Common Mistakes in Data Science Interviews

Read this article
⸻ Beyond Data

The most common reasons strong candidates get stuck in an interview are often easy to fix with the right focus

I lied!

Last week I promised to land the second part of the AotP Workshop write-up. I’m currently working through one with a customer and realise it needed updating a little. Once I’ve worked that through I’ll get that sent out.

Instead, this week we’re talking about

Interview Pitfalls

Now that we’re well and truly past the peak of the data and AI hype cycle, I can tell you hiring data scientists can be a painful and challenging process for both sides. Whether you’re a candidate trying to find the perfect role or an organisation seeking the right fit — there are plenty of pitfalls to watch out for.

For those getting into the role, there’s confusion about where to start. Trying to become the ‘Full-Stack Unicorn Data Scientist’ can feel like a task of mythical proportions. There’s just so much to learn. What’s worse, finding the right fit of skills for the role your going into is like playing Tetris set to invisible.

For organisations trying to find the right person, it can be just as overwhelming — especially if they’re going to be the first data scientist your organisation adds to the team.

While there are myriad paths to failing any hiring process, I’ve outlined some of the most common themes from my experience below.

Beyond understanding machine learning, statistics, programming, etc., there is a specialist skillset to be learned throughout your career as a data scientist.

You’re more than just a bag of technical tricks — knowing how to communicate and collaborate is paramount to your success.

The tech job market is more competitive than ever, and it should be no surprise for aspiring data scientists that even getting past the first hurdle can be challenging. It can be even more challenging if you don’t set yourself up for success. The advice below should help you avoid common pitfalls if you’re looking for a new job or a promotion at your current place.

If you’re looking to hire data scientists, you might find some of the points below useful to look out for.

So, what are the most common routes to failing the data science interview?

Machine Learning Buzzword Bingo

It’s all too tempting to tweak your resume to stand out. You read the job advert and see a few technologies you’re only tangentially familiar with. You think, No harm, I’ll brush up before the interview. This is a great way to pass the automated hiring filters used by recruiters and organisations, so you might think it’s a good step.

You need to remember, however, that many experienced data scientists will know how long these skills take to learn. Furthermore, they’ll want to know specifics and details.

It’s becoming ever more common to seek out specialists when building data science teams. I often speak about being T-shaped — have a broad knowledge of many tools and technologies but pick a topic or two and go really deep. This allows you to steer the interview towards your true strength.

It can become apparent quickly if you don’t honestly know your stuff. I’ve been in interviews where I’ve asked the candidate to draw up and talk through a high-level diagram of how CNNs work on the whiteboard, only to be met with shaky loose descriptions of the basics of neural networks.

You’ve wasted precious time you could have used to show your strengths and prove you can’t entirely be trusted.

Library Calls != Model Building

I’ve seen this so many times — a bright, talented candidate starts talking about all of their experience, and it seems reasonable at first, but as we start to dig a little deeper, answers to simple questions like:

  • What’s the difference between XGBoost and a Random Forest?
  • Can you explain the steps in building a decision tree?
  • How do you select the number of clusters when using K-Means Clustering?

Don’t seem to come quickly. Asking more profound questions about why to use one tool over another might even go unanswered.

The thing is, it’s so easy to load up the library or copy the notebook from Kaggle and get going that the aspiring data scientist can see some results without ever developing a proper understanding of what’s going on.

Now you don’t have to be able to derive the equations from scratch for every model you use.

As a rule of thumb, be able to draw up a diagram of the fundamental processes for the tools you use and talk your way through them confidently when asked.

Understanding how these models work will allow you to better spot when they’re suitable and when they’re not.

Model Building != Data Science

Machine learning is a big topic, so it might seem wise to spend much of your time and focus there. For many organisations, however, this can leave you without the broader skills required to make an impact.

It’s no secret that the vast majority of time in data science projects is spent on tasks other than machine learning. So it would help if you spent time building other technical skills, such as data wrangling, cloud computing, statistics, and analysis.

Designing suitable experiments and knowing how to test a hypothesis is more important than knowing which tools will get you the highest accuracy.

It’s worth your time to look outside the machine learning literature to some fundamentals.

Relating Technical Skills To Business Problems

This is often the most frustrating experience to encounter when hiring. A capable and talented individual who knows the ins and outs of technical skills gets stumped when talking about the commercial aspect of the role.

Like it or not, you’re there to add value to the business. Simply solving technical problems is often the first step. In a larger team, this can be compensated for by other members with more commercial understanding — but it’s still a red flag in many cases.

I’ve heard horror stories of companies having to let go of great data scientists on paper because their ego stopped them from even trying to engage with the business — staying lost in the clouds of the technical. There’s a certain intellectual snobbery that you see in many early-career data scientists that is always more harmful than good.

Take the time to understand how these tools and ideas can translate to real value.

Try to communicate everything in terms that matter to your business — cost savings and profits instead of accuracy and precision.

Specific Knowledge About the Organisation

I always find this one a shock as it’s so easy to get around. You get through the early stages of the interview, and it’s going well, but when asked about what a candidate knows, it becomes clear they’ve made a short skim of the website — and that’s it.

Many people in the tech sector (especially data) love talking about what they’ve done. So, if you’re going into an interview, try hard to find all the information you can on specific projects they’ve done. This will give you great insight into the skills and domains that are important to them.

Read through recent marketing material. Look for blog posts on the company site. Try to connect with the people that will be interviewing you.

I always recommend trying to find individuals on LinkedIn — to see if they have personal blogs or have recently done industry talks. This will make you stand out as having done your homework.

Final thoughts

Although this list isn’t exhaustive, these common themes often go unspotted. So keep in mind the following before your following interview:

  • Develop your fundamental skills outside of machine learning and understand the why.
  • Don’t inflate your application with tools you don’t have experience with.
  • Be able to confidently talk through the tools you use and focus on displaying a single strength over trying to cover as many areas of data science as possible.
  • Make an effort to understand the business context; read through materials outside the technical about companies that have seen great success from machine learning.
  • Seek specific information about what the company or individuals doing the interviews have done — it’s often more straightforward than you think.

Best of luck if you are searching — it can be tough out there!

All the best,

If you enjoyed this, consider sharing this newsletter with others that would find value from it or follow what we’re doing at Hypercube


When you’re ready, there are a few ways I can help you or your organisation:

What do you think of this issue?

I’d love your feedback to ensure I’m writing about the topics you want to read, let me know if this one hits the mark

Want to write for Beyond Data?

Have something you want to share? Interested in writing an issue or working with our team?