Any guidelines for assessing, rating Virtual Agent conversation complexity?

[Joe Tanenbaum]

Number of variables?  Branches? What thresholds do you recommend for distinguishing between low, medium, high complexity?