One point here - support emails do not help you identify problems that less invested users may be having with your product.
For example, I used to work developer relations on TensorFlow. We wanted to make the framework accessible to enterprise data scientists. The problem was that these users were not familiar with the tools that we commonly used to get feedback - GitHub issues, the mailing list, etc.
Most of them were using TensorFlow on Windows via Jupyter, which wasn't well-represented among the users that we had frequent contact with.
It was really hard to understand the universe of issues that prevented most of these users from getting beyond the "Getting Started" experiences. Ultimately, these users are better served by easier to use frameworks like PyTorch, but I think a big reason that TensorFlow couldn't adapt to their needs is that we didn't understand what their needs were.
Another big problem is that it takes a certain level of technical sophistication to know how to send maintainers a useful crash report. If you rely on this mechanism, you will have a very biased view of your potential user base.
For example, I used to work developer relations on TensorFlow. We wanted to make the framework accessible to enterprise data scientists. The problem was that these users were not familiar with the tools that we commonly used to get feedback - GitHub issues, the mailing list, etc.
Most of them were using TensorFlow on Windows via Jupyter, which wasn't well-represented among the users that we had frequent contact with.
It was really hard to understand the universe of issues that prevented most of these users from getting beyond the "Getting Started" experiences. Ultimately, these users are better served by easier to use frameworks like PyTorch, but I think a big reason that TensorFlow couldn't adapt to their needs is that we didn't understand what their needs were.
Another big problem is that it takes a certain level of technical sophistication to know how to send maintainers a useful crash report. If you rely on this mechanism, you will have a very biased view of your potential user base.