Algorithmic bias refers to systematic errors or distortions in artificial intelligence models that lead to unfair or discriminatory outcomes.

In other words, digital discrimination manifests itself as unequal treatment of certain groups of people through automated decisions based on biased data or AI models. Such biases arise when algorithms are trained on historical data that contains existing stereotypes and imbalances, or when developers unintentionally embed biases into the functioning of the system. As a result, artificial intelligence can produce false or biased conclusions, even though it operates on large amounts of data.

The problem is that such biases are not visible at first glance, but can significantly influence business decisions, especially in marketing. The MIM:AGENCY team conducted an analysis and found out how exactly this affects, what the danger is, and how to counteract this phenomenon.

How do AI biases affect marketing?

Algorithmic biases have a broad impact on various tools and platforms used in modern marketing.

Let’s look at the main areas where biased algorithms can cause harm:

Advertising and targeting

In digital advertising, algorithms decide who to show ads to and how to allocate budgets. If there is bias in the model, ads may be shown to a too narrow or one-sided an audience, excluding certain consumer groups. For example, studies have shown that Facebook’s algorithm distributed job openings for “male” professions (such as forestry) predominantly to men, while cleaning job ads were seen mostly by women. Such biases arise even without the direct intention of advertisers – the system itself learns to show ads to those who, in its opinion, are more likely to click. As a result, a potentially valuable audience may be lost, and the company may miss out on customers.

Content personalization and social media

Marketing increasingly relies on personalized content: social media news feeds, product selections, or articles tailored to the user’s tastes. Biased personalization algorithms can create information bubbles and reinforce stereotypes. Similarly, social media algorithms can promote content that matches a user’s past behavior while showing fewer alternative viewpoints or content from minorities, thereby narrowing the audience’s horizons. This is risky for brands, as their message may not reach part of their target audience due to hidden bias in the news feed or content recommendations.

E-commerce recommendation systems

Algorithms for recommending products and services, such as those used by Amazon, are also prone to bias. They suggest products based on past user behavior data. If this data contains historical inequality, the recommendations will reproduce it. For example, the system may recommend different products to men and women even without directly specifying gender, simply because that is how it has historically been. As a result, certain categories of products may not be shown to female audiences, and male audiences may not see ads for home or personal care products if the algorithm has decided that they are not interested in them.

Recruiting platforms

Employer branding and HR technologies also suffer from algorithmic biases. Automated resume screening systems should make the hiring process more objective, but in practice, this is not always the case. A well-known case: the algorithm was trained on resumes submitted over 10 years, most of which belonged to men (reflecting the gender imbalance in the IT industry). As a result, the system “decided” that men were better candidates and began to downgrade applications that contained the word ‘female’ (for example, in the phrase “captain of the women’s chess club”), and stopped recommending graduates of some women’s colleges altogether. This case clearly demonstrates how biased data and models lead to biased results, even if the attribute (gender) is not explicitly specified.

Thus, algorithmic biases can permeate various aspects of marketing – from who to show ads to and what content to offer users, to how to filter job candidates. Each of these areas requires attention so that automation does not turn into injustice.

Algorithmic bias in the Ukrainian market

Have there been similar situations in Ukraine? So far, there have been no high-profile scandals specifically related to algorithmic discrimination in domestic marketing or online advertising. However, this does not mean that the problem does not exist – rather, it is still under-researched and not always recognized. Ukrainian businesses are gradually adopting global practices: more and more companies are using AI to work with customer data, advertising platforms are the same (Facebook, Google), and use the same algorithms as abroad. In the HR sphere, the use of AI is also growing – large international corporations in Ukraine use automated selection systems, and local job search sites integrate smart algorithms. For example, the robota. ua portal has launched AI functions to generate interview questions and job descriptions.

Interest in the topic is also growing at the societal and state levels. The concept of digital (algorithmic) discrimination is already present in Ukraine’s legal discourse. Experts note that traditional laws do not yet contain direct norms against algorithmic bias, but the need for regulation is growing. The Concept of Artificial Intelligence Development, adopted by the government in 2020, declares the principle of non-discrimination, although so far, this is more of a declaration than a working mechanism. Therefore, Ukrainian companies currently operate under conditions of self-regulation and must take care of the ethical use of algorithms themselves.

Although there have been no high-profile cases in Ukraine, the risks are similar to those around the world. For example, marketers may find that the targeting system selects the audience in a biased manner. Or that the recommendation algorithm on the website filters out products in Ukrainian due to a lack of data, or the priority of Russian-language content in the past. Such biases are difficult to notice without special analysis, but they affect both consumers and business results.

How to reduce algorithmic bias: recommendations for businesses

Algorithmic risks can and should be proactively mitigated. Marketers and companies using AI should implement a series of steps to avoid or minimize bias. MIM:AGENCY has created several recommendations to help make AI tools more fair and safe for brands.

High-quality and diverse data for training

Most biases enter the model through data, so it is critical to monitor its quality. It is necessary to collect and use the most representative data possible, including different user groups. Avoid outdated or biased datasets that skew the model. As SAS Marketing Director Jennifer Chase aptly noted:

“Biased data and biased models lead to biased results.”

Therefore, algorithms should only be “fed” with verified, balanced data that reflects the real audience without bias. In practice, this may mean, for example, collecting additional data on a less represented group of customers before training the personalization model.

Regular auditing and testing of algorithms

You cannot run an AI system “on autopilot” – you need to constantly monitor its decisions for bias. Companies should implement audit processes: periodically check the results produced by the algorithm and look for anomalies. For example, analyze how ad impressions are distributed among different demographic groups to see if there are any systematic deviations. Today, there are special tools available – for example, Google has released Dataset Search to find more balanced data sets, and frameworks such as IBM AI Fairness 360 or TensorFlow Fairness Indicators help identify biases in data and models.

If a problem is identified, take immediate action: retrain the model on new data, adjust the algorithm, or decision-making rules. It is advisable to conduct audits on an ongoing basis because even a correctly trained model can “drift” in an undesirable direction over time.

The “human in the loop” principle

Full automation is not an end in itself, especially when it comes to decisions that have a serious impact on people. It is worth introducing human control at critical stages. For example, if AI has generated a list of candidates for a vacancy or an audience segment for a campaign, let HR or a marketer do the final review.

Several large companies have already declared a rule: important decisions that significantly affect people’s lives should not be made autonomously by an algorithm.

Humans can see what machines cannot “understand” – for example, that all candidates on the list are of the same age for some reason, or that an ad is not being shown in a certain region. We discussed the question of who understands humans better – machines or humans – in more detail in the article “Can AI understand consumers better than humans?”

Ethics policy and team training

Marketing and product teams should develop clear internal rules for the use of AI. These policies should include principles of non-discrimination, transparency requirements (as far as possible), and incident response procedures. It is important to train staff, conduct seminars on the ethical use of data, explain how algorithmic biases manifest themselves, and why this is a problem. If the whole team is aware of the risks, they are more likely to notice and correct them. A culture of responsible AI must come from leadership: top management should emphasize that ethics are no less important than KPIs. For a brand, this is an investment in long-term trust: consumers value those who act proactively, rather than just reacting to scandals after the fact.

Involving diverse teams and experts

Monotony in a team of developers or marketers can lead to “tunnel vision”–everyone thinks alike and can overlook biases. Therefore, it is worth striving for team diversification, especially those responsible for data and algorithms. People from different backgrounds are more likely to notice biases that are invisible to others.

The broader the perspective when developing and testing an algorithm, the less likely it is that unconscious bias will become embedded in it.

Limit the use of AI where the risks are high

Analyze scenarios for using artificial intelligence in your marketing and business. Identify areas of increased risk – for example, hiring decisions, granting credit, or evaluating a customer for an important service. If the consequences of a mistake are very serious for a person or potentially illegal, it is better not to rely entirely on the algorithm. It is worth leaving such decisions to people or using AI only as a supporting tool. As practice shows, sometimes “less is better”.

Conclusions

Algorithmic bias is not a death sentence or a reason to abandon artificial intelligence. It is more of a call for awareness. Marketers have always worked with human biases – studying stereotypes, preferences, and social effects. Now it is time to work just as carefully with machine biases.

Businesses that can combine AI with ethics and equality will not only have loyal customers but also a competitive advantage. After all, in a world where data and automation are a priority, the brand that wins is the one that consumers trust because they know they are being treated fairly and impartially.

Sources:

  1. SAS. AI Bias in Marketing: Risks and How to Avoid Them. SAS Institute, 2023.
  2. U.S. Department of Justice. Justice Department Secures Groundbreaking Settlement with Meta to Resolve Allegations of Algorithmic Bias. DOJ Press Release, 2022.
  3. MIT Technology Review. Discrimination in Online Ads: The Facebook Case. MIT Tech Review, 2022.
  4. Ukrainian Marketing Association. Digital Discrimination: Challenges for the Ukrainian Market. UMA, 2022.
  5. Robota.ua. AI Tools for Recruitment and Selection. Official Robota.ua blog, 2023.