All projects related to Data Science start with data and usually with a question. Typically a business question like. "Can we predict when and why a client will leave us?", "Can we predict that a new TV series will be successful?", "How can we sell more iPhones to increasingly affluent people aged 70+" and so on. The first reason for failure in the Gartner article top four reasons is the most critical: inappropriate or siloed data. Data scientists in real firms have to work with the data they have in front of them, but often the real data that they need is not made available for political reasons, timing reasons, security reasons, quality reasons or even just the sheer lack of knowledge to extract the data. Another reason lies in the communication of data science results. Sometimes a client will ask a question but would prefer the data science results to "confirm" their "gut feeling for the answer rather than give back a conflicting result.

So Emmanuel, can you predict what last week's weather will be?