Five Hypotheses as to why Artificial Intelligence and Machine Learning projects fail

There are numerous articles and published papers around why AI/Machine Learning/Natural Language Processing projects never make it to "production" or fail to deliver on the value proposed (Gartner predicts that through 2022, 85 percent of AI projects will deliver erroneous outcomes due to bias in data, algorithms or the teams responsible for managing them). The following are a few hypotheses as to why Artificial Intelligence projects fail and some observations on company reactions (technical and organizational) to AI projects lack of delivered value. Please note these opinions do not reflect on any current or prior employers, but are a synthesis of conversations with data scientists, engineers, product managers and architects across industries. Hypothesis #1: Data Science initial models don't scale or are too experimental to be used by internal or external customers. Projects often start here as companies hire on a few data scientists who build their models in Python or R, only to discover quickly that there is a difference in mindset between data scientists and engineers.