Data analysts decipher business problems, create algorithms, and turn terabytes of structured and unstructured data into valuable intelligence that businesses can understand and use. “Imagine all the comments companies receive from social media and emails,” said Brandon Purcell, data science team lead at Beyond the Arc, a customer experience consulting and advanced analytics firm based in Berkeley, Calif. “A data analyst has to create a taxonomy to turn all of that unstructured data into rows and columns which isn’t always easy.”
Job growth for data analysts of all stripes is expected to rise over the next several years, thanks to shortages in burgeoning industries such as healthcare and banking; those specializing in market research could see their profession grow 32 percent by 2022, according to the Bureau of Labor Statistics.
According to Purcell, interviewers for data-analytics jobs want to see if a candidate can not only crunch numbers, but also organize unstructured data and communicate crucial trends and important findings to business leaders.
How would you create a taxonomy to identify key customer trends in unstructured data?
- What Most People Say: “I’d probably read through a random sample of the data then partition it into training and validation sets. Then, I would create an algorithm and categorize the data based on what I see in the random sample.”
- What You Should Say: “First of all, I’d consult with the business owner from the outset to understand their objectives in categorizing this data. Then, I would use an iterative process, pulling new samples and modifying the model accordingly, evaluating for accuracy and inclusivity along the way. I’d follow the basic process of mapping the data, creating an algorithm and legend, mining the data, visualizing the data and so forth. However, I would tackle the project in segments in order to solicit feedback from the business stakeholder, and to continue enriching the model to ensure that I’m producing actionable results.”
- Why You Should Say It: You can analyze the data until the cows come home, but a model is meaningless unless it produces actionable results. An experienced analyst varies his or her strategy depending on the type of data being analyzed and the desired results. For instance, was a customer complaint retweeted on Twitter? If so, should that data be included or excluded? Plus, sensitive data has to be protected, so it’s best to consult with the business owner to make sure you’re following compliance regulations, disclosure laws and so forth.
How would you handle the QA process when you’re creating a predictive model to forecast customer churn?
- What Most People Say: “I would partition the data into two subsets and use the training set to build the model. Next, I’d evaluate the model using the validation set. Generally, I use lift and gains charts to validate the results, visualize the improvement, and to identify the best predictive model.”
- What You Should Say: “I’d partition the data into three sets: training, testing and validation. To eliminate bias in the first two sets, I’d show the results from the validation set to the business owner. I need input to gauge whether the model accurately predicts customer churn and provides actionable results.”
- Why You Should Say It: To operationalize analytics, data analysts need a collaborative environment and input from the business owner. They also need a repeatable, effective, efficient process to create predictive models and reliable architecture to deploy predictive analytic models into production. Without feedback from the business owner, the model will probably sit on the shelf. And no self-respecting analyst wants to create a one-and-done model.
How often should you retrain or refresh a model?
- What Most People Say: “I usually retrain a predictive model every six months or at the end of the quarter when the company releases its financial statements.”
- What You Should Say: “I’ll work with the business owner to establish an appropriate time period upfront. However, I would retrain a model immediately should the company expand into a new market, consummate an acquisition, or encounter emerging competition. Models must be retrained quickly to adjust for changing customer behavior patterns or shifting market conditions.”
- Why You Should Say It: A top-notch data analyst understands how changing business dynamics affect the efficacy of a predictive model. She’s not just a data cruncher: she’s a valuable consultant who uses her analytical skills and business acumen to solve root-cause business problems.
- 4 Interview Questions for Data Miners
- Data Expertise Needed by Marketing Technology Firms
- Despite Apps, Much of Data Analytics Still Drudgery
Image: Fernando Blanco Calzada/Shutterstock.com