Giving AI a set of numbers and asking for a summary is useful, but it is also the most obvious thing to ask.
Try asking it to be awkward instead. Which figures look unusual? What does not fit the overall story? Which customer, product or month deserves a closer look? What conclusion could be wrong if one assumption changes?
That tends to produce a much more useful management conversation because you are not asking AI to rewrite the dashboard in sentences. You are asking it to look for the things that deserve attention.
You still check the source numbers before acting on anything important. But “challenge this” is often a better instruction than “summarise this”.