Sergey Nivens - Fotolia
According to this month's LinkedIn U.S. "Emerging Jobs Report," data scientists had the third largest annual growth, with 37%, and have been at the top of the report for the last three years. That puts data scientists -- and data science skills -- in high demand.
The skills required for data scientist jobs include machine learning, Python, R and Apache Spark, according to the LinkedIn report. But once those fundamentals are in place, what does a data scientist do next to prepare for the future of their career? Experts recommend these primary skills for data scientists.
The most sought-after data scientists not only know how to use the technology but when to use it and why, experts say. That requires an understanding of the business benefits of the data, as well as the ability to communicate these benefits to executives and colleagues in other areas of the company.
"I've seen a lot of data scientists have a good understanding of the models and not have any business impact," said Andrew Fast, chief data scientist at CounterFlow Inc., a cybersecurity company based in Charlottesville, Va.
To address the issue, some professionals are zeroing in on specific use cases.
"We see that the market is getting more specialized," Fast said. "Instead of somebody being good at data science, you might say that someone is interested in data science for marketing or data science for IoT. Developing a specialty or focus on a particular area will help you get traction as data science and analytics grow in popularity. Having an area that you can call your own is very valuable for success."
Andrew FastChief data scientist, CounterFlow AI
An understanding of finance and how data science projects can help a company's bottom line can also be helpful.
"Many technical and scientific people have no feeling for money or economics," said Michael Feindt, strategic advisor and founder at Blue Yonder, part of JDA Software Group, a German AI software company.
In Europe in particular, he said, it seems that data scientists think of money as a bad thing. An MBA or other formal education program can help, Feindt said, but the best place to learn about business is at your current job.
"For people who are intelligent and analytical, learning about business is pretty easy," he said. "You have to want to learn it, but not be too arrogant, and have respect for people who are not too scientific."
Soft skills lead the way
Outside of technical knowledge and business familiarity, good communication and other soft skills are at the top of the list for necessary data science skills in the job market.
At True Fit Corp., an AI-powered fashion recommendations company, it's the nontechnical skills that make the difference.
"There's been a lot of emphasis on technical skills, on learning a specific algorithm or class of algorithms," said Rhonda Textor, True Fit's head of data science. "The tricky thing is that if you don't understand the problem that you're solving and don't understand how the business you're part of works, how it makes its money, how your algorithm feeds into the bigger picture to do that -- then it doesn't matter if you understand everything about neural networks because you won't be able to apply it to the right problems."
The company looks for good communication skills, she said. "And we also look for people who don't just understand data science but can map that to business value and business impact."
Even the most technically oriented companies, like San Francisco cybersecurity vendor Kenna Security, are starting to prefer data scientists who can look at the bigger picture instead of chasing the latest algorithm.
"Whether it's deep learning or random forests or regressions, those conversations are becoming less interesting," said Michael Roytman, Kenna Security's chief data scientist. "What I'm most excited about is figuring out the right way to structure research, data science and development together, so that from inception to production, a new product can see the light of day faster. To me it's all about the soft skills -- communication, people and teams."
It's all about the execution
Data scientists who know how to work with data engineers, or themselves have strong engineering skills, can make the difference between just having a showcase pilot project and delivering a strong ROI with a successful implementation.
This doesn't mean that data scientists necessarily need to retrain as software engineers, CounterFlow's Fast said.
"It's a tall order to expect a data scientist to be good at both those things," he said. "But they need to understand the language of software development."
Future-proof platform skills
The Python programming language will continue to be a fundamental skill for the foreseeable future, and data scientists can augment that with new techniques such as neural networks, deep learning, adversarial networks and transfer learning.
There's also a growing emphasis for data scientists on cloud-based deployments, which allow for easy transfer from testing to production environments, automatic and cost-effective scaling, and integrated and easy-to-use libraries of algorithms and training data. Plus, the platforms integrate well with the most popular data sources, such as commercially available data sets, as well as data sourced from in-house systems or business partners.
The three main stacks belong to Google, Microsoft and Amazon. There are also dedicated data science platforms, such as Databricks, DataRobot and Domino Data Lab, which allow data scientists to collaborate on building and deploying models. So it may benefit data scientists to become familiar with these platforms.
"These tools are trying to become a one-stop shop for everything you want to do, and they have the right integrations, the right functionality, and are highly flexible," said Ivaylo Bahtchevanov, head of data science at ForgeRock, a San Francisco-based cybersecurity company.
Most importantly, cloud platforms make it easy for companies to start small and then scale easily.
"You can easily expand the number of servers or resources you need without having to go and purchase actual hardware," said Charles Ng, vice president of enterprise AI at Appier, an AI-powered marketing technology company based in Taiwan.
"The AI and machine learning platforms are changing dramatically," said Myke Miller, dean of the Deloitte Cloud Institute, part of Deloitte Consulting LLP.
The new platforms simplify the work of the data scientist, he said, and add a tremendous amount of automation to the data science pipelines. Data scientists need to learn how to use the new platforms, but after that, the platforms actually reduce the need for certain specific data science skills.
"The need for Python or Firehose or any of the component products is diminishing," Miller said. "And the need for collaboration and understanding the business problem is continuing to rise."