Can causal analysis change business? Applying causality in AI and beyond

Neil Raden Profile picture for user Neil Raden October 16, 2023
The surge in enterprise AI has fueled interest in causal analysis. In this piece, I explore the threads that bind cause and effect - and how they can be applied across a range of industry scenarios.


The concept of causality is a one central pillar governing our understanding of the world. Causality involves delineating the relations between cause and effect, demystifying how and why phenomena occur by unveiling the threads that connect events, leading to informed predictions and deeper understandings.

This article is part one, covering the philosophical and scientific foundations of causal inference, statistical and machine learning approaches in general, and the largest section on applications in business decisions. The companion article, Part 2, Methods of Casual Inference, will take a deep dive into the techniques and methodologies of causal analysis.

Philosophical foundations of causality

Modern philosophical approaches, influenced by thinkers such as Hume and Kant, focused on human cognition and the limitations of our understanding in delineating cause and effect. Hume emphasized the empirical observation of events, claiming causality is inferred through constant conjunction, while Kant felt that causality is a fundamental framework through which we understand the world, transcending mere observation.

The philosophical foundations of causality delve into the deep-seated questions regarding cause and effect, exploring the nature, sources, and conditions of causation. Philosophers have long debated how we identify causal relations, what makes them necessary, and what kind of claim we're making when we say something caused something else.

Over the centuries, as our scientific understanding has advanced and our conceptual frameworks have evolved, so too have our philosophical interpretations of causation. While no single account is universally accepted, these foundations collectively shape how we think about causality in diverse contexts, from everyday life to cutting-edge scientific research.

David Hume asserted that our concept of causality arises entirely from experience. When we repeatedly observe that events of one sort (A) are followed by events of another sort (B), we begin to associate As with Bs. Over time, this regular succession of events gives us the idea of causality. However, Hume was also skeptical of our ability to know causal necessity. He argued that we never directly observe the necessity itself, only the constant conjunction. Rooted in Hume's philosophy, regularity theories argue that causation is nothing more than a regular pattern of events. A causes B if and only if whenever A happens, B happens as well.

Counterfactual theories of causation suggest that an event A caused an event B if, had A not happened, then B would not have happened either. This idea explores causation in terms of "what if" scenarios or alternate realities. Another perspective on causality suggests that we understand causal connections in terms of our ability to manipulate one event to bring about another. Such a view ties closely with experimental practices in science, where a controlled manipulation of one variable can show its effect on another. Given the inherent randomness in many physical systems (especially in quantum mechanics), some philosophers argue that causation should be understood in terms of probabilities. An event A might not always lead to B but might increase the probability of B occurring. On the other hand, pragmatic, and epistemic approaches, especially those influenced by scientific practice, argue that our understanding and identification of causal relationships are deeply tied to our practical goals, explanatory aims, or available knowledge.

Scientific Perspectives

In Newtonian Mechanics, causality emerged as a foundational principle, where actions are linked to equal and opposite reactions, forming a deterministic system that predicts outcomes based on initial conditions. Quantum Mechanics introduced a realm where causality was not always so straightforward, where particles could exist in multiple states and only settle upon being observed, introducing inherent uncertainty in the causal analysis.

In Epidemiology, the Bradford Hill criteria include nine viewpoints by which to evaluate human epidemiologic evidence to determine if causation can be deduced: strength, consistency, specificity, temporality, biological gradient, plausibility, coherence, experiment, and analogy. This established a framework for inferring causality from observed associations, encompassing facets such as the strength and consistency of the association, and the biological gradient, bringing causality into the heart of medical science.

Statistical Approaches and Machine Learning

Correlation Does Not Imply Causation: in statistics, it has been emphasized that "correlation does not imply causation," underlining the importance of distinguishing between a mere association between variables and a causal relationship. Developments like the Granger causality tests extended these principles further, offering statistical methods to infer causal relationships by observing temporal patterns in data.

Machine learning provides causality renewed interest, as researchers grapple with the challenges of understanding and leveraging causal relationships in artificial intelligence systems. Causal inference frameworks aim to uncover causal relationships from data, venturing to use these understandings to make robust predictions and drive decision-making processes in a myriad of applications.

Applications in Business Decisions

Causal analysis stands as a cornerstone in shaping customer retention strategies and price elasticity models. By understanding the intricate cause-and-effect relationships that drive customer behavior and demand patterns, businesses can implement strategies that resonate with their target audiences, optimizing retention rates and pricing models. Whether it's through A/B testing, predictive analytics, or root cause analysis, causal inference offers a lens into the dynamics of the market, steering businesses towards informed, effective, and strategic decisions. As the business landscape continues to evolve, the integration of causal analysis promises a pathway marked by sustainable growth, customer loyalty, and optimized revenue streams.

Causal analysis in marketing optimization

By analyzing data before and after a marketing campaign businesses can understand advertising impact and can measure the direct impact of their advertising efforts on sales, website visits, or brand awareness. This helps allocate resources more effectively. Typically, A/B testing, exposing two groups to different versions and measuring their responses, allows marketers can causally link certain changes to outcomes. Marketers can test variations of ads, emails, or landing pages to understand what drives better results. By

Customer segmentation discovers which marketing strategies work for which customer segments. This understanding is causal, as it ties specific marketing actions to outcomes within defined groups so businesses can personalize their marketing efforts.

Case Study: Coca-Cola frequently adjusts its marketing strategies based on causal analysis of campaigns. For instance, by analyzing sales after a specific ad campaign, they can determine its effectiveness and adjust future marketing efforts accordingly.

Supply chain management

Demand forecasting is foundational is supply chain management. By analyzing causal factors like marketing campaigns, seasonal events, or economic trends, businesses can more accurately predict product demand. Inventory management is key to supply chain performance and understanding the causes of inventory shrinkage, whether due to theft, mismanagement, or other factors, can help businesses optimize stock levels.

Supplier performance evaluates and predicts which suppliers consistently deliver on time, provide quality products, or offer the best terms, helping businesses optimize their supplier relationships. Causal analysis can take performance analysis beyond simple reporting to identifying the characteristic of good performers.

Case study: Toyota's "Just-In-Time" supply chain management relies heavily on understanding causal relationships. By analyzing how specific factors (like a machine breakdown or a delay from a supplier) impact the overall production, they can optimize their processes for maximum efficiency.

Product development

By using controlled tests where one group uses the old product version and another uses the new version, product testing provides businesses with a causally link of product changes to outcomes like user satisfaction or product performance. Feature importance, understanding which features drive user engagement or satisfaction can help prioritize development efforts. Causal analysis can differentiate between features users engage with because they love them, and features users engage with out of necessity. And  of course, after launching a new product or feature, Post-Launch causal analysis can determine its success by comparing key metrics (like sales or user engagement) before and after the launch.

Case study: Microsoft frequently employs causal analysis in its product development. For instance, before rolling out a major Windows update, they might test it on a smaller user group. By comparing the experiences of this group to a control group using the older version, they can causally link changes in the update to user outcomes.

Customer retention strategies

Customer retention refers to the ability of a company or product to retain its customers over a specific period. High customer retention means customers of the product or business tend to return to, continue to buy or in some way not defect to another product or business. Understanding root causes before implementing retention strategies, businesses must understand why customers are churning. Causal analysis can identify the root causes of churn, whether it's due to product issues, customer service challenges, pricing, or external factors.

Predictive analytics, using historical data, can predict which customers are most likely to churn in the future. This prediction isn't just about identifying high-risk customers but understanding the causal factors that increase the likelihood of churn.

Case study: churn is a continuing issue for telecommunications. When Company A noticed a spike in customer churn, using causal analysis, they found that a significant portion of these customers faced service disruptions in the past month. By addressing these disruptions proactively, the company reduced churn rates significantly.

Causal analysis in price elasticity models

Price elasticity measures how the quantity demanded or supplied of a good change with its price. It’s a critical concept in economics and is used by businesses to optimize pricing strategies.

Causal analysis helps businesses understand how sensitive demand is to price changes. For instance, if raising the price of a product causes a significant drop in sales, the product has high price elasticity. On the other hand, lowering prices may require a significant lift in sales, otherwise profitability may suffer. External factors, that influence price elasticity, competitive reaction or economic downturns, might make consumers more price sensitive. Causal analysis can help businesses understand external factors.

Not all customers respond to price changes in the same way. Causal analysis can segment customers based on their price sensitivity, allowing businesses to implement tiered pricing strategies. Much like in customer retention strategies, A/B testing can be used to understand the causal impact of price changes. By offering different prices to different groups and observing demand, businesses can accurately

Challenges in establishing causality

Complexity of systems: one of the challenges in establishing causality is the complexity of systems we aim to analyze. Many phenomena, particularly in the social sciences and medicine, are influenced by a plethora of factors, making it difficult to isolate individual causal relationships. The intricate nature of causal relationships means that causal analysis can be a complex endeavor. Leveraging technology effectively necessitates a deep understanding of the underlying principles of causal inference

Ethical considerations: furthermore, ethical considerations often limit our ability to perform experiments that would conclusively establish causality, necessitating the use of observational studies and the challenges that come with them.

Talent acquisition: the advanced nature of causal analysis technologies necessitates a skilled workforce capable of navigating these tools, highlighting the importance of talent acquisition and training.

Future directions in causal research

Technological advancements, including artificial intelligence, open up avenues for causal discovery, with machine learning algorithms capable of analyzing vast datasets to unearth previously unseen causal relationships. But promoting causal literacy in education could empower individuals to navigate a world of information overload more effectively, distinguishing between correlation and causation, and making informed decisions based on a nuanced understanding of causal relationships.

By leveraging data, controlled tests, and rigorous analysis, businesses can disentangle complex relationships, identify key drivers of success, and chart a course for future growth. These endeavors in marketing optimization, supply chain management, and product development underscore the profound impact of causal analysis on shaping business strategies and ensuring their success.  Soon we will see more automated solutions in causal analysis, where AI and ML technologies will offer businesses the ability to quickly and efficiently derive causal insights without the need for deep expertise in the field.

My take

Causal analysis stands as a powerful pillar in modern business decision-making, a beacon guiding businesses towards informed and foresightful strategies. Leveraging current technologies, businesses can navigate the complex landscape of causal relationships, transforming data into a goldmine of actionable insights.

In a world inundated with data, the ability to discern the causal forces at play can be a game-changer. From optimizing marketing strategies to refining product development processes, causal analysis technologies are reshaping the business landscape, offering a pathway to success grounded in data-driven insights and nuanced understanding of causal relationships.

As businesses stand on the cusp of a new era of causal analysis, the integration of advanced technologies promises a future where causal insights are at the fingertips of decision-makers, driving businesses s towards sustained success.

The road ahead is one of exciting possibilities, where causal analysis technologies will continue to evolve, offering businesses unprecedented insights and fostering a culture of informed decision-making grounded in the principles of causality.

In this dynamic landscape, the businesses that will thrive are those that embrace the power of causal analysis, leveraging current technologies to forge paths grounded in understanding and foresight. By delving into the causal forces that shape outcomes, businesses can unlock the full potential of data, steering towards a future of informed strategies and sustained success.

Causal analysis, empowered by modern technology, holds the reins of the future, driving businesses towards a horizon of unparalleled insights and strategic foresight. Let us embrace this evolving field, as we forge ahead in a world where data is not just king but a guiding star, leading the way to informed and innovative business strategies, honed by the powerful lens of causal analysis.

A grey colored placeholder image