How to Interview a Data Scientist: A Guide for Non-Technical Interviewers

Interviewing data analysts and scientists can be challenging if you don’t have a technical background; however, there are simple questions (to ask, not necessarily answer) that will vet a potential hire’s analytical acumen. These questions are designed to gauge how analysts conceptualize data and analytics, rather than test technical skills.

Certain questions below have been tailored for a data analyst and data scientist role. The specifics of these questions can differ — what is most important is that the answer captures what we want to assess.

Q1. Describe our database …

Sample questions

• What entities (or data classes) would you expect to see in our database? Examples include users, products, inventory, and orders.
• What do you think are the most important tables (or datasets) for our business? What fields (or columns) would you expect to see in these tables?
• How would you pull a particular report from our database? At what grains (or granularities) would you output the data? What dimensions and metrics would you include?
• What data sources could help us better target customers?
• What data would you add to our current data? What technology stack would you use to load that data regularly?

What the question assesses

The interviewee should be able to describe several tables in your company’s database. Data structures are common among industries — e-commerce companies have similar data, businesses using Google Analytics have the same web analytics data, and CRM data for B2C companies will be the same semantically.

If the Interviewee can answer this question well, they likely understand (a) how relational databases work and (b) the fundamentals of your business. This also gives insight into the types of data the interviewer has experience with (e.g. point-of-sale, online, syndicated).

Q2. How would you segment (or model) our users?

Sample questions

• Data Analyst Variation: What segments would you build to better understand our business? What types of segments would you use to personalize our CRM efforts?
• Data Analyst Variation: What type of targeting segmentation would have the greatest impact on a given goal (e.g. retention, upsell)? Why?
• Data Science Variation: How would you build a model to predict customers’ lifetime value? What assumptions are you making when building this model? What caveats should we consider? How would we operationalize the model?
• Data Science Variation: What are some ways we could draw executive-level insights from this model? (Note: most machine learning models cannot be interpreted for business purposes; however, the data scientist should know that a parsimonious model can accomplish this).

What the question assesses

This probes how analysts apply their knowledge to execute an analytical process. When discussing regressions with a data science interviewee, we’re trying to gain insight into two areas: (a) do they understand underlying statistical concepts, and (b) can they apply their analysis to your business? For instance, it’s important to know that one should only interpret variables with low p-values, but it’s also important to know why variables with low p-values are relevant for the business.

When answering the data analyst version of this question, demographic and recency-frequency-monetary segments are good starts. The next step is behavioral segments (such as an abandoned cart or cluster-based profiles). Also, the reason for specifying “to better understand the business” versus “to tailor marketing” is to test whether the interviewee can differentiate insight-oriented segmentation from activation-oriented segmentation.

Segments will often depend on the industry; however, this is a good question to gauge interviewees’ basic understanding of segmentation and business acumen. Just make sure their examples are business-relevant.

You can read more about enterprise segmentation here.

Q3. How can we use your segmentation/model to impact our business?

Sample questions

• How can we use your segmentation/model to better understand our business? How would you deliver your work to provide insights for our business (or product) strategy teams?
• How can we use your segmentation/model to make a business process better or more efficient?
• What caveats should we consider when operationalizing your segmentation/model? Are there situations when the model would not perform well?

What the question assesses

This question gauges whether the analyst has the ability to create and communicate business value from their work. We’re also testing analysts’ ability to (a) communicate insights and (b) design data products.

• Insights communication enables stakeholders to better understand their business, generally through ad hoc analysis or reporting dashboards. The candidate should discuss how they would present simple, relevant information based on their audience.

Conclusion

After writing this article, I realized that we discussed somewhat technical concepts above. But, to step back, we are testing three areas with these questions:

• Basic understanding of your business’ data
• Conceptual understanding of segmentation and/or modeling
• Applying analytics to your business

We’ve focused on segmentation since it’s a common application in B2C businesses. However, we can tailor questions to predictive modeling, natural language processing, geospatial analytics, or other topics.

Analytics is just as much strategic as it is technical. Therefore non-technical interviewers can “go deep” on the strategic aspects of technical topics — specifically, how does this apply to our business and what factors should we consider when applying it to our business?

By framing these questions from a strategic lens, we can (a) test whether interviewers can frame their analytical thinking around a business problem and (b) gauge how well they can align analytics with the nuances of your business or industry.

Happy interviewing!