With the development and adoption of AI accelerating at an unprecedented pace, the demand for data scientists and data skills in general is snowballing. To recruit successfully in this competitive talent market, TA teams must understand the specific data skills their organisation needs and the intrinsic career motivations of the data professionals they seek to engage with.
My guest this week is Akshay Swaminathan, Head of Data Science at Cerebral and co-author of a new book called “Winning With Data Science”. In our conversation, Akshay offers an insider view on the data science talent market, what motivates data scientists to move companies and the most effective recruiting process for people with this skillset. Essential listening for anyone recruiting data science professionals or working in talent markets with similar skills shortages.
In the interview, we discuss:
• The growing demand for data science professionals
• The specific data skills that employers currently need
• Is there any consistency in data science job titles?
• What would motivate a data science professional to move jobs?
• How should recruiters source and engage in this talent market
• What type of hiring process will drive the most successful outcomes?
• Gig working and fractional working
• Creating business value from AI
• What skills will be needed in the future as the development and adoption of AI ramps up?
Follow this podcast on Apple Podcasts.
Transcript:
Matt: Support for this podcast comes from Harver, the industry leading hiring solution helping organizations optimize their talent decisions. Rooted in over 35 years of rich data insights, backed by I-O psychology and cognitive science, Harver delivers a suite of automated solutions that enable organizations to engage, hire, and develop the right talent in a fast and fundamentally less biased way. Visit harver.com to learn how you can take the smart path to the right talent.
[Recruiting Future theme]
Matt: Hi there, this is Matt Alder. Welcome to Episode 594 of Recruiting Future. With the development and adoption of AI accelerating at an unprecedented pace, the demand for data scientists and data skills in general is snowballing. To recruit successfully in this competitive talent market, TA teams must understand specific data skills, their organization needs, and the intrinsic career motivations of the data professionals they seek to engage with. My guest this week is Akshay Swaminathan, Head of Data Science at Cerebral and co-author of a new book called Winning with Data Science. In our conversation, Akshay offers an insider’s view on the data science talent market, what motivates data scientists to move companies, and the most effective recruiting processes for people with this skill set? Essential listening for anyone recruiting data science professionals or working in talent markets with similar skill shortages.
Matt: Hi Akshay, and welcome to the podcast.
Akshay: Hi Matt, thanks for having me.
Matt: An absolute pleasure to have you on the show. Please, could you introduce yourself and tell everyone what you do?
Akshay: Sure. My name is Akshay Swaminathan. I’m a data scientist. I work on strengthening health systems. So, these days I wear two hats. One hat is as Head of Data Science at Cerebral, which is a national tele mental health company. And our mission is to democratize access to high quality, long-term mental health care. And my role there? I use data and data science to improve the lives of both our clinicians and our patients. And then my other hat is as a researcher and medical student at Stanford, where I work on deploying AI systems within hospitals and evaluating them and measuring their impact.
Matt: Fantastic. That sounds like some amazing work that you’re doing there. On top of all of that, you’ve also recently written a book. Tell us about the book, what it’s about and why you wrote it.
Akshay: Sure. The book is called, Winning with Data Science: A Handbook for Business Leaders. And this stemmed out of the experiences of myself and my co-author, Howard Friedman. And you know as data scientists, we’ve worked a lot with nontechnical folks. Business leaders, domain experts and what we found, this observation we had is that there are a lot of resources out there for people who want to become data scientists, for people who want to learn the technical stuff. There aren’t as many resources out there for people work with data scientists, but don’t do the technical work themselves. And what people don’t often appreciate is that it’s those folks that can make or break the success of data-driven projects.
Oftentimes, a data savvy business leader who can speak the language can save data teams weeks, months of work even just by helping us refine our question, asking the right questions, questioning our assumptions, challenging our approach. Both of us have had experiences where we worked with, say, product managers or marketing folks or clinicians who are able to really push us and use their domain expertise effectively to come together with the data teams. And so, we wanted to write this book for all those people out there who may not be data scientists themselves, but who work regularly with data teams, for people who have to make decisions using data to show them what does an effective collaboration look like between a business person, a nontechnical person, and a data person? And how can you both come together to get the most value possible out of data?
Matt: So, obviously, for a number of years now, data scientists have been in high demand for many, many organizations. Will the need for data science skills continue to grow as we move forward from here?
Akshay: It’s certainly going to continue to grow as companies are realizing that data is a valuable asset, not only for the purposes of helping them understand their own operations. And I often say data is a great way to check your intuition. And sometimes data can prove your intuition wrong. So, if you have some insight about, “Oh, I think this unit of my business is performing particularly well but then you look at the data, turns out that’s your lowest performing business unit, or maybe your most costly business unit.” So, I think more and more businesses are using data to gain insight into internal operations. But now what we’re seeing more and more is that data is becoming an asset in and of itself.
So simply having, if you’re able as a business, to set up your own proprietary database, where you’re collecting data on maybe your clients or the way people are using your product or your service, you can then use that data, one as a product on its own, if you can create a high-quality data set that maybe others are interested in using, but now you can also use it to train and fine tune AI models. And once you have an AI model that’s fine tuned on your own data, you basically have– it’s almost like the sky is your limit. If you have enough high-quality data, maybe you’ll be able to train an AI agent to perform the work of a number of humans or make your humans more effective. And so those are kind of the two use cases where data is going to continue to become more important in the business setting.
Matt: And what are the particular skills within that that are kind of most in demand? What are the sorts of precise skills that companies need to be able to achieve those kinds of things?
Akshay: Yeah. So, I like to think of it in three levels. If you’re a level one company, maybe you have some data, but data is not really a big part of how you run your operations. Maybe you don’t even have mechanisms to collect data at scale. So, for that level one data driven company, you want to invest in people who can help you start collecting data, who know the right questions to ask. And for this level, you don’t need highly technical folks. You just need someone who understands the basics of database management. Maybe you’re working with low-code or no-code tools like, say, Airtable to just start collecting data. Once you get into level 2, maybe you have a lot of data and you need to figure out a way to extract actionable insights from it.
So, this is where you need to start getting into the world of data analytics and analytics engineering. So analytics engineering is all about setting up the data such that it’s ready to be used for analytics. So, things like setting up a data lake, doing data modeling, and writing extract transform load code that can move data around from one source to another, and then surface it to a BI tool, a business intelligence tool, like a Power BI or a Looker. So those are kind of the tools that become important at that level 2 stage, and then at level 3 stage, that’s the stage where maybe you’re building models, you’re deploying models, you’re using data to fine tune models and continually iteratively improving the data-driven assets you already have.
And at that stage, you need people with expertise in data engineering, ML engineering. Maybe you need data scientists who are more knowledgeable about statistical methods, understanding what types of models to build using what data. But I think one of the most crucial and underappreciated skills is the skill set, kind of what I like to call translational data science or implementation science. So, if you have a model, how do you make sure that that model is getting used by whoever you need on the business side to actually use it? I mean, we see this problem a lot in healthcare that models get built, but no one actually uses them because it’s hard to get clinicians to change their behavior and start using these tools. So, I want to emphasize that don’t neglect the implementation side of things, which is not often a technical person. It can be a nontechnical person that can speak the language of data science, but also understand how people will respond when you ask them to, “Hey, I need you to change your workflow to start using this model or this tool now.”
Matt: And is there any consistency in job titles across all those sorts of different levels and skills, or do people sort of just generically call themselves data scientists?
Akshay: So, there are some titles that have a generally well-understood definition. So, I’d say data analyst is a pretty well-understood definition. If you see a JD for data analyst, you’re thinking, okay, they’re mainly working in SQL, they’re writing queries to extract data from a database. They’re maybe doing some dashboard building. Data engineer, I’d say, is also generally well defined. So data engineer, data architect, ML engineer. So, these are the folks that are basically in charge of the infrastructure that’s required to build, deploy models, the infrastructure required to bring in data, transform that data, and then surface it to the data analysts. The one title that is the most vague is data scientist, because there are so many things that could fall under the umbrella of data scientists. You could have someone doing observational causal inference, you could have someone building machine learning models, you could have someone doing data visualization. All those things fall under data science. So, if you have a job title that’s called data science, you have to look at the skills section, required skills. And so that’s where I think it’s fine that we use the title data science, but let’s be specific about the competencies. So, I think that would be my 2 cents there. If you’re using a vague umbrella term like data science, at least be specific about the skills.
Matt: What would motivate? Obviously, we’re talking about skill sets that are in high demand at the moment. Many companies wanting to recruit people with these skills and experiences. What would motivate a data science professional to move jobs? What’s the kind of intrinsic motivation that may be common across the sector?
Akshay: So, I think of three reasons here why someone might be looking for a new role. So, the first is technical growth and technical challenges. So, if I’m a data scientist and I’m really interested in growing my technical chops, expanding my methodological toolkit, maybe developing new methods, I’m interested in unsolved problems. I’m interested in, for example, if a recruiter told me, “Hey, we have this interesting problem. We have this type of data that’s coming in from this source, and we needed to do this task. But because of this bias in the data, or because of this way the data is collected, it makes it challenging for us to do that task, and we’re trying to develop methods to do it.” I mean, this was a common thing in my first role at Flatiron Health, where we collected data from patients undergoing treatment for cancer. In parallel, we had data that was used basically to give them genetic tests. And it turned out that merging those data sets and using those two data sets together to generate insights was not straightforward to do.
And so, there were a lot of interesting statistical methodological questions around, how do you do this robustly? And so that’s one type of thing that can attract people to a new role, unsolved problems. And so, the opportunity to innovate, build new methods, maybe even write papers if it’s more of a research role, writing tools to do that sort of stuff. The second thing I’d say is scale. So, data science is so powerful because you can use data to understand trends in large populations, you can use it to deploy technologies that affect huge numbers of people. And so, the scale of your data is massive, if you’re a huge company that has a huge database because you have a ton of customers or you just developed a really large data set that no one else has, that can be a huge draw for a data scientist. It’s kind of like a kid in a candy store phenomenon where the more data and the more interesting data that we have access to, there’s more cool stuff that we can do. So, the scale certainly is one factor.
The other thing is the scale of impact. So maybe you have a huge client base. And this was sort of my draw to Cerebral, where when I joined, we had close to over 100,000, 150,000 patients in our network, and we were able to deploy data-driven products that impacted all of those patients. And that was a huge draw for me just because of the sheer scale of that impact. And then the third thing I’d say is the team. So, some data scientists are looking for an environment where they can really be challenged technically. Maybe you have a team that is really high performing and they have a track record of great output, like working on great products, putting out really innovative, maybe research papers, or putting out really cutting-edge data products. Or maybe you have a team structure where you have a data scientist that gets to work with a ton of other units of the business. Maybe you have a data scientist working with doctors, working with product managers, working with designers and engineers, having that cross functional team dynamic. That can also be something that data science are looking for. So just to summarize those three things, the first one, technical challenges, the second one, scale, either scale of data or the scale of impact, and then the third is the team.
Matt: And I suppose leading on from that, how should recruiters be approaching data scientists? How should they find them, market to them, and actually sort of communicate with them?
Akshay: I think coming back to these three things that I mentioned, so there’s a stuff that would work for any sort of job recruitment process. You want a competitive benefits package, competitive compensation, all those things go. But I think starting, for example, if I think about what emails am I most likely to open or what messages am I most likely– I see it and I’m like, “Oh, this is interesting.” It’s one of those three things that I mentioned. So, an email that said, “Hey, Akshay, we have a data set of 5 million patients and their health records, and we’re trying to use it to improve child mortality rates or something. That’s a really interesting opportunity for me.” Another one relating to the team, “Hey, Akshay, our team just put out three models that are now being used by,–” I keep giving examples in healthcare because that’s my field, but we just deployed three models that are being used by doctors all over California help us on our mission or something like that.
So, something about the impact of the work and the technical track record of the team. That’s another one. And then the first one, the problem, so, I mean, OpenAI, I think is doing a great job of this. Right now, they’re working on the problem of super alignment. So alignment, basically aligning. How do we get AI agents to align with the goals of humans and human values. So, their whole call to action is help us solve super alignment. And that’s a really interesting unsolved problem for computer scientists and data scientists. So, I would say think of ways to add one of those three pull factors into any of your messaging.
Matt: That makes perfect sense. And in terms of the actual hiring process, what type of hiring process is likely to lead to successful outcomes?
Akshay: I think it’s a process that closely aligns with the type of work that candidates will be doing on the job. I personally am not a fan of hiring processes that ask these sorts of crazy brain teaser questions because it’s just not reflective of the type of– It’s not informative, as a hiring manager, I want to know how are these people going to perform on the types of tasks that I’m going to assign them to do? And so, we take that approach at Cerebral where we give them a take home assignment, which is reflective of the type of work that they would have to do on the job here. And then during the technical interview, again, we try to align it very closely to the types of tasks that they would do on the job. And then we also try to make– Cerebral is a cross functional place where you’re working with not just data scientists, but people from other domains.
So in some of our interviews, we’ll even invite panelists from other parts of the business, product managers, etc., have them do some sort of presentation or some sort of exercise where they’re forced to communicate with all these different stakeholders. So that would be my advice there. And then I think part of this is, the recruiter can play a really important role in this process, even though they’re non-technical and this is especially true at Cerebral. You need your data scientists to be able to communicate effectively to nontechnical people. And so, part of the process is, can this candidate compellingly and clearly explain what their work is about to our recruiter. And if the recruiter feels, “Hey, they said a lot of stuff, it kind of all went over my head.” That’s sort of a red flag to us that, “Hey, maybe this person is not the best communicator.” So, I’d say the recruiter also plays an important role here.
Matt: And is this an industry where you are seeing different ways of working come in? So, gig working, fractional working, different ways of thinking about work to try and deal with some of the skill shortages?
Akshay: Yeah. I think gig work or contracting roles have always played a big part in data science. I think especially now with AI, there’s going to be more and more of these types of opportunities come up, whether it’s helping an organization deploy their first large language model powered application, or whether it’s consulting with organizations to help them understand, “Okay, how can I use AI to improve efficiency in my company?” So, I think those things will continue to be a big part of it. I think also training and upskilling people within your company is also going to be an important piece of this. So, I think these days, everyone needs to have a basic understanding of prompt engineering. This for really everyone in the business– almost everyone can use an LLM to make their lives easier and to make them more effective at their jobs. So, I think everyone needs to understand the basics of prompt engineering as one example. So, I think upskilling is also a big piece of it.
Matt: You’ve sort of talked about AI and the things that OpenAI problems that OpenAI are working on. How is this likely to develop in the future? What are the skills of the future? How is AI going to sort of change and develop everything?
Akshay: So, you know, there’s a stuff that a lot of people are talking about. Now, there are even job postings for, you can see JD is out there for prompt engineer, and there are so many startups out there that are sort of building wrappers around large language models like GPT to accomplish certain tasks. But what I think is going to be really necessary and is going to distinguish the winners from the rest, it’s two things. So, one, can you effectively leverage your own internal data for improving your own AI system? So, leveraging your own internal data, can you build up a proprietary database of high-quality data that can be used to fine tune and maybe even train your own language models? That’s going to be a huge piece of this.
I’ve had many conversations with companies that have started out using OpenAI’s, APIs, but soon as they’re able to collect and curate their own data sets. They don’t need to rely on OpenAI anymore because they have enough data that they can train and build their own models. So, it’s really not the model itself that’s going to be the distinguishing factor. It’s going to be the data and the quality of the data. So, if you as a company can get the right team in place to curate and develop your own proprietary database, that’s going to be huge. And then the second piece is, can you successfully deploy AI applications in a way that generates business value? There are a lot of cool applications, a lot of interesting applications, but it remains to be seen how many of these are truly valuable. Some of them clearly are.
But I’ll tell you, in the healthcare setting at least, we need a lot more research to understand whether, say, giving clinicians access to AI, that helps them write their notes. How does that actually impact the business? Does it lead to clinicians? Does it actually reduce clinician burnout? Does it help them write more notes? Does it help them see more patients? So, it’s certainly a cool application of AI, but we need people and we need teams who are equipped to study these questions, basically think critically about what are the success metrics here. So, it’s not enough to just deploy some LLM application and say, “Hey, we launched this thing.” You need to actually measure the impact of it and you need to measure the right thing. The worst thing is you roll something out and you’re measuring the wrong outcome. So, I think that skill set which I talked about before, this implementation science skill set. Are you measuring the right thing? Are you training the right people? Are you designing your pilot or you’re A/B test in the right way? That’s a skill set that I think is very underappreciated but will become increasingly important.
Matt: Lastly, where can people find the book?
Akshay: To find the book, to hear more about everything we talked about, you can go to winningwithdatascience.com.
Matt: Akshay, thank you very much for talking to me.
Akshay: Thank you, Matt. Great to be on the show. Thanks for having me.
Matt: My thanks to Akshay. You can subscribe to this podcast on Apple Podcasts, on Spotify, or via your podcasting app of choice. Please also follow the show on Instagram. You can find us by searching for @recruitingfuture. You can search all the past episodes at recruitingfuture.com. On that site, you can also subscribe to our monthly newsletter, Recruiting Future Feast, and get the inside track about everything that’s coming up on the show. Thanks very much for listening. I’ll be back next time and I hope you’ll join me.