[danicgross]

Yes! In many ways, all your initial idea needs to be is a launch pad. Once you get your foot in the door, you can always refine based on what users tell you.

Per your point, it's important to learn from their workflow, not ask them what they want. When you ask someone what they want, they often perform attribute substitution (https://en.wikipedia.org/wiki/Attribute_substitution) and answer a different question.

[got2surf]

Great point. Also important that your customers may not even know what is possible, so they won't know to ask for it directly. Instead, if we ask why they want something, we can understand their intents - and translate that into technology.

We've seen this a lot with text analytics - instead of trying to repeat what humans do manually (reading/annotating), we try to understand the questions that clients are interested in (the drivers behind their current workflow).