This session submission is intended for the Best practices in Data Science and lessons learned track of SDS 2018.

Over the last years, my data science colleagues and I have had countless conversations with customers of all industries and sizes on potential data science projects with questions ranging from “what can we actually do with data science” to “how can we put what we have built into production”. In this talk we will share the 10 key findings we have distilled to help the audience in avoiding some of the common pitfalls.

The talk will be divided into the lessons learned before, during and after the actual implementation.

In part I we will cover lessons learned before a project gets implemented. We will first walk through the buzzword jungle and highlight that AI, ML, deep learning and data science are not synonyms but are topics intertwined with one another. Among other things we will also touch upon how the very first data science project can determine
the fate of data science as a whole in a company.

In part II we will have a look at lessons learned during the actual implementation phase. This will include (but is not limited to) the importance of using a methodology like CRISP-DM and that one should not a priori focus on a particular technique (e.g. deep learning) just for the sake of the technique, no matter the hype it comes with. (No free lunch, anyone?)

In part III we will finally claim that work is not done when the last line of code is written and the expected model performance is accomplished. Careful thought needs to be put into how business processes, applications and people now interact with the data product.

Time: 14:30 - 15:00
Track: Track 2
Speaker: Marc Schöni