Tips and Tricks
Two Approaches to Data Insights
When everyone involved in data analysis thinks the same way and uses information the same way, it can be easy to come to a consensus about how to operate. Having a small team of centralized data analysts is a great way to streamline analysis, but it comes at the cost of missing out on information from other teams.
At Periscope Data, we believe that data done right involves analysis that takes into account a diverse range of expertise. With data discovery for business, we’ve opened the door for non-technical business users to get their hands on data and start looking for connections that can only be made by someone with their knowledge and experience. More people analyzing information is going to require some operational changes to avoid confusion and allow everyone to get the most out of their data.
The purpose of analyzing data is to draw a connection between data and tactical business actions, using the data as both a measuring stick and a way to make intelligent recommendations. When a small data team owns the process, it starts with the data and progresses through a series of connections until ultimately a business recommendation is clear. When business users are invited to the analysis, they also bring in a new way to process data, one that starts with knowledge of the end results and works its way back to find a solution.
For example, imagine a company that is analyzing suggestions to minimize customer churn. A data scientists might start that process by examining the inputs that create the company’s primary churn metric. They will determine whether to prioritize retaining customers or revenue. They will look at whether the time period is the right one to be using. They might consider whether to measure against historical performance or an established industry standard. Once they have determined that the churn metric is measured accurately, they will begin looking at a wide range of variables until they find recommendations that are statistically likely to affect churn.
A business user tasked with the same job will probably look at the process in reverse. They’ll start by looking at the final churn rate, make a note of when it is going up or down and begin connecting the dots to decisions that were made at the same time. They’ll test hypotheses that are formed by their knowledge of business activities until they find an answer to what caused that change in churn.
Both approaches can pinpoint a variable that affects churn. The data team takes a wide net approach and sifts through every variable until significant correlations emerge. Business users start with specific hypotheses and test them until they come to a conclusion.
The point of this post isn’t to suggest one process as more valid or accurate than the other. Both user types are starting their search for insights with what they know best. However, as the pool of data analysts expands, it’s going to be important for everyone involved to accept data-driven insights generated from alternative approaches.
In the case of the churn KPI from earlier, maybe the right way to handle this is to create a dataset that gives transparency into the churn calculation. Instead of showing one number that indicates the rate of monthly customer churn, they can add transparency by listing out the individual components that are part of that metric. If the calculation is simple like (customers at beginning of month - customers at end of month) / customers at beginning of month, then make a column for customers at the beginning of the month, one for the customers at the end of the month, another for the difference and a final one for the churn rate.
If the calculation is more complex, increase the number of columns to show the complexity. Maybe you treat new customer churn differently. Maybe you want to analyze monthly customer churn by product line or by some sort of audience cohort. All of that can be handled beautifully across audiences, you just need to be as transparent as possible so that someone who starts their analysis at a different point can see the path you took to get to your results.
The goal of adding new points of view to the analysis process is to generate rich new insights. By breaking down the barriers to data and increasing transparency, both business users and data professionals can see how the other operates and arrives at conclusions.
If you’re looking for more tips to help you new citizen data scientists get more out of their analysis, check out our How to Chart Your Data Discoveries guide.