When optimisation models ignore the cost of being wrong — and how to embed financial consequence into analytical design.
Analytics models are evaluated on accuracy, precision, recall, and a range of statistical performance measures. They are rarely evaluated on the economic cost of their errors. A model that is 95% accurate on a classification task may produce errors whose economic consequences are asymmetric — false positives costing ten times more than false negatives, or vice versa.
When the cost structure of errors is not embedded in the model design, the model optimises for the wrong objective. It produces technically impressive outputs that generate economically suboptimal decisions.
In clinical research, a false negative — failing to detect a real effect — may mean delayed treatment. A false positive — detecting a spurious effect — may mean exposing patients to an ineffective intervention. These costs are not equivalent, and a model that treats them as equivalent is not fit for clinical decision-making.
The same asymmetry appears in credit risk, marketing optimisation, and fraud detection. Every domain has its own error cost structure, and embedding that structure into the model's objective function is a prerequisite for economic relevance.
Optimisation models that maximise a single performance metric without economic constraints frequently produce solutions that are locally optimal and globally destructive. A marketing mix model that maximises short-term revenue attribution may recommend channel allocations that erode brand equity over a multi-year horizon.
Adding economic constraints — margin floors, customer lifetime value thresholds, regulatory compliance requirements — transforms an optimisation exercise from a mathematical problem into a business decision framework.
Embedding financial consequence into analytical design requires collaboration between analytical teams and the business functions that bear the cost of analytical errors. This collaboration is not a one-time calibration — it requires ongoing feedback loops that update the economic parameters as the business environment changes.
The organisations that do this consistently produce analytical systems that are trusted by decision-makers, because the systems are designed around the same economic objectives that the decision-makers are accountable for.
The economics blind spot is correctable. It requires expanding the definition of model quality to include economic consequences alongside statistical performance — and building the cross-functional structures that make this expansion operational.
Explore the full library of 8 essays on decision intelligence, governance, and analytical architecture.