I worked for years as a BI consultant for some rather large Silicon Valley companies. My experience felt like 70% wasted time creating pipelines, data restructuring and visuals for insights that had minimal to no effect on the company's business.<p>30% of the work we did seemed to have value - but that was with a relatively targeted and preset set of questions to answer. It always felt as if someone higher up said, "let's get this data in shape to use for analysis" but never had any questions to ask of that data to begin with. I suspect that if LLMs can truly understand complex interactions across a multitude of datasets, the context within those interactions, along with dozens of exceptions, then it can be of use.