The system determines which user to update based on the match-on value (usually Employee ID, so we'll assume that for the rest of this example). If the datafeed processes two users with the same Employee ID, it believes that they are the same user. It would process the first row of data, updating the user with the specified name, email, etc. Then -within seconds- the datafeed would process the row with the repeated Employee ID, overwriting all of the values it just saved, essentially erasing the initial user.
Because the match-on value is the only way for the system to know which user it is processing, it is absolutely vital that these values be unique.