Skip to content

ACCA Strategic Professional · Strategic Business Leader · Finance transformation

Calder Insurance uses a machine-learning model to price motor policies. The model uses telematics, postcode and purchase data and gives lower prices to some groups. A board member asks the CFO whether the model's output can simply be accepted because its past predictive accuracy is high. Which response best reflects sound governance of analytics in decision-making?

The model should be kept under oversight: test it for bias, ensure it can be explained, document assumptions and keep management and the board accountable. High historical accuracy does not show fairness, because proxies such as postcode can produce discriminatory pricing outcomes.

  1. AAccept it, because high historical accuracy shows the model is free of bias and fair to all customers
  2. BReject it, because predictive models cannot be used in pricing decisions under any circumstances
  3. CAccept it, but only if the data science team alone owns all responsibility for the outcomes
  4. DRetain it with oversight: test for bias and explainability, document assumptions, and keep management accountable for outcomesCorrect

Explanation

Accuracy does not prove fairness; proxies such as postcode may produce discriminatory outcomes and opaque models hinder accountability. Good governance combines validation, bias testing, explainability and board and management accountability. Blanket rejection is unrealistic, and delegating all responsibility to the data team weakens accountability.

Did you get it right without looking?

One question tells you little. A timed set on Finance transformation shows your real accuracy, how long you take and where you lose marks.

Practise Finance transformationFree diagnostic mock

More Finance transformation questions