Their addition of AI into Quicksight (their BI tool) really cemented for me how much they're willing to bluff to look relevant. They launched the AI addition at least a year ago, but recently rebranded the whole thing as "Quick Suite", promoting new AI capabilities. It really just seems like a new skin, with more purple. So far, out of 4 people at my org who tried it, all were disappointed after upgrading. I get much better results from sending a CSV to ChatGPT and asking for a graph.

It just doesn't seem to work out of the box. I test drove it on 20 BI questions, and for trivial questions, it gets about 50% right, and for more nuanced questions, it gets about 20% right. For example, I have a table like "people" with a column like "job_title" which has 10 possible values in an uneven distribution, and I asked "what fraction of people are nurses?". It gives me a pie chart where each slice wrongly gets 10%, because of the 10 possible values. Or if I asked what the mix by decade was, it would just show a flat line, 10% each period.

Then you have the more tricky questions, like "show me the revenue for each of the top 3 products in each quarter". It calculates an all-time top 3 first, and then shows the same products each quarter, instead of redoing it within each quarter. Reasonable enough, but when you try to correct it, it just gets stuck, possibly with an even worse answer, and there's no way out of the loop.

Why would they let executives touch something so half-baked?