In a business environment where data abounds but decisions remain complex, the ability to transform information into concrete and transparent actions has become a key competitive advantage. Traditionally, organizations have relied on controlled experiments—such as A/B testing—to assess the impact of their policies. However, randomised trials are not always possible due to operational constraints, costs or ethical issues. This is where an innovative approach comes into play: prescriptive policies based on observational data, which allow inferring what action to take in each situation without the need to intervene directly in the real environment. This concept is at the heart of COAT (Counterfactual Optimal Action Tree), a framework that combines counterfactual estimation with large-scale optimization to make interpretable and actionable decisions under complex business rules. While the original use case focuses on pricing ancillary services in airlines—achieving significant increases in revenue—the underlying principles are applicable to any industry where personalization of decisions is critical: from the allocation of healthcare resources to the recommendation of financial products.
To understand the value of COAT, we must first explore what it means to learn a prescriptive policy from observational data. Unlike predictive models, which answer 'what will happen?', prescriptive models respond to 'what should we do?'. Let's imagine a bank that wants to offer a discount on a product to certain customers. If we use historical data where offers have already been applied non-randomly, past decisions are biased: perhaps customers with a longer credit history received better offers. To correct this bias, causal inference techniques are used that estimate what would have happened if a different action had been taken – the counterfactual result – and, from there, an action tree is trained that recommends the optimal action for each profile. The interpretability of the tree allows business leaders to understand and validate the rules, which is essential in regulated environments such as banking or health. In addition, the column-generation optimization process ensures that recommendations respect operational constraints such as maximum budgets, technical or regulatory capabilities, combining efficiency and transparency.
Implementing such frameworks would not be possible without a robust technology infrastructure. Businesses need systems capable of handling large volumes of data, running complex AI models, and deploying policies in real-time. This is where integration with cloud platforms becomes relevant: AWS and Azure cloud services offer scalable environments to train and serve models, while Business Intelligence tools such as Power BI allow you to visualize both the results and the rules generated, facilitating auditing and continuous adjustment. At Q2BSTUDIO, we understand that every organization has unique needs, so we develop custom applications that connect these capabilities. For example, we can build a system that ingests observational data from multiple sources, applies causality algorithms, generates optimized decision trees, and exposes them using secure APIs, all while adhering to the highest cybersecurity standards. Process automation and the use of AI agents for monitoring and updating policies complete the cycle, allowing decisions to dynamically adjust to market changes.
A crucial aspect of prescriptive policies is their interpretability. In sectors such as health or finance, it is not enough for the model to be right; The rules need to be understandable to professionals and, in many cases, explainable to regulators. Decision trees offer that clarity, as each node represents a condition (e.g., 'if the customer is under 30 years old and has a purchase history of less than 5 units, then offer a 10% discount'), allowing business teams to discuss and modify the rules collaboratively. In addition, the combination with counterfactual techniques reduces the risk of perpetuating historical biases, because the model learns from hypothetical results and not just from past decisions. For companies that already work with artificial intelligence for companies, incorporating this approach is a qualitative leap: they go from predicting behaviors to actively influencing them, maximizing objectives such as revenue, customer satisfaction or operational efficiency.
The pilot's success in the airline industry – with 6.9% increases in booking revenue and projections of $50 million to $150 million annually – demonstrates that investment in this type of solution has a tangible return. However, the real potential lies in its replicability. Any company that has historical data on decisions and outcomes can benefit, as long as it has the right technology in place to implement the full cycle: from data collection and cleansing to policy implementation. At Q2BSTUDIO we offer business intelligence services and custom software development to guide organizations on that path. In addition, our expertise in AWS and Azure cloud services ensures that the solutions are scalable and secure. Whether you need a proof of concept or a complete system, we can integrate everything from AI agent models to dashboards in Power BI that monitor the real impact of prescriptive decisions.
In conclusion, the ability to generate interpretable and optimal policies from observational data represents a significant advance in business decision-making. COAT is just one example of how the combination of causal inference, optimization, and transparency can transform the way companies interact with their customers and manage their resources. The key is to take a holistic approach that integrates technology, business, and ethics. At Q2BSTUDIO we are ready to help you take that step: from the design of custom applications to the implementation of artificial intelligence for companies, our multidisciplinary team can build the prescriptive system that your organization needs. The future of analytics is not just predicting, but acting with certainty.
