It depends on what kind of pretesting you are trying to do. If you are doing cognitive interviewing to assess the effectiveness of your question wording, the 10-12 interviews will be enough. If you are trying to establish the reliability of the items for a newly created scale, then you will need more.

The answer to the question lies in the usage of pilot tests. Pilot tests are generally used to test the questionnaire of the study and get in touch with some realities of the field that will be considered during the survey propre.

Ok this basis, i think even 20 interviews will be ok.

The basic rule is that the larger the sample, the more stable your results. If by pilot you mean that you want to get an initial indication of how reliable an instrument is, it will depend upon the types of metrics you are using. If the survey results based on objective type items, it could use as few as 30 or so observations to get your initial estimates. The key in this type of situation is that your sample include the full range of variability you might encounter in general use. If items are open-ended, it could be done with fewer observations but it could take more. The key is to keep going until a consensus is reached or you have exhausted the types of responses you might receive. For these types of items, it is most important that respondents are understanding the question in the same way.

You have the sample size you want before doing a pilot study, but one big reason for a pilot study is to determine your sample size requirements, so it seems the cart is before the horse.

James McLean has a good answer, but I'd like to note that a large sample size may not be as adequate as a smaller one well planned. If you took a stratified random sample you may learn more than from a larger sample which may be highly biased. Even "big data" can be of very low quality if it comes from all but a few important segments of a population. (Bias in statistics here means that you have not properly represented the population. There are ways to account for that with auxiliary data. There can be a bias/variance tradeoff. But I have noticed the word "bias" misused in every day language when someone doesn't want you paying attention to facts, and that distorts the actual meaning of the word "bias." Further, in statistics, as in other disciplines, there can be jargon too, where words have special meanings, so communication can be challenging. In statistics there can be different sources of bias, but a good analogy of the impact of bias could be throwing a basketball from the free throw line. If your basketball shots tend to fall to one side of the basket, then there is a bias.)