11 Ways to Ask Smarter Data Analytics Questions
Not everyone has been trained to think like a data scientist or a data analyst, but they can learn to think more like one.
![](https://eu-images.contentstack.com/v3/assets/blt69509c9116440be8/blt3bed4f0a8d8680cd/64cb5a1f1e75beaf4e8c0c9c/dataanalytics-Skorzewiak-adobe-cp.jpg?width=700&auto=webp&quality=80&disable=upscale)
It's been said that the quality of insights depends on data. However, the quality of insights also depends on the questions asked. While one of life's pearls of wisdom is, "If you want better answers, ask better questions," how does someone know what a better question is, let alone a better answer?
“When an experienced data scientist is working on a problem, they’re usually not looking to prove themselves right; they’re looking to prove themselves wrong," said David Robinson, principal data scientist at data science platform provider Heap. "They’re thinking, what if this data source is untrustworthy? Would this be different if I used the median rather than the mean? How could I tell if this conclusion is confounded by other variables? This habit of critical thinking helps you build trust not only with stakeholders, but also helps you trust your own conclusions so you can build on them further.”
A couple of factors to consider are context and the level of specificity that's appropriate. For example, 2020 involved many changing contexts including hoarding, supply chain issues, stay-at-home mandates, remote work and shifting consumer behavior that could have altered the way a question should have been articulated. Similarly, is the point to understand a general trend or a trend specific to a segment of a target market using a particular product? Bear in mind that different types of questions tend to require different types of data to produce appropriate answers.
"Looking at a problem from a variety of angles is a great way to approach an issue analytically," said Eric Blank, marketing analyst lead at marketing agency PACIFIC Digital Group. "Take into account multiple sources of information that play different roles in the space you’re analyzing. In doing so, the data will begin to tell a story and allow you to uncover where opportunities or issues exist.”
Data scientists and data analysts know they may not necessarily ask "the right" question and that's OK.
"Since this is an iterative process -- full of dead-ends -- the analysis is not always on the right track," said David Smith, VP of data & analytics at TheVentureCity, an international, operator-led investment organization. "But after a while, I start to understand the correlations between different variables, and which are the most important drivers of outcomes. Then it becomes easier to home in on the salient insights to glean from the data. It can also help to touch base mid-course with stakeholders to see how initial insights match up with their intuition and domain knowledge."
By talking with the people who want the problem solved (such as people working in lines of business) data scientists and data analysts can better understand what their "customer" wants to accomplish. An interactive approach, with other people and data, can help pinpoint what the better questions are.
"Analysts are indeed better-than-average in distilling a challenge into meaningful questions. Getting from 'Why aren't our websites converting better?' to 'Will including customer testimonials improve our website conversion?' takes a big leap into creating value through analysis," said Ilkka Petola, head of growth and former head of data science at online invoicing software provider Zervant and the company's previous head of data science. "This skill is best acquired by working on a problem that's meaningful to you, together with an analyst."
Other options that are not mutually exclusive include:
Taking a course or reading a book about how to think analytically
Learning how to use an analytics platform proficiently
Asking data scientists and data analysts for their suggestions on how to improve query quality
Reading articles on the subject
Learning by doing
Good, analytical thinking requires a shift in mindset that tends to be more skeptical than the average business professional tends to be with data.
"Critical thinking is the practice of thinking logically and analytically, and it underpins all scientific thought and process," said Peter Watson-Wailes, founder of scientific product management solution provider Hirundin. "Problem solving is thinking to analyze data and find ways to find solutions to challenges you encounter. These two modes of thought are the basis of all the work you do in research and analysis, and both are things you can get better at."
Following are some suggestions from more data scientists and analysts.
Analytical problem-solving should be tied to a business objective. Are you trying to increase revenue? Reduce costs? Attract customers? Retain them? Optimize or predict the likelihood of something? Analytics for analytics sake won’t get you far.
"Being formal and explicit about objectives is the first step. I personally like to start with the end goal and work backwards -- breaking it into smaller problems that can be tackled sequentially," said Miguel Araújo, director of data science at unified targeting company Semasio. "However, at the end of the day, seeing how others approach problems that seem out of our reach is critical for growth because while technical prowess plays its part, that's rarely what defines the success of real-world projects."
One thing that sets data scientists apart from mere mortals is their dedication to the scientific method which begins with a hypothesis to be proved or disproved.
"You'll know you're on the right track to finding something like the truth if you're spending time discovering causes which could explain your data, but don't stand up to testing," said Peter Watson-Wailes, founder of Hirundin. "Remember, you're always trying to be as precise as possible in your explanations of causes. It's not just about being right, it's about being precise and right. Progress in all research fields is about narrowing down what isn't, until you're left with just what is.
The quality of an analytical result depends on the data. The question is whether a person has the data required to answer a particular question.
"Seek to understand sources of data, and all the interactions that make up the source. A lot of times you'll gain more from an analytical perspective if you know where data comes from, where it's been and who's touched it, versus what it tells you in the end." said Michael Schwarz, VP of data at data, technology and digital transformation firm KSM Consulting.
Alicia Frame, director of graph data science at graph database platform provider Neo4j, said if one lacks the data one needs to answer their questions or test their hypotheses, then it's time to think about your organization’s data strategy in the first place and what you can do to ensure that you have the information available to act in a data-driven way.
Finally, not all data is equal. Brittany Davis, head of data at data platform provider Narrator.ai, said focusing on customer actions instead of attributes increases the likelihood of finding something useful. "People who are new to data tend to focus on the attributes about their customers (job role, industry, company size) as valuable drivers for their KPI, but seasoned analysts will tell you to look for behavioral indicators instead because they know from experience that the behavioral traits are usually the most predictive of future action."
People who have not been trained to think analytically may get overly excited about results too fast, particularly when those results align with one's own point of view.
"Data analysts and scientists form hypotheses based on what the data tells them. For example, if a manufacturing line isn’t producing enough units, a data analyst will turn to the data from the line to find the bottleneck, then dig further into the data to understand why the bottleneck exists," said Sameer Gupta, VP of data at Drishti, which specializes in factory video analytics for manual assembly lines. "Conversely, a non-data data analyst or scientist will largely rely on his or her expertise and past experience to hypothesize the cause, then look to the data to validate that assumption. If Pete is known to be slow and he’s working that day, a non-data analyst will assume Pete is the bottleneck and look for evidence in the data."
Citizen data scientists and other business users should do the same. "Thinking analytically requires you to let the data tell you the story rather than trying to figure out how to make the data fit your narrative. Trying to fit your data can blind you from finding real insights that can solve your problem," said Michael Pugh, business intelligence manager at The Lifetime Value Company, a portfolio of brands that helps people discover, understand and use data.
Blake Burch, co-founder of serverless data operations platform Shipyard, said ordinary people make the mistake of providing the results of an analysis without providing the why. If they do provide the why, they don't always verify whether the why is the real why. Instead, they find a convenient fact that could explain the results.
"Advertisers are extremely guilty of this. They may say that advertising drives online sales, therefore the advertising is effective. However, they don't always do the appropriate testing to verify if the sales would have happened without the advertising or if other business factors contributed to the interest and therefore the sale," said Burch. "A true data professional will dig deeper to determine if the why is repeatable. Does this outcome typically occur if these factors are present? Does the same outcome occur if these factors are not present? Do the results change over time? Are they specific to a certain period of time? This type of thinking ultimately results in predictive modeling."
Thomas Wood, founder and chief data scientist at AI and machine learning consultancy Fast Data Science, said a good data scientist tries to solve a problem in the simplest way.
"Sometimes a linear regression model is all that is needed. Or the data science investigation may point towards a non-data-science solution, such as adding a form to the client's website. A lot of data science is just common sense, backed up by numbers," said Wood. "A person who lacks training, such as a novice data scientist, or a non-technical manager, might have heard a buzzword such as deep learning, and jump into spending far too much time and money on that particular tool, but perhaps you have a simpler tool in your toolkit? An experienced data scientist knows when to bring out the flashy tools and when to use Occam's razor. Good data scientists don't think too much about technologies, but rather about the problem itself."
Frank McSherry, chief scientist at SQL streaming database provider Materialize said one tends to be on the right track when your queries start to get simpler. "It isn't too hard to write complicated queries that find something interesting; however, as your queries get simpler, they focus attention on the ‘reasons’ for the interesting answers."
Data scientists often visualize data at an early stage to understand what they're working with. They also use data visualizations later to explain to an audience what outcome of a query is. Alexander Zograph, founder and senior data scientist at customized data science and AI solution provider isofti.com recommends that instead of just trying to understand the data, one should also strive to understand their dependencies.
"One column of data wouldn't give you anything, but two columns between two parameters would give you much more of a sense. All sorts of dependencies could be analyzed such as correlations, trends or variance," said Zograph.
Francisco Arceo, senior data scientist at ecommerce payments startup Fast thinks the best way to learn how to think analytically is to apply structure to an unstructured problem.
"Applying structure to an unstructured problem forces you to understand the process, the underlying mechanisms that are moving things, the quantification of the right things and, ultimately, the way to solve the problem," said Arceo.
Fast's Arceo said one way to determine whether you're on the right track analytically speaking is to measure success and develop a solution that incrementally improves your outcome,
"The most important piece is measurement," said Arceo. "Without that, you have no ability to make progress. The first solution doesn’t need to be perfect; it just has to be improved with data. Measuring success allows you to create a solution that can be continuously improved."
Jackie Sun, senior engineering data analyst at Fivetran, said before jumping to a conclusion, it's wise to check the baselines.
"I think people who are not trained in data or [how to] analyze it frequently often forget about baselines. They may see something that they think is unusual, but it may actually not be if they were to consider the baseline. A request I often get as a data analyst is to provide some analysis around some unusual behavior that the company is seeing -- revenue growth is slowing down, or employee productivity is decreasing -- and one of the first things I check is the baselines to establish whether what we’re seeing is actually unusual or not.
Gut feel isn't always accurate, and neither is data analysis. If you get a result that doesn't feel right, it might be an indication that you could ask a better question, that something in wrong with the data, or something else has gone wrong.
"Part of the beauty of data analysis is that there isn’t a prescriptive way to do it. It can be a very creative process! It’s easier to tell if you’re on the wrong track, which often comes from built up intuition around the dataset," said Fivetran's Sun. "Do a gut check of your results. Do they actually make sense? Try to poke holes in your conclusions. If possible, try to write your queries in different ways -- do you get the same results?"
Check out other InformationWeek slideshows.
Gut feel isn't always accurate, and neither is data analysis. If you get a result that doesn't feel right, it might be an indication that you could ask a better question, that something in wrong with the data, or something else has gone wrong.
"Part of the beauty of data analysis is that there isn’t a prescriptive way to do it. It can be a very creative process! It’s easier to tell if you’re on the wrong track, which often comes from built up intuition around the dataset," said Fivetran's Sun. "Do a gut check of your results. Do they actually make sense? Try to poke holes in your conclusions. If possible, try to write your queries in different ways -- do you get the same results?"
Check out other InformationWeek slideshows.
-
About the Author(s)
You May Also Like