Updated: Aug 1, 2019
While everyone is chasing after the always expensive and often data scientist, I have been advocating for organizations to instead consider building a data science team. Here are a few reasons:
First off, there are just not enough data scientists out there to meet the surging demand. Based on the data there is at least a 3 to 1 data scientist demand to supply ration right now across the U.S.
Second, there is still a lot of debate about what it takes to be a true data scientist as opposed to someone doing data science. Statistical analysis, predictive model building, machine learning all in one person, who also has subject matter expertise, good data visualization skills and can break findings down to non-tech people is a lot to ask.
Third, there is a lot of risk associated with bringing in one person and depending on them to completely transform the way data is optimized in your organization. The data scientist turnover is pretty high as expectations of both employee and employer are often out of alignment.
My approach is to fill out a team to do the work of a traditional data scientist but spread the work out into traditional data positions in an organization.
This approach is less risky, covers all aspects of a good data science solution and is much more achievable for organizations with lots of data, but with less resources.
If you break down the data science life cycle in most organizations, you end up with the following job families:
1. Data Steward/Data Gatherer
2. Data Engineer/Database Admin
3. Data Modeler/Statistician
4. Business Analyst
When you have strong contributors in each of these roles, you can often do as much if not more than a traditional data scientist. You might sacrifice some of the more advance model building and machine learning, but for most companies that is down the road stuff.
Now if your organization really is ready for building prescriptive analytics models, machine learning and process automation using A.I., then you will likely have to go out and fight for and spend a lot on a true data scientist. One with at least a master’s degree in stats, math, or some comparable program, who can code in R and/or Python and who can visual data using Tableau or Power BI would meet the test.
One last caveat though, if you do hire a data scientist, make sure your data is well governed. The leading cause of attrition for data scientists is forcing them to spend most of their time cleaning up and organizing data. If your data is a mess, clean that up first then hire a data scientist once that’s done.