Michał Kosiński
Data Scientist
Every research & data project must obey the same rules as any other project in the company - it must bring value. Only then it makes sense to invest. Data science is tricky, though: as opposed to an engineering project, uncertainty is much higher, and progress is non-linear. To succeed in data science projects and bring real value to the business, the team must encompass unorthodox techniques and rules.
Without data, there is no science in data science. The real world is messy, and so is the data obtain by measuring that world. Data in raw form is often unstructured, has hidden bugs, missing values. In most cases, data has to be obtained from multiple sources, carefully defined and described, dealt with missing data, and if proven valuable, stored and maintained for future use.
Analytics is a vital part of the data science cycle. It allows to uncover bugs in data, define directions for data-based features, help prioritize engineering work, or help drive business efficiently.
Machine learning is the cherry on the top of the data science, and so it has to be considered - not as a base of everything we do, but as an extension. We always should strive to build features that allow an end-user to achieve his goals with the least friction - if machine learning can help us, that's great, but machine learning per se should never be treated as the end goal.