“If you know the enemy and know yourself, you need not fear the result of a hundred battles.” -Sun Tzu
Business processes result in the production of tremendous amounts of data which pose multiple administrative issues and opportunities. Customers service intake, account management, purchasing behavior, inventory states, quality control, accounting records and product returns obscure basic details critical to business decision making, such as “how profitable is the company” and “how can we be more profitable?”. This data may be incredibly valuable if useful insights can be extracted to drive change within the organization. For example, venture capital funds identified that when they ranked their most successful ventures, the return on investment fell off rapidly with each additional company following a Pareto distribution. Their top investment returned as much money as the rest of the fund’s investments put together [1]. Funds now only invest in companies which they believe will make them more money than the size of the entire fund [1]. Identifying these insights and leveraging them to drive change in your company behavior requires creativity on the part of the analyst and techniques from a variety of scientific disciplines. The analyst supports company strategy by being the mechanism through which the company can know itself.
An analyst cannot merely grab a single data point and ask themselves why it is occurring, nor can they throw the data into a predictive machine learning model and expect to understand the cause of different behaviors across the business. Neither of these approaches identify variables that cause certain business behaviors. Instead, causal techniques utilize counterfactual instances in observational data to isolate the drivers of change to singular causes [2]. Since the business collects so much data, any potential changing data may have many contributory factors. Building metrics and KPIs support isolating these factors but cannot be the end of the story. Statistical techniques and predictive machine learning models allow us to understand if changes in metrics and KPIs were significant, how large the changes were, anticipating customer behaviors and purchasing requirements from historical data, and projecting business costs [2]. Predictive factors should not be confused for causal factors, as a variable which is highly predictive for a certain outcome may change with the true underlying cause alongside the outcome. Causal inference is much harder to conduct and may not improve with additional data in the way machine learning can. Causal inference also requires some creativity by building a conceptual model of what may be happening under the hood, establishing metrics and tests to confirm or invalidate the concept, then using supported concepts to answer why some behavior is occurring. Some causal inference tests can include A/B testing, quasiexperimental design, statistical matching, difference-in-difference modeling, and sophisticated recommendation algorithms such as uplift modeling for marketing campaigns and content delivery [2]. Each of these techniques can contribute to existing business strategy and inform future directions.
How do we go about knowing ourselves? First, business information must be accessible. Companies like Meta maintain distinct analytics databases where information from other processes get stored and accessed without interrupting critical business processes. Second, a conceptual model about behavior in business must be developed. We can break down these conceptual models into units of responsibility like departments or individuals, the entire business, or even processes outside of the business. In the case of a marketing department, you might say “advertisements effect revenue”. You can use that to inform a testable hypothesis like “increasing ad spend on YouTube leads to greater revenue”. You can then build metrics to measure this hypothesis, like measuring revenue and money spent on YouTube ads over time. Lastly, an experiment must be developed and implemented to check if increases in YouTube ad money causes an increase in revenue after the intervention. If this hypothesis is supported, you can measure by how much each dollar of YouTube ads makes in new revenue. This amount then serves as an actionable insight into a mechanism by which the business can be more profitable. This is just one of many potential project iterations that drive continuous improvement, inform efficient internal processes, improve profits.
[1] Thiel, P. A., & Masters, B. G. (2015). Zero to one: Notes on startups, or how to build the future. Pg.85 Currency Books.
[2] Rodrigues, J. (2021b). Product analytics: Applied Data Science Techniques for actionable consumer rights. Addison-Wesley.