the server the email was opened will not fire. However, if the recipient clicks on a link in that email, the UI report, if I'm understanding Kenny's response correctly, infers an Open as well whereas the export of the report may not have an actual Open activity time-stamp only a click.
Does that make sense?