You’re looking for a senior data science leader with a proven track record of success over 10 to 15 years. Where do you start? I’m going to help.
It can be confusing when navigating the sea of the latest data-science buzz words. Do you need someone that knows the difference between Scala vs. Julia? Or if PyTorch is more useful than TensorFlow? In looking for a data science leader, do you weigh learning intelligence above emotional intelligence? Is the candidate experience-building teams more beneficial than just managing an existing team? Does the ability to build and foster relationships matter more than technical competence?
Each of these questions is difficult. The answers may vary depending on the specific role. But, there’s one common element across all these questions: It will benefit you to think about these answers before you start receiving resumes and evaluating candidates.
Data-science is about creating value
It’s exciting to ask questions about Snowflake, Pig, Hive, or NLP. It feels like the company is more progressive if it asks about a candidate’s knowledge using Hadoop.
However, it’s more applicable to ask for specific examples of how data has been used to create value. The answer could apply to biologics. The response might have origins in finance. The setup may support healthcare. Examples might be eCommerce-based. The shocker is, it doesn’t matter.
You’re not evaluating a technology. You’re assessing the candidate’s application in context. This is the essence of data science—extracting insights that otherwise weren’t possible to access. What was the business problem they started with? How are they using data for continuous intelligence? What can the business do today that it couldn’t do yesterday? What capability has been enabled or expanded?
Data science leaders always start with the business opportunity.
Terminology and skills that exist
Whether we’re talking about dual processors, 80-node clusters, or a multi-core environment, the terms mean similar things at different points in time. Terminology changes over time. Are you referring to CPU capacity or elastic compute? Guess what; they’re the same thing. Why do I point this out?
If you’re looking for data-science-terminology references in someone’s experience from 2003, it’s going to be hard to find. That term wasn’t popular then. To identify skills needed today, you need to explore different keywords to find relevant work experience from 10 years ago that applies to data science today.
Here’s an example of high-performance computing and how terminology has evolved over the years.
High-performance computing began in 1998 with dual processors and various MIPS RISC microprocessors. That evolved in 2000 to high-performance computing clusters that matured into AlphaServers by Compaq, which provided a total capability of six teraflops (six trillion calculations a second). By 2005, computational sciences were excited about the potential of elastic compute and leveraging 7+ GFLOP clusters (seven-plus billion calculations per second). As 2011 approached, multiple technologies emerged from the latest quad-data-rate Infiniband to multiple cores of over 8,000 with a peak performance of 84.3 TFLOPs (84 trillion calculations per second). By 2012—if the technology wasn’t changing fast enough—the Joule Supercomputer surfaced, comprising 24,192 cores with a peak performance of 503 TFLOPs. And, just five years later, Joule 2.0 boasted 74,240 cores with a peak performance of 5,750.8 TFLOPS. Earlier this year, IBM built a supercomputer capable of 200 PFLOPs (200 quadrillion calculations per second).
For reference, a 2020 iPad can support about 5 TFLOPs (five trillion calculations a second).
What’s useful to extract from this review is that it’s unlikely you’ll know each technology handoff from year to year. That’s okay. Focus on how data was used to create value—this concept transcends time.
Locking in 10 years of data-science experience
Let’s assume you have to fill a data-science role on your growing team. If you’re looking for 10 years of high-performance computing, you’re not going to find it. Ten years ago, high-performance computing wasn’t common. It would be better to search for “big-data processing,” which started to get popular in May 2012. In fact, until May 2011, the concept of “big data” wasn’t even a household term.
To find an executive with a background in data science, hiring managers need to use terms that were cutting edge at the time. For example:
- Data warehousing in 2004
- ETL in 2005
- SaaS in 2006
- Business objects in 2007
- Cognos in 2008
- Apache Hadoop in 2009
- NoSQL in 2010
- Python in 2011
- Tableau in 2012
- Data analytics in 2013
- Artificial intelligence in 2014
- Serverless computing in 2015
- PySpark in 2016
- Augmented analytics in 2017
- Automation of data management in 2018
- Automation of data science in 2019
- Continuous intelligence in 2020
Note: The above dates are when technologies first because popular, not when they initially launched.
Using the right terms when reviewing candidate resumes will ensure that the leader you select has experience leading in cutting-edge innovations over time. To have breadth and depth in data science today, the candidate would have been focused on data warehousing 20 years ago not data science.
Data science from 2000 to 2020
To make finding your new data science leader easier, I pulled together a list of when technology terms became popular. This is also the most likely window when progressive leaders would have started to apply and leverage these technologies.
2000 to 2005
- Actionable analytics
- Call-center analytics
- Clustering methods
- Data acquisition
- Data engineering
- Data harvesting
- Data integration
- Data mart
- Data mining
- Data processing
- Data processing
- Data quality
- Data storage
- Data warehouse
- Decision science
- Distributed computing
- Enterprise decision management
- Image analysis
- Large datasets
- Parallel architecture
- Pattern recognition
- Predictive coding
- Relational database management systems (RDBMs)
- Speech recognition
- Transactional database
- Unsupervised algorithms
- Visual analytics
- Visual statistics
- Web analytics
2005 to 2010
- Business intelligence
- Business objects
- Cloud computing
- Cloud services
- Data analytics
- Data steward
- Elastic computing
- Exploratory data analysis
- High-performance computing (HPC)
- Network analytics
- Non-relational databases
- Operational data store (ODS)
- Predictive modeling
- Regression analysis
- Smart data
- Social analytics
- Software engineers
2010 to 2015
- Analytics platforms
- Apache Hadoop
- Apache Spark
- Artificial intelligence
- Augmented analytics
- Big data
- Causal inference
- Cognitive analytics
- Data analyst
- Data collection
- Data collection
- Data engineer
- Data science
- Data Science as a Service (DSaaS)
- Graphics-processing units (GPUs)
- Hybrid cloud
- Internet of Things
- Linear regression
- Machine learning
- Machine Learning as a Service (MLaaS)
- Natural-language processing
- Predictive analytics
- Quantum computing
- Sentiment analysis
- Structured data
- Unstructured data
2015 to 2020
- Actionable insights
- AWS Lambda
- Data automation
- Data platforms
- Deep neural network
- Descriptive analytics
- Pre-trained model
- Quantum computing
- Serverless computing
- Virtual reality
The unspoken dirty secret about data science
The most essential skill for a data science leader to be successful isn’t technical masterly. It’s relationship management.
The team needs to partner with business leaders outside technology to identify business cases. Without data, you have nothing to apply science and intelligence to. To get this, you need buy-in. That buy-in, credibility, and trust are earned through relationships.
If you’re a biotechnology company, IT partners with scientists. If you’re in the healthcare space, IT partners with physicians. If you’re in the retail space, IT partners with the store and regional managers. That store manager isn’t going to care that you know the nuances between Keras and Caffe. They’re going to care that you understand their business case.
It’s more important to build deep and meaningful relationships to share the data-science capability organizationally than hiring a whiz-bang technical team of introverts. This might sound obvious to some and backward to others. This is the difference between the theoretical application of data science and a practitioner with experience implementing it organizationally to change behavior and add new business insights.
Hiring data science leaders with data
A recent study by Heidrick and Struggles stated that a Ph.D. is neither a ticket-to-entry nor a required qualification for a data science leader, and being the smartest person in the room isn’t enough. Likewise, an MBA is considered a nice-to-have.
In this study, 16 years was the average time post-bachelor’s degree for a data-science leader to be impactful and have acquired the necessary commercial experience. The post-degree commercial data-science experience did vary based on the size of the company:
- In small companies (1-50 employees), the data-science leader had >5 years of experience
- Medium companies (51-10,000), 11-20 years of experience
- Large companies (>10,000 employees), 16+ years of experience
The study also noted that data science leaders with computer science and engineering degrees transitioned into the role at a faster pace than those with other academic qualifications.
Hiring for outcomes not for product knowledge
You may have heard that it’s important to have foundational knowledge to be an effective data science leader. I fully agree with this. Computer science, data technology, visualizations, statistics, and mathematics are foundational to data science. If you have a degree in computer science or engineering, you likely have experience in all five. This gives you a great base of knowledge.
Alternatively, leading data-science teams and building cross-functional, impactful, organizational relationships isn’t the same as being the technical expert. If the team is stuck on a technical problem, there’s a myriad of consultants that could swarm in to assist. Unfortunately, if you erode a relationship down to the nub, there isn’t much a consultant can do to rebuild that credibility and reengage that business partner.
Data science is about cross-functional teams leveraging data, processes, and algorithms to generate new insights. Can we ask questions we didn’t even know we should be asking? Do we have access to intelligence that’s now actionable? Are we able to generate predictive insights based on data trends?
Exceptional senior data science leaders can provide a stream of never-ending examples of how data was used to generate new capabilities and offer unique intelligence. Ask for examples of how data was used to create value.