Somebody had to say it ...
It's a common approach in data analytics to provide insights by creating colorful dashboards. These tools have become very powerful and the charts or graphs are very impressive. But is this really analytics ...?
Most data drive analysis works like the following:
Training: Historical Data + Algo => Model
Application: Real Time Data + Model => Prediction, Regression, Classification, .... depending on the algorithm
So what is the Algo part in Dashboards ...? Correct, it's the developer or user. Humans still have an unmatched capability in image processing which included classification and pattern recognition. The training of our human algo machine is the experience, skills and expertise we gain by working with data for a long time. This is evolution at its finest ... good to know what's a mammoth and what's a sable-tooth.
(Somewhere on LinkedIn I read an article that a group has developed a system that can detect 90% of pedestrians crossing a street ... I'm pretty sure I can do better ....)
And even though this is a powerful combination, it is an outdated concept. Nobody has time to stare at a screen for prolonged time, especially not in 24/7 operations. It might also be a waste of your most valuable\experienced resources to perform simple classification tasks.
There is no way around it: Most of the analysis part have to be performed by algorithms. The system has top be set-up in a way that information flows to the operator and doesn't have to be retrieved from screens. IMO this will lead to an enormous modeling effort during the transition - which has already started - , but will at the end lead to much more robust production.
I think i agree on the most basic point of your post: never underestimate the humans. But you also point to the exact reverse: humans are really bad at some things. While looking at a dashboard is not the best way to support decision making, you have to account for the time it takes an organisation to mature. It might be true that a good set of rules or pattern detections might alert the users there is something to act on proactively, it is not at all easy to get at that point. And in many, many cases a good dashboard is a great step towards a better control of your plant.
It is difficult nowadays when everybody likes to see machine learning, predictive analytics, but when analyzing the real needs you find that a good report, a smart set of KPI's and a well organised set of people can do much more than the buzz words could ever.