What player and monitoring system are you using to judge the export?
To eliminate as many variables as possible, try this:
- Bounce the Master bus to a new track (I usually name this Master Bounce)
- Set the output of that track direct to Main Outs so it's not re-effected by the Master bus, but plays in parallel with it.
- Solo the bounced track, and group the solo button with the unsoloed Master bus.
- Play the project and A/B the live mix with the bounced track.
- If they sound essentially the same, you can try ungrouping the Solo button, and inverting the phase of the bounced track to hear how well it nulls with the live mix. Depending on what plugins you're using, and whether their processing has any randomness, you may or may not get complete nulling, but it should be close.