We've all heard the news: being a data scientist, an Machine Learning specialist, or a Digital Transformation Specialist are among the sexiest jobs of 2021. You picture your company's data team working in a dark room with large screens, moving graphs on them, and a whole Tony Stark setup that looks like a dream job to many. But in reality, companies constantly face turnover from displeased employees who, in your opinion, are not keeping up with the end of the bargain. So let's backtrack a bit. Forget about the fancy computers and the Avengers headquarters set up and go back to the basics. Why is your data team quitting? If finding this role in today's world has turned more accessible, given that the job is rising in the number of available talent, why are they not staying? And most importantly, why aren't they delivering what they promised they could do for your company?

Reality vs. Expectation

The first reason why data scientists quit their job is that companies think that their infrastructure will be in place and running to reap the benefits of AI by hiring them. 

Whether large or small, any organization that wants to introduce AI must first guarantee that they have real useful data to work. Paired with that is having the right team, the right leaders, and the right mindset. It sounds like a lot, but data is not about one person. It is about culture. 

  1. Usable data: your business may be satisfied with simple log data from an app or website. But companies who want to leverage AI to improve business must ensure that the data they gather is in the correct format. The right sort of data depends on the industry and company. That may be log, transactional, customer, or categorical data; your data team will determine this.

Apart from collecting loads of data, companies need to ensure that they are gathering usable data. For example, if you have an e-commerce presence and receive orders online, you may be recording transactions. Your team needs to first and foremost decide what type of transaction is "a transaction." Is it a paid transaction? Or is it a transaction being processed? Once you label the data you're collecting, you can ensure that your logs store the information YOU need.

This is what is called the Cold Start Problem. The ancient philosopher Confucius has been credited with stating that you should study your past to know your future. In particular, supervised machine learning relies on the availability and use of labeled data (things past) to classify previously unseen data (things future). Without labeling data from the past, how will you label for the future?- Knuggets

"Study your past to know your future."

How do you do this? By visually looking at your data. Companies take a step back and put their data out for their team to visualize it very few times. A practice that positively impacts all departments tremendously, taking a look at the data before using it is vital to know what you have gathered before running projects.

The right team: Once the data is analyzed, stakeholders can start communicating the KPIs (Key Performance Indicators) they want to improve. An organization can then select what reporting or dashboards are required for teams to make educated decisions based on those KPIs. And concentrate on automating the systematic computational analysis of the data or statistics. 

Data Teams are not gods.

Being a data analyst doesn't mean being a data scientist and a database expert. They're all different roles, and understanding each one is vital to know what each one can bring to the table. It is widespread for executives to assume any data professional knows everything about Spark, Hadoop, MySQL, Python, Tensorflow, A/B Testing, etc. Considering that, it only says your company may have no idea what the data strategy is. Seldom to say that because someone is senior doesn't mean they know all this stuff. They don't because they focus on a handful of skills for each role. It is your job to determine who can do what for the goal in mind.

Don't isolate your data team.

In the case of services offered by a company, we can immediately recognize a successful data project when the user experience is intuitive. Furthermore, the results of the said project are perceived by the user with offerings that are tailored to them. This happens when the data team is integrated into the business, thanks to a collective goal and execution, which was achieved with the work of the data team and sales, product, marketing, and so on. 

Picture your data team working in isolation to run machine learning and put machine learning models, which will most likely take most of their time. They will undoubtedly get isolated from the rest of the company. Even though that is a necessity, you may be setting them to fail. 

Thankfully, technologies exist to relieve their time spent learning new machine learning algorithms by providing easy-to-use tools to build and put these models into production without writing code. Leaving them with the much-needed time to send the outcomes directly to their customers and reach those much-needed goals.

  1. The right leaders and mindset: If the leaders and the company won't have the urgency and philosophy of a data-driven company, then the data team will fail. Data is not nice to have. It is strategic; treat it as so. Please include it in your strategic plan, train, communicate and embed it in the culture. Consider even having a VP of data or if you are a small company, have it as one of the primary considerations when making decisions. In our case, we defined that making decisions based on data was our north star.

Ask yourself these questions, and you may find out why you have high data talent turnover. Of course, the nature of your company may face other challenges. Still, generally speaking, these are the most common reasons your data teams are not staying loyal. Talk with them, find frictions in their day-to-day and apply these findings to their answers to ensure you sail smoothly towards bettering your KPIs.