Data Science Hiring Theory

Data Science Hiring Theory

The ultimate process does not exist. Only compromises.


Share:          
  1. Interview Style
  2. Productivity and Knowledge
    1. Summary
  3. Graduate and Professional Productivity History
    1. Professional
    2. Graduates
    3. Summary
  4. Current Skill Set vs. Learning
    1. Lateral Transfers
    2. Summary

Interview Style

For starting a career as Data Scientist (or moving laterally, pivoting) you need to go through interviews such as for any other role. Either way, Data Science has its very own rules, pitfalls, and pre-conceptions that are hard to navigate. Sometimes, the process is more similar to Software Engineering, sometimes closer to Business.

Candidates/applicants want to know what is special about a Data Science interview. Hiring managers want to know which is the right or wrong way to interview.

Having had conversations with friends and family in the technology sector, linkedin.com, anonymous platforms such as reddit.com, and teamblind.com, I came to the conclusion that there is no universally right way to interview. This is super important for both hiring managers and candidates to understand.

Why? It is all about making compromises. There are no absolutes:

  • Shorter, more succinct interview processes introduce a higher risk of hiring the suboptimal candidate. Longer, more in-depth interview processes, introduce a higher risk of false negatives (i.e., rejecting a candidate who would actually be great for the role), and a higher risk of just rejecting qualified applicants.

  • A strong focus on quizzes / coding tests / quick reaction questions / take-home assignments can introduce a high risk of giving good scores to applicants who either prepared super well how to interview or simply got lucky and knew the answers to those specific questions. Focusing instead only on their work experience introduces a high risk of hiring someone who cannot improvise.

Productivity and Knowledge

Having said that, in my experience there are a couple of universal facts in a statistical sense that really help with interviewing:

A history of continuously being productive is with high likelihood an indicator for being productive in the future. Applicants with a history / résumé that lack productivity are unlikely to become productive in a new role.

This is, of course, a relatively sweeping judgement, I know. Let me explain. Many candidates look great on paper, had the right classes, know the right things, talk convincingly about their abilities… – but for some reason, when they get hired, would struggle to get things done. Everything would take too long, people overthink things, don’t know when to ask for help, etc.

And that is when the red flags showed up (retroactively): The candidates knew all the facts, but had never produced output. Whether in grad school (publications, projects,…) or at work. The ratio of productivity/knowledge is out of balance.

According to some people, this is mostly an inherent trait. Some candidates just find ways to get things done, and other candidates find ways to NOT get things done. Many people believe that there is very little you can do as a manager to change that in a person.

This leads to the realization that if a candidate who aces every part of the interview but has nothing to show for in his previous roles becomes a concern - unless they have an explanation as to why. Most of the time there is no satisfactory reason why.

Summary

Hiring managers:

Focus on evidence of productivity in candidates.

Candidate:

Make your résumé scream “I will get stuff done”. Achievements outweigh knowledge.


Christian Haller phone call

Photo by @magnetme on Unsplash

Graduate and Professional Productivity History

Everyone is going to be compared against a benchmark, but the benchmarks are different. Someone fresh out of a Masters degree is going to get graded on a different scale than someone with two years of experience. But either way, the candidate needs to look productive relative to their peers.

Candidates do not need to provide excessive detail to show they were productive. Ideally, a candidate can write/say something along the lines of “Delivered $2M in revenue by implementing a [some ML] model in [language]”. If a professional candidate is not allowed to disclose those details, they can remove the quantities and keep it vaguer. What’s critical is that they provide a cumulative list of projects / impact that looks compelling. If the candidate has worked for an employer for two years, they need to either have one to two really impressive accomplishments, or four to six more moderate/minor ones. They need a combination of volume and impact that stands out.

Professional

Something candidates often miss: it is not good practice to combine or aggregate multiple projects (e.g., “Delivered $[x]M in value across several projects”). Instead, listing every single project is likely to be more successful. Do not say/write things such as “Collaborated on several cross-department initiatives”. List every single one and what was accomplished in them – even if minor.

For example, compare these two lists:

  • Delivered $[x]M in revenue as part of a multi-year, strategic plan to overhaul sales analytics process.

vs.

  • Generated $[xxx]K in costs savings by optimizing ad spend across channels using a linear programming approach.

  • Identified $[xxx]K in additional revenue opportunities through targeted price increases.

  • Mitigated COGS increases of [x]% by consolidating volume across multiple brands.

  • Helped leadership identify $[xx]K in redundancies by providing ad-hoc analysis to identify redundancies.

What would you prefer as hiring manager? Conversations with DS hiring managers yielded that they are almost surely going to prefer the second list, because it tells them the candidate has generated value in many different ways and times. That means more evidence that proves “I can accomplish things”.

Graduates

As a fresh graduate, it is critical is to make sure to cover all papers, reports, documents, projects, etc. Descriptions to avoid are:

  • Conducted cross-functional research in the area of [something]

And instead break that down into details:

  • Performed literature review (>20 manuscripts) in [something] design, [type of] algorithms and [something] optimization.

  • Conducted weekly research meetings with researchers in the Department of [something].

  • Prototyped novel model using combination of [language] and [framework].

  • Delivered monthly status updates to advisor.

Summary

Focus on detailed accomplishments and results. If possible, provide numbers.


Christian Haller code

Photo by @casparrubin on Unsplash

Current Skill Set vs. Learning

While it is a platitude, it is still a universal truth:

“The only constant in workplaces is change.”

For many hiring managers very high on the priority list is finding out if someone can learn exactly those things quickly they will need them to do now and in the future.

Other hiring managers ask themselves “What would this person do today?”, and then focus on finding people who are doing exactly those things today.

If the applicant pool is large (e.g., the manager needs them to code in [language]), then it is a reasonable approach to be very specific. However, if the candidate pool is small (e.g., they need people with experience in one specific algorithm), then things become difficult. This is because the odds of finding someone who is both an excellent interviewee and has that exact experience is miniscule. This is where the search principles come back to the productivity history screening. Hiring managers should put more emphasis on finding productive people than finding people with the exact experience they want.

Why?

Experience can be acquired. People can learn. Not only that, what they may have to do today may not be what they will be doing half a year in the future. Perhaps that new employee comes in and they themselves are able to suggest a new approach that works better than what was done before, rendering that skill set obsolete.

A candidates’ current skill set is largely dictated by their current job, and should not be taken as a fixed, static skill set.

Lateral Transfers

When hiring managers consider candidates that have been in a single environment for several years, then they look at the résumé and assume that what the candidate is doing now is just who they are. “Well, this applicant has done a lot of reporting as Data Analyst - not what I we are looking for. We need a Data Scientist building [some type of] models.”

And that is the wrong approach. Because what all this indicates is that candidate is doing reports, which is what they need to do in order to do their current job well.

So, what can a hiring manager do with a candidate that has done mostly Data Analysis-work but wants a position as Data Scientist training models?

Answer:

  • Measure productivity. If the applicant’s previous job was data analytics reporting, then how successful were they at that? Did they differentiate themselves in that field? Did they go above and beyond what others would have done in the role?

  • What experience does the applicant have with modeling - not in this previous role, any time before? How well can they describe that experience? How willing are they to return to it?

  • What experience does the applicant have picking up new skills? In their current role, did they have to learn new technologies or languages?

Summary

Hiring managers can take the risk-averse position and not even entertain the thought of hiring someone without the exact résumé and experience they are looking for. But they are overwhelmingly likely to miss out on some great candidates.


© 2023. All rights reserved. Hosted on GitHub, made with https://hydejack.com/