Grappling With Uncertainty — In Business and Data Science
How data science principles can be applied to better decision making in business
The role of data and data science in business
Over the course of my career, I have been fascinated by decision-making and how to help organizations make better decisions. There is the human element of this question which is a piece of the puzzle that is studied by psychologists, organizational behaviouralists and leadership experts. Another piece of the puzzle is the practice of using data to inform decision making.
“Data-driven decision making” is a phrase that I have often heard used in describing an organization’s culture or an individual’s preferred leadership style. To me, the flaw with saying that an organization is “data-driven” is that data becomes thought of as a source of truth that can make decision making clear and straightforward. I do not share this view. In fact, I believe that data is fickle; data can be easily manipulated to fit any story so more data often confuses the decision. So, how does data help?
I believe that the thoughtful use of data is the best way we can approximate the truth. Data can help us narrow in on a wide world of possibilities as being possibly the truth to a nuanced view of likelihoods of different truths. Data helps *inform* our decision-making; it doesn’t make a decision for us.
Why do we care about uncertainty?
Every business is built upon a series of decisions: where should I open shop? should I sell this product? how should I price this product? when should I order more inventory? should I hire this candidate? These decisions are consequential, not only to the success of the business but also to the wellbeing to those who work with the business. Critical to making these decision is understanding the effect of the decision on the results we’re interested in be it higher profits, less product wasted or a healthier team. The challenge is that the truth will never be as predicted. In my experience, this leads to a mistrust of predictive models and of data in general.
Data, thoughtful interpreted and analyzed, can help us understand uncertainty. Just because a prediction isn’t exactly right, we shouldn’t throw out the entire process. When interpreted and analyzed well, we can use data to wrap our heads around future outcomes in nuanced ways that don’t just give us one prediction but a range or predictions. We can identify what outcomes are *possible* and how *plausible* those outcomes are. As an example, if I were considering buying a car, I would put little trust in the claim that the car would break down exactly 5 years from now. What I would believe is that there is a 30% chance of failure within the first 3 years and a 20% chance of failure in years 4 and 5. Comparing these probabilities to that of another car model I’m considering gives me helpful and believable information to help me make a decision.
The concept of trying to understand a range of possible results is not new to most of us. Many of us do it naturally, seeking out qualitative and anecdotal information to help with our decision. When considering a car purchase, we will scour the web for reviews and talk to friends and family about their experiences. When it comes to decision making in the organizational context, we need to formalize the measurement of uncertainty. We can do this by expertly combine quantitative techniques with our intuition to make important decisions.
Demystifying uncertainty — a series of posts to come
In an upcoming series of blog posts, I will outline analytical techniques that help us gain a better understanding of uncertainty. Techniques will range from those from the business analyst toolkit of modelling scenarios to those in a data scientist’s toolkit which leverages statistical techniques to estimate uncertainty.
Many organizations will have already had experience with testing their decision under different scenarios to make important decisions, although sometimes sporadically. The next step to improving decision making is to systematically use statistical techniques to estimate uncertainty. Organizations who do this well leverage their better understanding of uncertainty to expertly trade off risks in order to make better-informed decisions and to prepare for a range of likely outcomes. They also empower all levels of the organization to use data and its nuances to make decisions for themselves, supporting the shared mission.
As someone who started a career in management consulting before being exposed to more rigorous analytical techniques in data science, I understand the instinct to brush off statistical techniques. I have shared these beliefs before, that statistics is too rigid to be applied to the real world, that it’s too complicated to explain to decision makers, that it doesn’t work when there is limited data. Over the past year through my studies and work, I have found these beliefs to be unfounded, especially with simulation-based techniques that have become possible with today’s computational power. Over the next several weeks, I hope to convince you of the same.
