The Art of Simplicity

After ten years in Academia, I joined Corporate America and its business reality. At the University, I was trained to deliver perfection in my analysis, spending years in improving our methodologies and reducing statistical errors (more data points) and systematic errors (smaller biases and variations). A typical research would take many years. As it was fundamental research, the applicability of findings was highly questionable. They don’t need to be. Indeed, although there are only few applicable key findings from fundamental research, the outcome that those results generate mostly overcome their cost at a global level. For example, think of the invention of the laser and all of its application today. We believe that for every $1 invested in fundamental research, the return will be $7, in the long term, many decades or centuries later. Hence, only international collaborations of states can provide such investment with such time horizon.

When I joined eBay, I discovered that this academic approach was inadequate for the business world. The inflow of daily questions was far superior to our ability to answer them. Some questions were fascinating from an intellectual perspective and could have benefited from a Ph.D. thesis. We only had a few days and not a few years to answer them. So, I had to relearn and adapt. The saying “Keep It Simple Stupid” was true in my situation. Kelly Johnson, a lead engineer at the Lockheed Skunk Works, created this principle. The word “stupid” is not intended in the sense of erroneous or illogical. It describes the lack of complexity to fix a problem or a failure.

The goal of a business question is to quickly assess potential changes. Either proactively before a new product/feature is launched, or reactively with a “what the heck did just happen?” My preferred archetype example is a real example coming from PayPal where the question of the color of its web button yielded to the transition from the original blue button to the current orange button. I’m sure one could spend two years analyzing this question to the smallest details… yet, one could also take half a day analyzing the right data, and coming to the same business conclusion.

So, the real question here is: How complex should an analysis be?

To find the answer, let’s consider the level of complexity for a given business question. The average impact that the finding will have is most likely a function like this:

Where the X-axis is the level of complexity, the Y-axis is the output in dollars. Here the impact on the company is measured in dollars. As we increase complexity, the Impact curve rises quickly at the beginning, but then reaches a plateau to a maximal impact that the finding of the question can have. At some point, after many efforts, there is not much that one can do to improve significantly the impact that the analysis can have.

From an academic perspective, without considering resources, the best analysis of a question would be the one with the most complexity given the time frame allocated - a year maybe for a master thesis, four years for a Ph.D. thesis.

Now, let’s consider the cost of the analysis. Here, I’m not only considering the cost of collecting the data and analyzing them, but also all ancillary activities like explaining the findings to others and implementing the outcome. This cost will exponentially grow with complexity:

Indeed, the more complex the analysis, the more complicated and lengthy the explanation and the harder it will be to convince key stakeholders that the findings and the business conclusion are correct.

Hence, if we subtract the impact of the results with their cost, we have the profit curve:

 There are two interesting points in this profit curve:

Point A where the profit is maximum. Point A would define the optimal level of complexity one should follow to answer a specific question.

Point B is the current limit of the capabilities of an individual and the tools used. Indeed, we all have limited intellectual capabilities that restrict the level of complexity we can use to solve a problem. One can increase this limit through education and experience. This limit is also defined by the tools used, and their limited capabilities (usually in term of storage and CPU).

This point A is much lower than the level of complexity that the vast majority of analysts and data scientists would suggest using. Primarily, this is coming from their academic background where cost is ignored. The maximal complexity is often chosen. However, this also comes from a human behavior where people always tend to want to learn new things and push the limit of their capabilities.

In my experience, I always suggest using an 80/20 rule where one would apply the “Keep It Simple Stupid” principle to 80% of the questions and provide the answer as rapidly as possible. However, in 20% of the cases, one should spend more time investigating new methodologies and tools to satisfy this human need to learn and discover new things, as well as enable our talents to stay fresh and up-to-date with the latest technologies and methods.

Moreover, it is also critical to investigate specific questions with more scrutiny and complexity, to find innovative approaches and answers that would provide a disruptive application of an existent (& old) business problem.

This is the point where science becomes art. Which project to deep-dive? One needs to use intuition and gut feeling, to select the 20% of business questions that need to be investigated further in order to unlock significant innovation and disruption. This is where the Science of Simplicity becomes the Art of Simplicity.