Public facing hard to guess identifiers

This might take some explaining of the actual problem. When applications are reporting information I would like there to at least not be possible to guess an identifier starting at 1 and have the data end up on some other users account. My goal isn't so much to guard against somebody who is intentionally trying to do misreporting, but to make it hard enough to do for all but the determined attackers.

So how do you do this, one way is to just use GUID's for every identifier but I have always hated that and it leads to bad database design at least in my opinion. So my suggestion is to just use a simple integer identifier counting upwards internally. However whenever the identifier required to report for this is displayed to an end user I take this ID and encrypt it using a secret key with AES-256. This results in a pretty much random 16 byte array that you then encode using base 64 and present to the user. Once any reporting is done you simply do the reverse so that the ID needed to be used is the base 64 encoded encrypted value. This means that it will be almost impossible to guess a valid identifier for anything coming from the outside but internally you can still just deal with regular integers of varying size for everything.

The performance hit should be negligible since AES is implemented in hardware in recent CPU's and even without it AES is really fast.