Learning Disability
I like my Sense monitor and I would buy it again. I love the concept of monitoring electricity consumption, by device, using machine learning.
I can't defend this product as actually working though, and I can't recommend it to anyone other than early adopter's and enthusiasts who are willing to look past the shortcomings.
The primary issue is that the Sense can't seem to "learn" the majority of the electrical loads. I heat my house with an electric heat pump and two air handlers. Sense hasn't learned any of that. I heat my water with an electric heat pump too. Sense has maybe learned to recognize a portion of that (the fan or something), making for dirty data. "Unknown" would be better than incorrect data, in my view. Sense has learned my electric car, but the data doesn't align with what the car itself is reporting (if only there were a way to pipe the car's data into Sense...).
I am willing to buy more equipment for better monitoring, and I have added several Kasa smart plugs to my house simply for their monitoring capabilities. Those are only available for lower amp, 110v loads though, so I have precise metering of my AppleTV (3w) and no visibility into my heat pump (probably multiple kws). 60% of my consumption is "other" (unknown), so between Sense and a bunch of smart plugs, I can monitor a minority of my electricity consumption.
Remember how Tesla was trying to have a 100% automated factory but couldn't get a robot to lay a sheet of cloth over a battery? They gave up the goal of perfect automation and supplemented it with alternatives that were known to work (in that case, humans). Sense should continue to build up its AI/ML as its primary strategy, but please stop hoping AI/ML will solve everything near term. It's time to enable direct monitoring of 220v, high amp circuits.