Skip to content

ACCA Strategic Professional · Strategic Business Leader · Enabling success: disruptive technologies

Brelton Bank uses a machine-learning model trained on ten years of lending data to approve loans. An internal review shows approval rates are significantly lower for applicants from certain postcodes, although postcode is not a stated criterion. The CEO wants to keep the model because its default prediction is accurate. Which assessment is most appropriate for the board?

The board should treat this as proxy bias carried over from historical data and act: review model inputs, test outcomes across groups and keep human oversight. Predictive accuracy alone does not remove legal, reputational and ethical risk, and confidentiality or wholesale data deletion are not appropriate answers.

  1. AKeep the model, because accuracy on default prediction is the only relevant test
  2. BRemove all historical data, since past data is always invalid
  3. CTreat it as proxy bias inherited from historical data, requiring review of inputs, outcomes testing and human oversight, given legal, reputational and ethical exposureCorrect
  4. DDisclose nothing, since the model's logic is commercially confidential

Explanation

Postcode can act as a proxy for protected characteristics, and models trained on historical decisions can reproduce past bias. Accuracy alone does not remove discrimination, ethical or regulatory risk. The board should review features, test outcomes across groups and maintain oversight. Discarding all data is excessive and confidentiality does not excuse unfair outcomes.

Did you get it right without looking?

One question tells you little. A timed set on Enabling success: disruptive technologies shows your real accuracy, how long you take and where you lose marks.

Practise Enabling success: disruptive technologiesFree diagnostic mock

More Enabling success: disruptive technologies questions