In the context of the rapid introduction of artificial intelligence (AI) and machine learning algorithms into the marketing activities of companies, the importance of the problem of algorithmic bias is growing. This bias, reflecting the social and historical inequalities embedded in the training data, can systematically distort the principles of fairness, reduce the economic effectiveness of marketing decisions, and lead to serious corporate and regulatory risks. The purpose of the article is to systematize the mechanisms of bias in the most common marketing models — segmentation, response forecasting, recommendation systems and dynamic pricing, as well as to assess the consequences for businesses and consumers, with a special focus on the realities of the developing market in Kazakhstan. The methodology is based on a structured analysis of current Scopus publications (2020-2024) and a comparison of international approaches with the specifics of Kazakh practice. For the first time, a conceptual model combining "source of bias — mechanism — marketing and social consequences — adjustment measures" is presented. The results demonstrate that algorithmic biases are systemic in nature, amplified by limited local data, linguistic heterogeneity, and an imbalance of digital footprints (as in the case of regional discrimination or proxy bias). Uncontrolled bias translates into financial, reputational, and legal risks for companies. The directions of reducing bias adapted to the regulatory framework of the Republic of Kazakhstan are proposed: audit of data explained by AI, development of equity metrics and implementation of management measures (creation of cross-functional groups). The study forms the basis for subsequent empirical work in Kazakhstan, contributing to the transition to transparent and fair models.