Nail Your Interview and Get a Data Scientist Job: Be the Anomaly
This is not a blog post on improving your job search process (read this instead).
This is a blog post on what we (my colleagues Joshua Loong and Lewis Davies and I, as hiring managers for advanced analytics at Best Buy) are looking for in an applied data scientist. If this helps you land a job, great. A more fulfilling career, amazing. But full transparency: I’m writing this with the hope of improving the quality of candidates in our recruiting pipeline.
“Looking for a data scientist that’s a solid average.”
Have you ever heard someone say that? No.
We all hope to find top anomalies. That’s the case in both my experience recruiting for data scientists and analysts at RJMetrics (a business intelligence software startup acquired by Magento/Adobe) and at Best Buy Canada (a F500 omni-channel retailer). We’re looking for a needle in the haystack. Someone that is in the top 95 percentile among peers.
But what is the KPI? What is “Good?” What defines a top candidate among the sea of people that write python, can clean data, run regressions, and develop apps?
In the context of recruiting for an applied data scientist capable of solving real business problems using data (think automation, optimization), here are three abilities that my peers and I are really impressed by:
“I am one of the only data scientists that can…”
#1 Generate my own PR
A unicorn data scientist should be able to articulate the business problem, how it aligns with organizational or strategic goals, why the solution proposed is the best, and the expected impact. All in laymen terms. And not rely on product managers to manage stakeholders and clients,
We’ve interviewed many analytics professionals who can take technical requirements and translate it into analyses and models, but never ask themselves why they’re doing it. Why is it important to build a forecast model for financial budget this year, as opposed to forecasting salesperson performance?
We believe data scientists deserve a say in what they work on: Prioritize projects and tasks that bring the most value to the organization. That starts by understanding why we’re asked to tackle a given initiative. It ensures we’re working toward projects with the most value (revenue or cost saving potential). What does success look like? The CEO is constantly asking you about new ways to insert data science to boost the bottom line.
We’ve also come across data scientists that just can’t speak in laymen terms, resulting in stakeholders not understanding how we work. It’s especially frustrating for our bosses when deadlines are missed because new discoveries need important model changes, but they don’t get why.
It’s much easier to push back on unfeasible requirements and timelines if you can communicate with executives directly. Explain why X problem is too hard to model, the wrong thing to do, or why Y approach may take more time.
So, when the business asks to create an app to personalize web experience, let’s first ask why:
- Do a root cause analysis
- Paint a picture of the problem
- Outline the impacts on the business
- Explain why solving it will help us hit our strategic priorities
- Then draft a solution, and estimate the impact it’ll have on the KPI (Try this Amazon Press Release exercise)
In other words, create an executive summary like this mad lib:
“Our business faces a challenge where __________________, so we’re developing a solution that _______, which we expect will have an impact of ____. We used the following advanced analytics methods and modern analytics tools____, including components and infrastructure that are now reusable for ______. The old way was___ . The new way is better because____.”
Don’t be surprised if you’re tapped to lead more and more data science projects, and get more resources and freedom. Leaders will trust your judgment if they understand you.
#2 Be your own devil’s advocate
Show that you’ve checked your blind spots.
Data science interview processes often involve a project component where candidates work through an analysis problem. There’s likely data cleaning, ETL and modeling components involved. Most candidates do OK on all those parts, but stumble in two key areas:
- They fail to outline limitations and caveats of their model (i.e., why their model is the “best” in a sea of other options), and
- Forget to propose improvements for future iterations (what would you try if you had access to other datasets or more time?).
There’s never a “perfect” model, and we all think of limitations and alternatives when building it. Yet many candidates fail to communicate them.
Yes, it’s scary to highlight why the model we just proposed may not be the best, as it may invalidate it altogether. But the alternative is much worse: using a suboptimal model without a clear idea of how to improve it.
At Best Buy, continuous improvement and constructive feedback are core values that drive our approach to development. The entire technology organization uses agile and lean principles. Candidates that are unafraid to pitch half-baked ideas and debate with peers are most desired. So go ahead and show that you can play our own devil’s advocate; demonstrate critical thinking. Show that you are a great resource to bounce ideas with.
Here’s a blog post that outlines critical thinking strategies. I highly recommend making some of the questions outlined as part of routine product development. I’d also like to pause and recommend data scientists to read Thinking Fast and Slow, a book on the research of two Nobel prize winning psychologists on human irrationality and biases. The book really helped me become more aware of biases when interpreting data.
#3 Imagine and Shape the Future of Data Science Here
I’m of the opinion that everyone on the team plays a key role in shaping the future of a company. Not just the CEO or VPs. This translates into acknowledging the company vision (e.g. At Best Buy, we work to Enrich people’s lives through technology), and think of how we can advance toward it. Give ourselves room to be creative.
In the context of data science, this could be looking at new tools or development processes or even team structure that enable easier deployment of machine learning into production, or faster time to MVP, etc. What could improve both our lives as data scientists, and the lives of our customers who benefit from optimized experience?
However you envision the future of data science, take leadership and shape it. Share with prospective employers your vision, why it excites/appeals to you, and one way we can get there. It doesn’t need to be right. The goal is to demonstrate you’re thinking about tomorrow. Because the average and median data scientist is too busy reacting to the present, creating reports with fun facts of no value.
Don’t Bury Your Accomplishments
Final note on resumes. Chances of getting past screen are much higher if you avoid bullet points on that only describe what you did: e.g. “Setup ETL pipeline using DBT…”
Instead, highlight what results/impact you achieved: e.g. “Increased productivity 30% by setting up an ETL pipeline using DBT that…” So start with the R in the STAR format.
Nobody is looking for a perfect resume, but let’s show employers that you don’t just work to work. You are a data scientist with a purpose: You exist to maximize value to the business. So celebrate the impact you had. Celebrate you.