DON’T CHUCK NUMBERS AT PEOPLE. GIVE THEM ANSWERS.
Executive Summary: This blog post sounds a bit preachy. And it is. But we have a saying at Sentier that has served us well: “Don’t chuck numbers at people.” It can mean all the difference when getting decision makers to accept advanced analytics and using the important insights that we generate.
It is fair to generalize and say that good data scientists are incredibly bright and creative, and they generate insights that can have a positive impact on their business. It is also fair to say that what we deliver, and what can be consumed by decision makers, are often two very different things. We are all guilty of not taking the time to clearly articulate the conclusions we are drawing and how we got there. And, most importantly, showing why what we have is the best answer out there.
We spend weeks of sleepless nights figuring out the answer to the next big question only to deliver a spreadsheet full of numbers or parameters that would take someone not closely linked to the analysis another week to figure out. So at Sentier we have a saying: “Don’t chuck numbers at people.” What does that mean? When is the analysis really done? And what can data scientists do to maximize our impact on decision making?
WHAT DOES IT MEAN?
Pictures say a thousand words. So do you think an executive can better act on this:
or this:
WHEN IS THE ANALYSIS DONE?
The analysis is done when the results can be communicated to an executive in five minutes or less and to the point where they feel confident they have all the information necessary to make a decision. Five minutes sounds tough but it is the reality in which we work. So deal with it.
WHAT CAN WE DO BETTER?
First, answer the question. Many times, as analysts, we get caught up in all the things that could negatively impact the results and spend a lot of time talking about that upfront (i.e. the time frame for the analysis is not 100% indicative of where we are now; the data used had some issues, etc). If the doubts that accompany all analytic results are bugging you that much, you should go back to the drawing board. Otherwise, start by answering the question that the executive asked. In the example above, we clearly answered the question, what is the optimal level of spend for Promotion A?
Second, present the results in a way that a person reasonably familiar with the type of results you are delivering can understand. If you are thinking of delivering the results of your analysis in a wide (or even small) table—don’t. Analytic results more often than not can be presented in a visual manner that can be consumed much quicker. In the example above, executives don’t care about the l, k, and x0 values or the machine learning that went into generating them. But they do care that there is room left on the response curve for profitable spending and that concept is easy to grasp.
Third, prepare a concise description of where the analysis came from. Executives need to quickly assess if the results are trustworthy or not. It is important to include a brief description of where the analysis came from, any assumptions you had to make, and what confidence you have in the results for the intended purpose. If available, it is always good practice to also show the results that the same analysis yielded in previous deliveries explaining the variations, if they are material.
So there—we have preached. Advanced analytics are not for the faint of heart and are not always easy to interpret and present to executives. But they are powerful, can offer better insights, and uncover decision paths that can greatly improve the business. So as data scientists, let’s be sure we are giving our hard-won results the best chance possible by following the three rules above. Give them answers, not numbers.
Sentier offers innovative analytic solutions that take out clients from data to insights in industry-leading time:
Marketing & Sales Effectiveness: How do you best allocate your budget and resources?
Customer Experience: How do you deliver the best experience to your customers?
Mission-Driven Analytics: What are those questions that no one has been able to answer before?
