Roman Remora, Executive Director of Data Science – Chewy Operations Applied Economic Research
Over the past five years, I have been frequently asked how to unlock the value of data science and big data programs in various applications in finance and supply chains. This question becomes even more important as economic growth slows, with COVID-19 being an unpredictable and exceptional event.
For companies looking to get the most out of their data science projects, there are two key aspects related to data science projects that you need to keep in mind:
(1) Functional components are divided into operational activities and strategic activities
(2) The technical part is divided into infrastructure and science.
Operations and Strategy”
Where and when should I invest?
Functional areas of data science projects are typically divided into purely operational activities or strategic activities. While they are not necessarily mutually exclusive, they often conflict with each other when it comes to allocating and funding plans for the coming year.
Data science activities around operations are necessary to conduct day-to-day business. They involve the establishment of automated decision-making systems and the appropriate calibration of the inputs to these systems. For example, in a supply chain, such a system could be a forecasting system that relies on machine learning algorithms or an inventory management system that relies on operations research techniques.
Data science activities around strategy are used to identify new opportunities from data or potential risks that need to be mitigated. Some application examples include pricing strategy, long-term scenario planning, or product lifecycle strategy.
Driven by data science, pursue growth amid economic uncertainty
When deciding to fund your data science project, take a look at where you are in the economic cycle. My definition of a business cycle is based not only on real GDP growth, but also on the state of consumer spending, the labor market, inflation, and the ease with which businesses can access debt/credit to fund innovation and households to fund consumption.
Keep in mind that the ROI of infrastructure and scientific capabilities will vary with time/complexity and the maturity of the organization 
When the economic cycle is at its trough, leadership has traditionally emphasized operational considerations at the expense of strategic considerations. This necessity is primarily driven by maximizing short-term shareholder value “in the next year.”What I’m trying to say is that this strategy will yield suboptimal results
Keep in mind that the ROI of your infrastructure and scientific capabilities will vary with time/complexity and the maturity of your organization9
It feels like this will hurt your long-term growth prospects; let me explain. The most important decisions are often made in a procyclical manner, meaning they are essentially made based on current economic conditions to optimize short-term gains. Ideally you want to identify new opportunities and risks in a counter-cyclical manner well in advance (ideally at the peak of the economic cycle or at the latest on a slight downturn) to give you the opportunity to implement the relevant action plans on time. You also want to understand who your customers are and be able to predict how they will react at different stages of the economic cycle. This way you can take advantage of the upside of the economic cycle without missing out on the upside momentum.
science and infrastructure
Where is my return on investment?
First I’ll show you my view on ROI profiles as a function of time/complexity from the time you invest (I assume they get more complex over time ).
When I talk about infrastructure, I’m referring to the sophistication of the data science pipeline, which includes data warehousing solutions, ETL tools, and software engineering capabilities that facilitate large-scale scientific development and deployment (whether it’s MLOps capabilities, simulation capabilities, or real-time state-of-the-art Optimization). I will leave the cloud vs. on-premises debate to a future article. When you invest in data science infrastructure, it’s sometimes difficult to see the benefits right away. You start with data infrastructure; then, you move on to improving modeling capabilities and, finally, improving model orchestration and performance monitoring processes. With end-to-end automation of complex modeling tasks, the value of your infrastructure ultimately becomes more apparent as you can translate it into SG&A savings. If done right from the beginning, you can minimize technical debt and create actionable value for the future as your infrastructure can be extended to other business applications.
When I talk about science, I am referring to the level of sophistication of analytical abilities. It starts with simple data visualization or statistics, all the way to custom algorithms developed specifically for your use case. Science is also about your people. There are many types of data scientists, including (but not limited to): machine learning scientists, predictive modeling experts, operations research scientists, optimization experts, and quantitative economists who often sit at the intersection of these fields. Quantitative economists can be used to solve operational and strategic problems because they combine skills in causal reasoning, experimentation, and statistics. They can estimate causal parameters used as inputs to decision-making systems or support the identification of risks and opportunities using custom econometric models tailored to your specific problem and data. Scientific results can be quantified almost immediately, especially if your organization has a lower level of data science maturity. However, as scientific complexity increases, scientific ROI may not scale linearly with time/complexity. It may plateau as you continue to explore alternative scientific methods, but it will definitely rebound when you invest in features customized to your business and work on understanding your customers.
in conclusion
Pursuing growth in uncertain economic conditions requires understanding the economic cycle and who your customers are so that you can predict how they will react during each phase of the economic cycle. My advice is to always reserve some funds for your strategic data science capabilities to stay innovative so you can take advantage of the upswing in the economic cycle.
Keep in mind that the ROI of infrastructure and scientific capabilities will vary with time/complexity and the maturity of the organization. The more mature your organization is, the more challenging it will be to generate the incremental benefits you’re looking for, but as long as you focus on understanding customer behavior and adapting, it’ll be totally worth it in the end.
