Make things that matter
Everyone in a modern company wants to add value. This should be no different for a data scientist working on an idea, a project or a product. But how can one know that their activity adds value?
By focusing on things that matter: matter to engineers, business leaders, product teams, customers, stakeholders, everyone. If data scientists lock themselves in their ivory towers creating proof-of-concepts, doing "research", cleaning and labelling data, running experiments far away from frontline KPIs and user feedback, no one should be surprised that their activities pure and simple: doesn't matter. This will manifest itself by lack of interest, resources and involvement from other stakeholders of the business.
So how would one avoid falling into this trap? Pre-startup era software engineering were marred by the same problems as machine learning today. Slow iteration, waterfall, lack of client feedback loop. To answer these the agile/lean movement was born and facilitated an unprecedented boom in software product development. It would be a waste of opportunity of not examining the mechanisms that enabled this and apply it to machine learning.
In subsequent posts I will describe which techniques can be applied, modified or should be avoided to facilitate MVP style ML product development. Machine learning by nature adopts the waterfall style so there will be several hoops to jump. Please stay with me throughout the journey.