top of page
Writer's picturePreet Minhas

How to not be a statistic: Tips for data scientists on generating quantifiable impact.

Updated: Feb 5, 2023


Photo by Towfiqu barbhuiya on Unsplash

During my training as a data scientist, I initially believed that the primary factor determining the success of a project was the data scientist's mastery of various machine learning algorithms. I thought the sole objective was to achieve the best possible performance of the model. However, I soon realized that this was a naive assumption. While a strong understanding of machine learning algorithms and a high-performing model are crucial elements, they are not the only determining factors of a successful data science project.


What I have learned is that a project’s success, in fact, lies in the ability to convert analytical insights into tangible business results and revenue for a company. This translation process is not straightforward, and as noted in a Harvard Business Review article by Kevin Troyanos, Gartner predicted that only 20% of analytical insights will lead to business outcomes through 2022.


That was a chilling statistic. One of the things I care about as a data scientist is to make a difference for my company (or for my clients). The thought that 80% of my work could be ineffective was concerning and disheartening.


What can we, as data scientists, do to ensure our analytics turn into business outcomes? From my experience, it’s all about the initial steps we take in order to understand the problem and the data.


Every budding data scientist should be trained on how to formulate valuable business questions: keeping the company’s business strategies and decision making processes in mind, what is that we are trying to solve? Why is solving this problem so important to the company? Who are the stakeholders of the problem? Will finding a solution to this problem lead to specific actionable item(s) that will generate measurable value?


Moreover, a valuable business question is always business driven, never technique driven - meaning it should never start with “IT has developed this amazing new technology, how can we use it?”, rather “We want to promote customer growth through marketing, how can we make that happen, using new technology?"


Looking back at my early career, I can think of a time when a client came to us with a technique driven question. The client wanted to increase customer engagement through various channels, including email and the web. The client had already done previous analysis and had specific insights on how to increase engagement; however, they wanted to use the automated feature engineering aspect of our in-house ML tool to gain more insights and predict future engagement. The question was focused on the technology and how it could help them gain additional insights and not on how gaining these additional insights would help their company.


When starting with technology as the primary driver it’s easy to lose focus on why it really matters to solve a problem. Diving a little deeper, we may discover that the real reason to solve the problem of client engagement is to increase the sales of a product, and only a specific group of customers are of interest. Then, a more valuable question to ask is: How can we increase the sales of product Z through improving customer engagement of customer group X, over a timeframe Y?


Keep in mind that formulating a valuable business question isn’t something done solely by the data scientist. It is a process that requires teamwork. It is necessary to have subject matter experts (SME) present in the discussion, as they know the business processes and data best, and can help in identifying gaps and inconsistencies in the existing business processes that can be improved to generate additional value to the company.


SMEs were very valuable to me and my team in helping us understand that some of the missing data was a result of faults in the way the data was initially collected, and then later helped us identify target leaks in our model.


Throughout the process of selecting a question to focus on, confirm alignment across all departments, from marketing to IT, on the company’s business drivers and strategy to ensure that the questions being explored also line up.


Once there is candidate business question to work with, it is worth liaising with other teams within the company (or within client company) to explore if the required infrastructure and business processes will be in place to support the implementation of your data science outcomes. Of course, the change of management and implementation is well beyond our wheelhouse as data scientists, and is quite probably a strong contributor to why so few analytics are translated into business outcomes.


The bottom line is that formulating valuable business questions is the first step towards adding quantifiable value to the business and to not end up part of that 80% statistic. In my opinion, investing extra diligence in this step will prevent wasting time and money.

Before moving forward with the valuable business question, identify those who have the power to change business processes and those that have the power to implement the change so that they can be responsible for integrating the insights (that we as data scientists discover) into the day-to-day work and decision of the front-line workers. Work on a valuable business question for which it is clear how the solution can be implemented.




Thank you for reading!

I would love to hear your thoughts!



References: Forbes, Janet., Leighton, Danielle., Brin, Lindsay. "From Theory to Data Product: Applying Data Science Methods to Effect Business Change." Strata Data Conference. 11 Sept. 2018. Lecture


54 views0 comments

Comments


bottom of page