As programmers we really like to automate away the mundane stuff, and we're not too fond of human interaction. I'm grossly exaggerating, of course, but for a lot of us, this was the push and pull that originally drove us to hack away at computers. Creating a big and shiny automated processes is almost a no brainer for us.

Yet, I believe that you are absolutely right. We should fight our inclination to see everything as a nail for the hammer that we master. The ROI on automating everything from the outset is not worth it, and if the business flops, it's just wasted effort. Worse yet, it's a missed opportunity for learning about the customers and about the domain.

So yes, do things that don't scale. If and when you get in the air, profile the business processes and only then automate the bottlenecks. It's the only sane way when you think about it.