Issue 17 / Home

August 22, 2022
An abstract image of bubbles that look like frog eggs, which contain a scene that appears to be a view out a peephole

Apartheid by Algorithm

Julien Migozzi

The infrastructural legacies of apartheid have made modern South Africa into an especially stark illustration of the New Jim Code.

It’s a crisp Saturday morning in Khayelitsha, Cape Town’s largest black township, and a thirty-something woman named Noxolo wants to buy a home. She spots a property on a local real estate agency’s Facebook page. It looks rather nice for a starter home. But property is so expensive these days, even for a salaried state nurse. To put a foot on the property ladder, Noxolo will need a mortgage.

To get a mortgage, Noxolo must go through a credit check. So she gives her payslips, bank statements, and a copy of her national ID card to the secretary at the real estate agency. Once scanned, all of the documents are emailed to a mortgage originator, who will then try to find a bank to give Noxolo a loan.

A couple of days later, the answer comes back to the agency in the form of an encrypted PDF. All the banks have declined: Noxolo’s credit score is too low. “She did not qualify,” the secretary tells me as she throws the documents into a large pile of declined applications stored in a box back in the office kitchen. Noxolo’s dream of homeownership will have to wait—at least until she can get her score back on track by settling a late payment owed to a clothing store, which the algorithm picked up right away.

Noxolo’s story illustrates how access to housing in South Africa requires navigating a set of filtering mechanisms structured on a continuous flow of data orchestrated by credit-scoring technologies. This data is abundant, and viewing it is easy. All you need is someone’s ID number and a few rands to get a full picture of their financial situation: gross salary, net salary, existing loans, credit accounts, late payments, and so on. These various inputs are used to calculate a credit score that has become, on the formal market, the ultimate arbiter of housing access in South Africa. A good score is crucial for getting a mortgage and, increasingly, even to rent.

Real estate agents, banks, and corporate landlords buy data mined by credit bureaus for the purpose of selecting home-seekers. Screening prospective residents is nothing new in housing, especially for the mortgage industry. But the vetting is now done with scoring algorithms and automated dashboards.

This phenomenon isn’t limited to South Africa, of course. Real-estate technologies are becoming popular in other countries as well, especially the United States. Institutional landlords have entered the single-family rental market, relying on software to automate property and tenant management, as documented by the geographer Desiree Fields. Meanwhile, fintech mortgage lenders tend to perpetuate racial disparities in their lending practices, despite promising to “democratize” finance with technology, as the scholar Tyler Haupert has shown. Moreover, credit-scoring technologies have become increasingly elaborate, relying on more advanced statistical techniques like machine learning and incorporating more diverse streams of data. And, as these technologies have grown in scope and complexity, they have become more socially harmful: as scholars like Martha Poon, Marion Fourcade, and Jenna Burrell have pointed out, the new algorithmic scoring technologies amplify inequalities of race and class.

All of these dynamics are on display in South Africa. But there’s something distinctive about how they operate there. Algorithmic scoring technologies are especially advanced in South Africa because of their deep historical roots in racist social control. The country’s history of colonialism and apartheid has bequeathed an information dragnet of formidable sophistication and depth to the real estate and banking industries. And this information dragnet has enabled the re-segregation of housing markets in the post-apartheid era of formal equality. Contemporary South Africa thus presents an especially stark illustration of what the sociologist Ruha Benjamin calls the “New Jim Code.” It shows how supposedly neutral technical systems often rely on structures inherited from past eras of formal segregation, and work to produce similar results.

Skeletons of the State

From the first days of Dutch settlement in the seventeenth century, colonial institutions in South Africa gradually structured cities and societies on the basis of made-up racial categories. As the ideology of social separation gained ground in the late nineteenth century, the colony of the Cape of Good Hope adopted a racist classification system, used to sort the population into different racial groups endowed with different political rights, with the first modern census in 1865. This in turn generated the statistical data that guided the first efforts at formal segregation implemented by municipal elites in Cape Town at the start of the twentieth century, resulting in the creation of all-Black townships such as Ndabeni in 1901.

In 1913, the all-white parliament of the Union of South Africa, established in 1910 as a self-governing dominion of the British Empire, took a further step by passing the Natives Land Act, which proclaimed urban areas as white. Any Black people in cities were required to carry a permit justifying their presence. The Union also embarked on a new census to count its population using a complex classification scheme that labeled people into main categories such as “White,” “Native or Bantu,” and “Mixed and Coloured.”

Following the triumph of the National Party in the 1948 elections, the South African government began to implement apartheid. This regime advanced the art of population statistics to the next level, aiming to achieve a fully segregated society through strict control of the housing market. As part of the Population Registration Act of 1950, which categorized all South Africans according to race, the government created a centralized population register linked to a new national ID system, where two digits indicated the race of the cardholder according to a four-fold classification scheme.

In a similar vein, the Group Areas Act of 1951, another legal pillar of apartheid, sorted residential areas into exclusive racial zones: for a title deed to be approved, the race indicated on the buyer’s ID had to match the race assigned to the neighborhood. Home-ownership and mortgages were only accessible to those defined as Indian or Coloured—a loosely defined racial category encompassing brown people descending from the indigenous Khoisan, slaves from Africa and Asia, and their white masters—and only in their respective assigned neighborhoods. Black people—people belonging to ethnicities such as the Xhosa or Zulu—were denied property rights in urban areas until 1986, and only allowed to rent state-controlled housing units in designated all-Black townships. Homeowners and tenants whose government-determined racial identity did not match the one decreed for the neighborhood were disqualified and brutally evicted, as exemplified by the destruction of District Six in Cape Town from 1976 to 1981, where bulldozers annihilated decades of racial diversity.

From an early stage, digital technologies formed an integral part of this state-driven classification of people and places. The apartheid regime loved computers, as the South African historian Keith Breckenridge has shown. IBM computers were used to count and categorize people, to store millions of fingerprints, to issue ID cards, and to centralize databases. The government’s willingness and capacity to classify the population with digital tools survived the transition to democracy. The pretext was the need to identify the beneficiaries of welfare programs. In fact, South Africa pioneered the use of biometric governance: in the late 1990s, the government launched the Home Affairs National Identification System (HANIS), a smartcard program that directly expanded on the population register inherited from the apartheid era. In the name of fraud detection, banks were granted automated access to the HANIS database in 2010, using ID numbers and in-house fingerprint readers for real-time identity verification.

The technical infrastructures developed during apartheid and updated by the democratic state endowed the real estate industry with an algorithmic skeleton structured around ID numbers and devoted to the collection, storage, and processing of personal and property data. But to add the flesh, it would take the dramatic expansion of consumer credit markets. This happened in the 2000s, with the explosion of debt-driven mass consumption.

Consumer Flesh

In the late 1990s, the South African government undertook a number of neoliberal reforms that liberalized the economy. By the 2000s, the economy was growing quickly, turning South Africa into a major emerging economy alongside China and Brazil. Despite rapid growth, however, salaries began to stagnate. People turned en masse to loans to meet their consumption needs. The explosion of consumer credit led, in turn, to widespread overindebtedness. Most workers became trapped in a debt hole, at the mercy of loan sharks and microlenders.

The extent of the crisis was illustrated by the Marikana massacre in 2012, with the police firing at mining workers striking for a pay rise, resulting in thirty-four deaths and more than seventy severe injuries. A subsequent investigation revealed that most of the workers were drowning in debt, with loan sharks in possession of their bank cards and ID documents to pressure them. By then, the recession following the global financial crisis had set in, making South Africans even more reliant on credit as economic growth slowed.

The expansion of consumer credit in these years produced an unprecedented datafication of the population, structured around ID numbers. The rapid adoption of information technology by both the real estate and banking industries in the early years of the twenty-first century made it much faster and cheaper for the credit bureaus to generate data about borrowers. Each time a credit account was opened, each time a payment was made or ran late, the credit bureaus collected and stored the digital traces—bit by bit, this activity added up to compose a comprehensive picture of someone’s financial history. Moreover, the National Credit Act, implemented in 2007 in the name of fighting reckless lending, made it compulsory to conduct more thorough credit checks—whether to buy a car, a couch, or a house—and had the effect of further strengthening the position of credit bureaus as data collectors and data providers.

As one engineer excitedly put it to me, “South Africa has a wealth of data. We have a lot more data than what is available in some of the other countries.” In September 2021, the National Credit Regulator (NCR), a government agency, counted 26.42 million active credit consumers owning more than 85 million credit accounts—double the adult population. These accounts generate an immense amount of data for the credit bureaus, which real-estate and banking professionals use to determine who gets to own or rent a home, and where.

C and Below

During the apartheid era, the racial classification specified on ID documents would qualify or disqualify someone from home ownership and access to housing finance. Redlining was the rule for Black South Africans, who represent about 80 percent of the country today. In the era of formal equality, banks are legally required to cater to everyone. Yet mortgages remain a luxury product. Only people earning more than $1,000 dollars a month—less than 20 percent of the population—can hope to qualify.

But that’s not all. Avoiding late payments and other negative traces on your credit history is equally critical. In other words, the digital record of your financial behavior must be spotless. Such spotlessness is rare: according to the NCR, bad debt is prevalent in South Africa, with 38 percent of consumers in arrears on debt repayments. Banks now use algorithms to select borrowers based on affordability tests and credit scoring. They source the raw data owned by credit bureaus to feed into a scorecard system that helps them assess mortgage applications. The national ID number is the lynchpin of this process, used both to source consumer data and to access centralized government databases from police and tax departments. As one real estate agent summed it up to me, “Banks: they want your DNA.”

During fieldwork, I heard countless stories of clients disqualified by their credit history. One real estate agent told me:

I had a Black woman coming to the office, she was driving a massive 4x4. She was a magistrate, earning like 60,000 rands a month [$3,800 USD]. A very, very nice salary. She wanted to rent a house. I said to my agents, “There is no way she is going to rent. She is gonna buy.” What we discovered: she had a [debt collection] judgment because she had an account against her with Edgars [a popular retailer]. She was a magistrate, but she couldn’t get a bond [loan]!

Yet having no history is just as bad. Overindebtedness affects scores negatively, but so does the absence of credit history. Home-seekers must leave digital traces of debt to build up their score. They must be visible to the system.

This visibility is not just required for homeownership. It’s also increasingly required to rent. Indeed, South Africa has arguably pioneered the use of credit scoring for screening tenants. In the 2000s, new companies emerged to offer products for tracking tenants’ financial performance and rental history. First offering credit checks in the form of SMS queries using the applicant’s national ID number in partnership with established credit reporting companies such as Experian or TransUnion, these companies quickly evolved—on the legal side, into their own credit bureaus, and on the tech side, into rental platforms. As their customer base grew among real estate agencies, their databases increased in volume. Their dashboards now connect real estate agents to traditional credit bureaus, but also integrate payment data from previous rental leases and check for eviction court orders. Running a full report on someone costs less than $5 USD.

Segregation 2.0

Officially, there is no racial bias to this system. Racial discrimination in mortgage lending and renting is illegal. But if scoring algorithms are race-neutral on paper, in practice they produce the same sorting mechanisms that the apartheid regime used to maintain a segregated housing market in South Africa.

Credit scores are highly racialized. Bad credit scores are mostly held by Black and Coloured people. This is not only because of low salaries and precarious employment—which lead them to take out loans that they struggle to repay—but also because the real estate passed down through the generations has relatively little value. Housing values remain systematically lower in Black and Coloured townships, a direct legacy of the apartheid era. This hinders the ability of Black and Coloured people to access housing finance, leaving the divisions of the segregated city intact.

Conversely, being born into a white family means one may inherit property in the best neighborhoods. Coming from this privileged background enables people to leverage existing equity and family assets to soften the effects of credit screening by reducing the mortgage amount with a substantial deposit, or simply by paying the entire amount in cash. Most of the real estate agents I interviewed in the most affluent, and still predominantly white, areas of Cape Town describe these neighborhoods as a “cash market.” None of them seemed to recall a sale which failed due to bad credit.

On the rental side, the adoption of credit scoring is likewise reinforcing racial inequalities. It is doing so as part of a broader shift of the market towards a more rent-based structure. During apartheid and the early years of democracy, the rental market was considered a risky investment by institutional investors because of a long history of street protests against evictions, rent boycott campaigns, and other forms of housing activism. But by 2010, rising rental prices were making the market attractive to institutional investors, while new technology gave them ways to mitigate the risks of bad or unruly tenants. In particular, they could now conduct tenant screening with credit-scoring software, filtering out anyone who might become a problem. As a rental manager confessed to me, “We don’t take tenants under 622.”

Corporate landlords have appeared in Cape Town, Durban, and Johannesburg in the form of real-estate investment trusts, investment funds, and large-scale developers. These landlords like technology: screening tenants with credit-scoring software reduces the risk of late payments, minimizing vacancy rates, and optimizing their portfolio’s performance. As one rental manager explained to me, “We only take A and B type of tenants. ‘A’ means that you don’t have debt, so a very good profile. ‘B’ means that you have debt, but you are a good payer. ‘C’ and below, you are not good.” So far, the holdings of corporate landlords are not extensive: only a few thousand homes. Still, the racial impacts are clear. Extracted from a mostly Black and Coloured middle class priced out from the buyer’s market, rents are becoming a lucrative source of profit for largely white-owned and white-run firms.

Open Veins

South African capitalism has historically been rooted in the extractive industries. Its biggest companies and cities were born from mining: Johannesburg emerged from a gold rush, while Cape Town and Durban grew from the global export of minerals which, to this day, underpins the country’s economy.

Perhaps the datafication of home-seekers represents a new frontier in South African extractivism, transposed to the digital realm. Personal data, mined from the digital traces of people’s lives, has become a valuable commodity. And just as mining companies invest in new technical processes to drill into subsoil in search of new veins, credit bureaus dig deeper into people’s lives in search of “alternative data.” Court archives, tax records, credit accounts from retailers, bank accounts, telephone records, censuses, social networks, tax information—everything is excavated.

This extractive logic is, crucially, inseparable from the need to discipline workers and consumers. In the Marikana massacre, the mining conglomerates sent the police to shoot the protesting miners in order to force them back to work. Corporate landlords and banks use subtler means, deploying credit-scoring technologies to encourage consumers to act the right way. This is an important point: such technologies don’t merely record behavior, they shape it. As one rental platform CEO explained to me:

People have adjusted themselves amazingly to… getting access to credit. The product itself has produced a better consumer. The banking system and the rules are producing the results that it wants: a consumer that is aware of its financial position, and that will take responsibility for what they have committed to.

This may remind you of the famous episode of the popular TV show Black Mirror in which people rank each other on an app after every social interaction. Lacie, the main character, desperately tries to increase her score so she can rent a property in Pelican Cove, the estate of her dreams. (She fails, and is eventually evicted.) Have credit-scoring technologies turned post-apartheid South Africa into a real-life Black Mirror? Not yet. But you may be surprised to learn that the episode was shot in Pinehurst, in the heart of Cape Town’s affluent northern suburbs. The fictional Lacie and countless real South Africans share a common fate: a simple score can prevent them from calling a place home. Previously enforced by public institutions, segregation is now big business, driven by big data.

Julien Migozzi is a postdoctoral fellow at the University of Oxford, where he studies real estate technologies and urban segregation.

This piece appears in Logic's issue 17, "Home". To order the issue, head on over to our store. To receive future issues, subscribe.