Build Successful Data Strategy: 5 Mistakes To Avoid

Photo by Mark Fletcher-Brown on Unsplash

I worked for different companies for my last 5 years as a Data Scientist and I saw data science project fail and other success.
Here I will explain the reasons why promising projects never saw the light of day in production and how we can avoid this kind of failure.

I. Having the wrong guy as Manager & Tech lead

I saw that a lot of times. Having a data manager or a data tech-lead who have proven their skills on others fields like developments is the big mistake because:

• He don’t understand machine learning
  He don’t understand how machine learning works and when the company will introduce him a new need that the team should resolve he can’t say if it’s doable or not and he will bequeath this responsibility to his team. They will have to assist to different meets to understand the need and explains the best way to do it. This will create a feeling of doing more than their scope and it is which leads to a feeling of misunderstanding.
  Furthermore, he will not be capable to make decision when the team member does not agree on a technical choice.
• He never deploys a machine learning model
  This emplies that he doesn’t know how the team should interact with the whole chain:
  - What is the role of data scientist
  - What is the role of data engineer
  - What is the role of dev/ops
  Dealing with this different person without knowing exactly the scope of every one could be difficult.

The aim of a manager & tech lead is to put in place a strategy that allows the members of the team to all move in the same direction while working on various subjects and to ensure that the prerequisites are well respected.
In this case, we realize that it will not be the case if you don’t choose the right manager & tech lead.

II. We want deploy something perfect in production

A data scientist almost wants to deploy a perfect model with the best accuracy, precision, recall … but this is a bad idea sometimes.
By dint of wanting something perfect, you will take a long time before deploying a first version and each time you identify a possibility of improvement you will postpone the delivery date and this is…