We use Net Promoter Scoring on our Quarterly and Annual Services Survey. I am not directly involved in deciding what the "statistically relevant" quantity of responses is, but I can tell you that we send out 11 surveys to about ~20K people every quarter. Every Survey has the NPS basis question on it, among others. From those 20K we get a average of 1.2K total responses.
We do not directly involve ourselves in compiling the survey results, as we do not want any possibility of perceived influence in the scoring. I have talked to the people at the company who compiles our scoring and they tell me that for our purposes, they would be comfortable with 1K responses in generating the NPS but are happy with 1.2K.
Of course, ideally we would get a response from every survey mailed--which would give us a complete picture. But even if we got that, there would be people out there who would survey with better and worse scores for the same time period if they were surveyed, so no survey can truly be 100% reliable, even with a 100% response.
Can the sample be too big? I think the question is really reliant upon your ability to process the information. If you find yourself spending huge amounts of time entering response data, you may want to do more research on how much is needed to be relevant.
If I got a 100% response rate from 20K surveys, would it really make a huge difference from the 1.2K average response? Experience tells me probably not. I have had survey cycles with as high as 2.3K responses, and only a 1 basis point shift in scores. On that projection, 20K responses could only shift scores about 3 or 4 basis points (based on doubling of the response rate = .01 point shift). So the effect of 20K vs. 1.2K is minimal.
Look around the Net for "statistically relevant". Even if statistics is not your cup of tea, it's a good concept to have at least a ground level understanding of. I know that when I started my job here, it was a concept not understood by many who looked at the Surveys regularly. A little education went a longway to solving some very confused opinions.
