AI bias is a pattern of unfair or skewed results from an AI system. It usually comes from the data the system learned from: if past decisions, reviews or customer records favour some groups or situations, the model tends to repeat that.
Bias can creep in at several points: unrepresentative training data, labels that reflect old human prejudices, features that act as stand-ins for things like age or postcode, or simply testing the tool on too narrow a group. The model does not intend anything; it reflects what it was given.
In retail and hospitality, bias risks show up in CV screening tools, automated rostering, loyalty offers that exclude some customers, voice ordering that struggles with certain accents, and fraud or loss-prevention alerts that flag some shoppers more than others.
To manage it, ask vendors how the tool was tested for fairness, test it on your own customers and staff, keep a human able to review decisions that affect people, and track complaints. Be especially careful with hiring and anything touching discrimination law; seek advice where needed.
It matters because biased outcomes harm real people and can create legal and reputational risk. A tool that works well for most customers but fails others quietly erodes trust and sales.