What was the source of frustration? What steps, if any, has your team taken to fix the problem?
1. The massive hype around ML has produced unrealistic expectation.
We often face the issue, that customers are unhappy with our results, because their high expectations: "Uh, your model has only 78% accuracy, can't you do better?" (No, not without adequate data and resources!) You basically disappoint people very often.
2. There are lot's of fraudsters in the game, that might get the fame.
I have seen data scientists being applauded, because they claimed to get "99,7% accuracy" for an regression problem. How the hell did they calculate that? Accuracy isn't even a good metric for regression problems! Of course those models usually don't stand reality, but that doesn't seem to be relevant...
3. The work might not have a relevant impact.
Often, we do prototypes to tackle problems, that don't even have enough impact to ever become profitable. We are set on it, because some manager has been told to "Leverage AI to improve Business". As a consequence, when it becomes obvious, that the resources needed to run something in production will never create a positive ROI, the project remains a prototype. Of course our Team knows & communicates that often from the beginning, but it doesn't even seem to matter, as long as anyone can put "working on AI" on some PowerPoint slide.
4. Most of the work is "boring" data preparation.
The "cool" modelling part of our work, where you design architectures and evaluate algorithms usually is ~5% of our implementation time. Most of the time is spent in preparing the data to be suitable input for the model. (I myself actually like data prep as a part of the process, but I know lots of colleagues don't).
This is my experience from working in a very large but not digital native company. I'd expect it to be different at Amazon, Netflix, Google etc. And I'd be interested to hear if data scientist from other companies face similar or different issues...