I consider this to be the end of a successful experiment. They tested an idea: can we use AI for this? They then backed it up with “NI” – Natural Intelligence – to step in when the machines were unsure. And they were fully invested in it, deploying to half their stores, and stuck with it long enough to yield meaningful results.
The outcome was that despite years of work, machine vision wasn’t good enough for this use case. But just because the outcome didn’t deliver the desired results doesn’t mean the experiment itself wasn’t done well.
