No matter data talent or any other type of IT professionals, being learning agile is the most important attitude due to the changing nature of technology.
Data Analytics has become the top priority of CIO's agenda at any forward-thinking organization today. However, high-quality data talent is still in serious shortage: Should data scientist a domain expert? Does a statistician need to be a domain expert to perform effectively? Does a programmer need to be a domain expert to perform effectively? Does any technical expert need to understand the business domain?
The problem is that today most companies see the data scientist as a single role. Few companies sees data science the ways it should be a broad spectrum career path that goes from the business analyst and a subject-matter expert to specialists in statistics and modeling. Because in the end, they should take responsibility to provide decision support to the management. Unless they know the domain they won't be able to contemplate and analyze the situation. If they have domain knowledge, they know which variables to include and which ones to exclude; and why are they important and how to investigate better. Complementary roles of domain experts and data scientists have to be acknowledged. The domain expertise is useful on two levels - first to tackle the problem and second to understand the client. In particular it is necessary to determine how to best serve the client's needs in light of the scope. But more on a theoretic level, a scientist should accommodate emergent needs that have not been articulated as such. In the latter case, domain expertise forms the beginning and end of any analysis shaping what gets considered and what becomes invisible. If a person does have analytical skills without domain knowledge then he/she can acquire the domain knowledge. In general, you need to know the basics, but most importantly, you need to be quick on your feet and adapt if necessary. It's useful to know "all" statistical methods - but you might just use something you hadn't heard of before.
It certainly doesn't hurt to have domain expertise. Being a domain expert not only reduces costs but also improve analytics efficiency, so it’s important to show how to make all the pieces (business side, product development, technical) fit together. Companies should understand this before investing on professionals. Fail to understand this will render all investments useless they will be frustrated by the lack of good results. You need to be quick on your feet and adapt if necessary. It's useful to know "all" statistical methods - but you might just use something you hadn't heard of before. Same goes for domain expertise: you need previous experiences that you can leverage and compare with to be able to understand the context quickly, and it's imperative to do so. A domain expert with data analytics expertise can facilitate in 1). Data interpretation,
2). Data visualization,
3). Choosing better data analysis technique whose result will carry some meaning to user,
4). Market the result by explaining it to the user as he /she is one amongst them, and
5). Improve the technique with user inputs.
The answer is almost always - it depends on. In software if domain expertise is prioritized to the detriment of technical ability, your project will accumulate a mountain of technical redundancy that may never get paid off and cost huge amounts with time. After a while, you also realize that there are broad similarities that underlie many different industry domains, regardless of what specific expertise is required, and often times you can rely upon this to better establish your own credibility even when you don't know the domain. This is especially true with both data modeling and data analytics, which both involve understanding the deeper structural connections of information first, and only then getting into the weeds of domain level content. At some point, the value of the expert/decision-maker to the scientist/decision-facilitator is just to implement the results diligently and timely. Unfortunately you can lead a horse to water but they may choose not to drink, at which point you must coerce them with logic, KPIs, and pretty plots which still may be seen as a foreign language. There are two scenarios: domain experts come up with some hypothesis (some candidate patterns or regularities) which may or may not be statistically significant (spurious pattern). Validity may be tested using Data Science algorithms. The other scenario is that some hypothesis is generated automatically by "Data Science" algorithms, but that may be based on correlation (may not be causal etc.), and therefore it needs domain experts' nod if it makes sense. 
1 comments:
Such an ideal piece of blog. It’s quite interesting to read content like this. I appreciate your blog
Data Science Training Bangalore
Post a Comment