Office Hours
If you’re new to hiring data scientists, it’s likely you’re terrible at it. Hiring is a huge responsibility and easily one of the most important things you do as a manager. Even a single bad hire can drag down your team’s productivity, kill morale, decrease team retention, and generate a significant amount of work (and headache) for you as a manager. From the job description to the interview to the offer, this series of posts is intended to memorialize hard won wisdom in hiring data scientists over the last few years, focusing on how to enhance the candidate interview process and get less…you know…terrible,
N.b., these are my opinions and not representative of those of my employer. This (these) posts are simply to memorialize some of the things that I’ve learned in an informative manner. While I hope you find them useful, they’re not official statements of any kind.
Authors: Eli Goldberg, PhD, MSc, Director of Data Science in Clinical Analytics, Analytics and Behavior change at CVS-Aetna Benjamin Goldberg, PhD, Instructor in Humanities and Cultural Studies at USF, currently trapped in Bulgaria on a Fulbright.
If you’re new to hiring data scientists, it’s likely you’re terrible at it. Hiring is a huge responsibility and easily one of the most important things you do as a manager. Even a single bad hire can drag down your team’s productivity, kill morale, decrease team retention, and generate a significant amount of work (and headache) for you as a manager. From the job description to the interview to the offer, this series of posts is intended to memorialize hard won wisdom in hiring data scientists over the last few years, focusing on how to enhance the candidate interview process and get less…you know…terrible.
Here’s some organization for these articles. I’ve bolded the part you’re on for convenience.
Part 1. Creating better job descriptions brings in better talent.
Part 2. Create a clear interviewing process.
Part 3. Good interviews have structure.
Part 4. Good companies respect their candidates time.
Part 5. Don’t get emotional about compensation.
Part 1.Creating better job descriptions brings in better talent.
A job description is often a candidates first exposure to your company and the role. If it’s bad, you’re going to attract the desperate or those that fail to read for detail. However, if you create job descriptions that position the impact the role can have on the company/world, and clearly set expectations for your candidates on what you’re looking for/need, you start your discussion with a new candidate at a significant advantage – they’ve already bought into your vision.
Data scientists are useful creatures within any institution, but their role and job descriptions are often poorly defined. Thus, job descriptions for data scientists often come across vague, filled with red flag tropes like, ‘work hard, play hard’, or may even advertise your ignorance in the field (e.g., asking for more years experience in a software/package than is technically possible).
As an aside, here’s a great article from thinkful summarizing the differences between software engineering and data science (data science vs software engineering). We’ll talk more about how to evaluate a great data scientist when we get to Part 3: Good Interviews Have Structure, which will help you design and structure the interview to suss out data these skills.
Assuming you’ve convinced yourself that you need a data scientist, here’s a common sense series of steps to consider when crafting a job description.
Step 1: Establish the ‘why you’. Make people care about you, your company, and your mission.
Step 2. Stop describing responsibilities. Start describing opportunities.
Step 3. Establish the ‘what’ by articulating key actions of the role and background experience needed, not technologies.
Step 4: Describe your company without sounding pretentious.
Step 5: Proofread and be a human.
I’ll provide an example for each step below.
Step 1: Establish the ‘why you’. Make people care about you, your company, and your mission.
If you want to attract (and keep) high performing data scientists, you need to make people care about what you’re doing in a way that speaks to their fundamental motivations. I have observed that most strong data scientists are driven by their motivation to make large-scale impact, to innovate/create something new, and to accelerate science.
By definition, data scientists are interested in solving hard problems with smart people. What this means is that if your company is not known for data science, or innovation, or solving hard problems, you need to clearly establish why a data scientist would want to work for you.
Here’s a basic template for how to overcome the ‘why’ question:
- Lead by selling the large-scale impact they could make.
- Align the business goals with the need for data science.
- Position your strategic advantages.
- Set high expectations using the accolades/accomplishments of the team.
Here’s an example from a position I’m currently trying to fill that I think works nicely.
Do things that matter. Diabetes, congestive heart failure, COPD, back pain, asthma, pulmonary fibrosis, chronic kidney disease, Alzheimer’s, multiple sclerosis, cancer… everyone knows someone affected by these chronic conditions. Our goal is to decrease cost and improve health for members with chronic disease. We do this by using data science and advanced analytics (e.g., condition-specific predictive models and clinical data mining) to improve their care journeys and outcomes, and bend the business case to favor the right members getting the right interventions at the right time.
Mission aligned to improve health. With +9,900 stores, >1,100 minute clinics, 1,100 health hubs, and millions of American insured lives, more than 80% of Americans live within 5 miles of a CVS. Together CVS-Aetna is an unparalleled platform to make a significant impact in healthcare in the US and beyond. https://cvshealth.com/about/facts-and-company-information
We have actual big data. The combined data assets of CVS and Aetna are massive. Claims. Pharmacy. Health care records. Commercial store and transactional data. Challenge yourself to innovate in ways that others simply cannot.
An elite team. You’ll be joining a team of MDs, PhDs, MBAs, and MSc from top tier universities and industries that bring their clinical, technical, and economic experience to bear on some of the toughest targets in healthcare – chronic disease.
Step 2. Stop describing responsibilities. Start describing opportunities.
Many ‘old school’ managers make the mistake of leading with a role’s responsibilities instead of a role’s opportunities. Position the opportunity and what you’re looking for in loose enough terms to give the role flexibility, but not so loose as to be unclear. Here’s an example:
Our group’s core competency is data science. However, we’re looking for people with additional talent and experience in 3 key spheres:
Healthcare data wizards. These magical folks help translate clinical hypotheses into data and interventions. We’re looking for data scientists with deep experience working with claims/healthcare data – particularly those that have an understanding of its implications, and its scientific and/or clinical value.
Client-service orientated chimeras. You must wear a lot of ‘hats’ to make a product work. You excel in translating business and stakeholder needs into data science direction, and vice versa. You have great interpersonal skills and an ability to reduce, and then present, complex business/data science concepts into digestible components. You care deeply about credible measurement and experimental design, and you are old hat at developing business cases and presenting to executive leadership.
Research and Development data scientists. You have deep experience architecting and evaluating cutting edge mechanistic and empirical models. They need to have deep experience coding and developing pipelines, in coordination with our other spheres and our data engineering colleagues.
An important note on the ^^: The worst thing you can do here is oversell, if you’re looking for a data scientist level 7 wizard to take command of the corporate scientific strategy – great. However, if what you actually need is someone to write SQL queries you’re going to set yourself up for failure. These days, there’s no shortage of websites where people can ‘yelp’ your company/interview/you. Not everyone wants a grandiose role. Those that do, are often unsatisfied with anything else, meaning that you’re likely going to be looking for a new person sooner than you might otherwise.
Step 3. Establish the ‘what’ by articulating key actions of the role and background experience needed, not technologies.
After you’ve established the opportunity, follow-up with some fundamental actions that someone successful in the role will take. The specific actions, obviously, depend on the role. However, here’s a general list of ‘actions’ you should answer in the ‘what’ section.
-
Be specific – what will they do? …Try to tie this into the mission. Are they doing causal inference in python, or are they using a massive claims dataset and causal inference to estimate the subjective ex ante treatment effect of an intervention on members health? One of those sounds cool (at least to me).
-
With whom, and with what internal/external teams, do they collaborate and interact? …If your candidate is intimidated by an elite team, then they’re not right for the job. However, make sure they know that there are others who will help them onboard, get up to speed, and innovate.
-
How will their key skills be used in the role? What are the skills that they’ll need to affect change internaly? … be realistic about your stack, but technologies change rapidly. Unless you’re bound to a technology for some reason (which can happen, particularly in large institutions), articulate what you’re looking for without specifically referencing technology.
There’s an ML joke embedded in here somewhere, but someone’s past performance is likely indicative of their future performance. Similarly to (3) above, you want to clearly articulate the background experience you’re looking for without specifically referencing technology. Technology can be learned (and often needs to at a new institution, regardless). Here’s an example:
Background Experience The complex chronic care group within clinical analytics employs cutting-edge ML, signal analysis, and time-series modeling to move the needle on healthcare using claims, commercial pharmacy, and front-end store data. As a principal data scientist supporting a portfolio of chronic disease oriented internal products, you spike hard in both client-service and research and development spheres.
Specifically, you: …Are an expert (+5 years) in experimental design, causal and empirical modeling methods, model explainers, and healthcare data, and …Have 5 or more years of technical data science leadership, focusing on application of ML to large datasets for healthcare or similar applications, and …Have deep (+3 years) clinical and research understanding of chronic diseases and their and associated disorders and comorbidities, and …Have deep (+2 years) technical experience in healthcare business case development, and med cost savings estimation, and …Have strong (+2 years) consultatory, clinical and commercial stakeholder management, and leadership skills, and …Are passionate about helping patients by integrating data science research into application.
Step 4: Describe your company without sounding pretentious.
If they’ve read this far (or you for that matter), you have an opportunity to position your companies goals in a way that reinforces everything you’ve laid out above. Here’s an example:
At Aetna, a CVS Health company, we are joined in a common purpose: helping people on their path to better health. We are working to transform health care through innovations that make quality care more accessible, easier to use, less expensive, and patient-focused. Working together and organizing around the individual, we are pioneering a new approach to total health that puts people at the heart.
Ok…so this is a bit pretentious, but it’s healthcare and some excuses should be made. That said, if you’re looking for a data scientist to help you sell cars, you might consider dropping words like ‘passion’ in favor of ‘impact’ and other relatable factors (improve a frustrating experience, help those in need, etc…).
Step 5. Proofread and don’t forget to be a human.
Finally, if you want people who are passionate about something (e.g., patient healthcare, experimental design, coffee, whatever), SAY IT. Some of the best conversations I’ve had with candidates have come after they’ve expressed that this role ‘spoke to them’ because you took 20 seconds to articulate that you want a real human to work with, not data science terminator…
Also, proofread. In making this article, I found spelling/copy-paste/grammar errors in my own job descriptions…so yeah. #irony.
Summary:
Writing a job description is hard. Make sure you really need a data scientist. If you do, follow these steps when drafting a job description.
Step 1: Establish the ‘why you’. Make people care about you, your company, and your mission.
Step 2. Stop describing responsibilities. Start describing opportunities.
Step 3. Establish the ‘what’ by articulating key actions of the role and background experience needed, not technologies.
Step 4: Describe your company without sounding pretentious.
Step 5: Proofread and be a human.
If you’re interested in having me proof read your job description, let me know! Feel free to reach out to me on LinkedIn (https://www.linkedin.com/in/eli-goldberg/).
.
.
.
