Andrew Morgan, CFA

A common issue in defining goals at senior executive level, is the lack of understanding of what is feasible with ML, given the realistic state of the data available. Therefore I believe goals need to be defined first at a level irrespective of the solution, be it traditional software or utilising AI. Only after an initial exercise is done through exploratory data analysis and crafting the first set of ML hypotheses, can clear, measurable goals be set out - and these can be more directly linked to minimum accuracy thresholds of models. All too often we see ML projects fail because initial goals are not realistic, or require levels of investment far exceeding the value because an initial data assessment and testing is not done first.