Obviously a distribution would be best, but for a single number, the mean is better. For example, if groupon says they have 100,000 customers and they will have 200,000 customers next year, and the mean # of groupons per customer is 2.5, then I can compute that they sell 250,000 groupons per year, and if they get 200,000 groupons, they will sell 500,000 groupons per year. The median tells us none of that.
