Why women in Africa’s services industry must outperform men to overcome customer bias

Editor’s note: For a broader synthesis of themes covered in this article, check out Issue 1 of our VoxDevLit on Female Labour Force Participation.

Discrimination in the workforce based on age, gender, race, or other protected characteristics can have wide-ranging negative effects on both individuals and organisations. It can harm hiring, wages, and promotion practices, as well as reduce worker satisfaction, damage organisational culture, and lower overall productivity. Workplace discrimination also exposes enterprises to legal risks and reputational damage.

While many studies have examined workplace discrimination, they have typically focused on employer behaviour, identifying disparities in wages and hiring outcomes among individuals with similar productivity levels. However, this approach has limitations, as it does not fully account for other forms of discriminatory treatment that can impact productivity (Glover et al. 2017, Sarsons 2022, Egan et al. 2022).

Less well understood is how customer biases affect their interactions with employees and how these biases influence the worker productivity we ultimately observe. Service workers are often evaluated based on customer interactions, with performance frequently measured by overall sales. However, their performance can be negatively impacted by customer bias related to worker characteristics, such as gender, which is the focus of our study. While individual instances of customer bias may seem minor or go unnoticed, their cumulative effect can be significant.

Our study seeks to better understand how discriminatory customer behaviour influences service workers' labour market outcomes.

Context: The service sector in sub-Saharan Africa

1. The Service Sector: Our study takes place in the service sector in sub-Saharan Africa. Service sector jobs have grown significantly over the past two decades. Between 2011 and 2019, the share of working-age individuals employed in services increased by 12% across the region. Women have driven much of this growth, with their share of the workforce rising by 16% (World Bank 2022). As internet connectivity continues to expand across the continent, service-sector workers are increasingly engaging with customers online. Online shopping has also grown rapidly, increasing by 18% annually between 2014 and 2017 (UNCTAD 2018).
2. The company we work with: We partnered with a travel agency operating throughout sub-Saharan Africa (in countries where female labour force participation is relatively high and comparable to that of men). The agency facilitates flight and hotel bookings through local sales agents who interact with clients via phone and online chat. Within the travel agency studied, two-thirds of the sales force are women, and they account for 83% of client interactions—consistent with industry standards. Customers initiate interactions with sales agents by clicking a chat button on the agency’s webpage. Clicking the button opens a chat window displaying the agent’s first name and automatically sends a short, formulaic greeting message from the agent to the customer. After this initial exchange, either the agent or the customer can send subsequent messages.

We analysed marketplace interactions over a nine-month period between six sales agents and over 2,000 customers, 87% of whom were from African countries. The agents worked in one of two offices located in the West African and East African regions, respectively.

Capturing the impact of customer biases on workers

1. The randomised control trial: We assigned agents randomly chosen names with implied genders each day. Customers received no additional information about the agents’ personal characteristics beyond the assigned names. We also ensured that agents were unaware of their assigned names. This was achieved using a web plugin that masked the names from the agents’ view, ensuring their behaviour was not influenced by their knowledge of name assignments.
2. The methodological contribution: This research design overcomes challenges with two common experimental methods to study discrimination: audit studies and correspondence studies. Audit studies—in which actors, who are as similar as possible except on one dimension, engage in a task like applying to the same job—struggle to control for all other differences between the actors. Actors may also be subject to “demand effects” because they are aware of their treatment status and do not have real incentives to perform well. Correspondence studies—in which fictitious applications with different sounding names are sent to a possible discriminator like employers—can only measure indirect outcomes such as job application callback rates rather than actual job hiring (Bertrand and Duflo 2017). In our setting, the daily name randomisation eliminates omitted variable concerns, and the name masking alleviates concerns about demand effects. Furthermore, by studying real workplace interactions we can collect the ultimate outcome measures of interest—e.g. the likelihood of purchase—and specifics of the interaction—e.g. customer engagement with the sales agent.

Findings: Discrimination by customers negatively impacts workers

1. Customers purchase less from women: We found that randomly assigned female names reduced the likelihood of customers making any purchase by 3.8 percentage points—a 50% reduction relative to the baseline purchase rate of 7.6%. Additionally, we observed declines in both the number and value of purchases made.
2. Customers are less engaged with women: An analysis of agent-customer interaction transcripts revealed that while customers did not bargain differently with female-named agents or engage in harassing behaviour, they exhibited greater disinterest in engaging with them. For instance, customers responded more slowly to agents with female names, often requiring multiple follow-ups before replying, and were less likely to transition from general inquiries about flights or hotels to actual purchases. These patterns were consistent among clients from both Africa and outside the continent.
3. Women must out-perform male counterparts: We then leveraged company data to explore how male and female sales perform outside the scope of the experiment. We found that agents were equally productive when evaluated based on the value and number of sales made. However, our study demonstrates that women achieve these results despite facing significantly more discrimination. This suggests that female agents must perform at a higher level to achieve parity with their male counterparts. In effect, absent customer bias, female employees would be more productive than their male peers.
4. This discrimination may be internalised by companies: This raises interesting questions: why would a company employ female workers when customers are more likely to make sales when engaging with men, and why would female workers choose to stay? The first question may be explained by the firm's perception that female workers are equally productive as men. As a result, the company may not view itself as incurring a loss by hiring female workers at wages similar to those of male employees. These findings align with the behaviour of a non-discriminatory firm that has internalised the discriminatory preferences of its customers. Second, female workers may remain in these jobs for two main reasons. First, they may face worse discrimination in other occupations, limiting their alternatives. Second, female workers may possess a comparative advantage in this role or derive non-monetary benefits from the work.

Regardless of the reasons, our experiment highlights significant constraints that prevent workers from escaping discrimination and inhibit the optimisation of their potential productivity.

Implications for how policy can protect women in the workplace

1. Changing customer norms: From a policy perspective, the most direct approach to tackling this problem is to change customer norms around women in the workplace. Governments may use programmes that increase the representation of women in positions of power, exposing the general population to women as authority figures. Similarly, if firms believe they could capture future benefits by sensitising customers or choose to because of pro-social intentions, these same firms may seek to change customer norms themselves. Nevertheless, norm change is likely to be a difficult and slow process, and since many customer services positions are already predominantly female, simply hiring female sales representatives is unlikely to lead to norm change.
2. Interim measures: While changing cultural norms is a long-term endeavor, interim measures can be implemented to mitigate the effects of discrimination. One strategy, as shown in prior research (Chan, 2022), is to obscure agents' identities. In our context, using gender-neutral names or omitting names altogether could eliminate the effects of discrimination. Another approach would be to avoid individual-based incentive pay schemes tied to the number and value of sales. While such measures could reduce inequality, they also potentially perpetuate the bias that creates these inequalities in the first place.

References

Bertrand, M, and E Duflo (2017), “Field experiments on discrimination,” Handbook of Economic Field Experiments, 1: 309–393.

Chan, W (2022), “The AI startup erasing call center worker accents: Is it fighting bias or perpetuating it?” The Guardian.

Egan, M, G Matvos, and A Seru (2022), “When Harry fired Sally: The double standard in punishing misconduct,” Journal of Political Economy, 130: 1184–1248.

Glover, D, A Pallais, and W Pariente (2017), “Discrimination as a self-fulfilling prophecy: Evidence from French grocery stores,” The Quarterly Journal of Economics, 132: 1219–1260.

Kelley, E M, G V Lane, M Pecenco, and E A Rubin (2024), “Customer discrimination in the workplace: Evidence from online sales,” National Bureau of Economic Research Working Paper 31998.

Sarsons, H (2022), “Interpreting signals in the labor market: Evidence from medical referrals,” Job Market Paper, 141–145.

UNCTAD (2018), “UNCTAD B2C E-Commerce Index 2018: Focus on Africa,” Technical report.

World Bank (2022), “World Development Indicators,” https://databank.worldbank.org/source/world-development-indicators.