There are no firm answers available to that question. But we can offer a few ground rules.
- Be sure to analyze your need for interviews bottom up. That is, you need to build up your sample from the lowest group of interest. If you really want to know how young women 18-29 years view your site, it doesn´t make sense to set an overall target of 1000 interviews (which is common in many market research studies). Instead, you should aim for, say, 250 interviews with young women and then monitor the number of interviews completed in this group (by exporting the data-file from UserReport to, for example, a spreadsheet where can see all individual records).
Following this, a workable rule is to say that you need 200-250 interview in any of the groups in which you have a specific interest (often such groups are defined in demographic terms: male/female, young/old etc.). If you want valid results for both men and women (but no additional breakdowns to, for example, young men and young women) and your site has an equal proportion of both sexes among visitors, then you are okay with an overall sample of 500 interviews. You may find also useful to use customer personas to better define the groups you are most interested in and follow their answers’ patterns. You can read more about customer personas here.
When you make tracking surveys, meaning that you track website performance from, say, quarter to quarter, it´s a good idea to keep a low-frequency survey running at all times instead of collecting all your interviews in one or two spikes. This makes evaluations less vulnerable to one-off events such as good/bad media coverage, individual campaigns and so forth. The UserReport system makes it possible for you to aggregate interviews throughout any time-period you desire.
Conversely, if you wish to measure the exact effect of a specific campaign or news event you should, of course, make sure to launch the survey immediately after (and/or during) the event. If possible, do so also before the event in order to isolate the impact of the event. This is particularly pertinent if you want to measure how people evaluate changes on the website.