Ghosts in the algorithm
Microfinance was once sold as a revolution. Its evangelists promised precision-targeted capital delivered to the "bottom of the pyramid" — a phrase that, in the optimism of the 2000s, seemed to imply the pyramid might eventually be flattened. A generation later, the pyramid stands largely intact. India has digitised its banking infrastructure with genuine speed and ambition. What it has not managed to digitise away is caste.
The persistence of caste-based financial exclusion in an era of algorithmic lending is not merely an irony. It is an indictment — of institutions that have adopted the aesthetics of modernity while preserving its oldest hierarchies, of regulators who have measured financial inclusion by the number of accounts opened rather than the fairness of credit extended, and of a political culture too timid to name discrimination when it wears a risk score rather than a saffron thread.
The transition to algorithmic lending was supposed to be the great equaliser. By replacing the gut feelings of a loan officer with objective data models, the industry hoped to launder prejudice out of credit scoring. The research says otherwise. Algorithms do not simply reflect the world as it is; they inherit the world as their trainers perceive it.
Frontline loan officers remain the primary gatekeepers of the data that feeds lending models. Their Social Dominance Orientation — a psychological measure of preference for group-based hierarchy — shapes what they record, how they frame it, and which applicants are quietly filtered out before the algorithm has a chance to act. A Dalit applicant rejected at the desk never appears in the model's training set as a successful borrower, even if she would have been one.
"Given its intangibility, statistical discrimination has not received the level of attention or condemnation that other forms of marketplace discrimination have."
— Meshram & Venkatraman, audit of 14,235 rejected loan applicationsThat word — intangibility — deserves sustained attention. When bias is delivered by a human with a briefcase and a sneer, it can be named and challenged. When it is delivered by a risk score formatted to two decimal places on a bank's internal dashboard, it acquires the sanctity of mathematics. Institutional prejudice has found, in algorithmic lending, its most durable disguise.
"Institutional prejudice has found, in algorithmic lending, its most durable disguise — a risk score formatted to two decimal places."
When formal banks fail marginalised communities, the informal market steps in — but with a predatory logic that makes the failure of inclusion actively profitable. OBC households participate in credit markets at 60%, and SC households at 56%, compared with 46% for General Caste households. These are not signs of financial confidence. They are signs of desperation routed through whichever channel will take them.
Research by Sangwan and Saha identifies what they call a "negative unexplained component" in lending to OBC borrowers: moneylenders extend credit to this group beyond what their observable economic characteristics would justify. The explanation is not generosity but enforceability. Marginalised borrowers, lacking the social capital to resist informal collection methods — panchayat pressure, reputational shaming, communal ostracism — are, from the moneylender's perspective, excellent credit risks. The very powerlessness that makes banks reject them makes informal lenders seek them out.
India has, in other words, constructed a two-tier credit system in which marginalised borrowers are too risky for the bank and too easy to exploit for the loan shark. This is not a market failure. It is a market design.
For most excluded groups, the primary barrier to formal credit is self-exclusion: communities with long histories of rejection eventually stop applying. This "learned hesitation" is structural — the scar tissue of systemic discrimination. Reservation policies can mandate fairness in approval rates once an application arrives. They cannot mandate the act of applying in the first place.
Scheduled Tribe communities carry a heavier burden still. Unlike SC or OBC applicants, for whom application rates are the dominant bottleneck, ST borrowers who do apply face dramatically lower approval rates — 77%, compared with 85–88% for other groups. Their exclusion is therefore double-gated: a social barrier that suppresses applications, and an institutional barrier that rejects those who clear the first hurdle. Economic success does not reliably dissolve either gate.
The persistence of these biases is inseparable from a simple fact: the people who design and operate India's lending institutions look almost nothing like the people those institutions are meant to serve. Bahujan representation at senior leadership levels in public and private banks is vanishingly thin. Reservation policies, where they exist, operate at entry level. The apex remains a monoculture.
The consequences are not incidental. A credit-scoring culture built and managed entirely by dominant-caste professionals will treat lower-caste identity as a proxy for risk — not necessarily through explicit prejudice, but through the quieter mechanisms of shared assumptions about collateral, character, and creditworthiness. It is a system that perpetuates itself by ensuring that the people most harmed by it have the least say in how it operates.
India's Gramin Bank SC-ST Welfare Association has documented a hostile internal work environment that prevents marginalised employees from reaching the decision-making roles where reform might originate. Congress leader Rahul Gandhi has noted, with characteristic understatement, that SC/ST employee advancement is frequently blocked "under the pretext of performance issues." This is not anecdote. It is policy by other means.
Y Puran Kumar, 52, was a senior Dalit police officer in Haryana. He took his own life, leaving an eight-page note naming the state's Director-General of Police and the Rohtak Superintendent of Police as architects of a sustained campaign of harassment. His note documented the denial of leave to attend his father's final rites, the withdrawal of his official vehicle, and annual appraisals "loaded with biased comments." Kumar's rank — attained through decades of service — provided no immunity from the social hierarchy his institution refused to reform. His case is not an outlier. It is the logical endpoint of a system that treats professional advancement as a threat to the natural order rather than its fulfilment.
The Delhi High Court recently ruled, in a case involving Axis Bank, that the SC/ST Prevention of Atrocities Act cannot be invoked to curtail a bank's lawful mortgage rights. The ruling is defensible on narrow legal grounds. As a policy signal, it is dismal. It reveals the extent to which social protection law and commercial law inhabit parallel universes — and the extent to which the courts, when forced to adjudicate between them, tend to find in favour of property.
True inclusion cannot be legislated in piecemeal interventions against individual bank officers. It requires structural reform of an institution — an entire financial system — whose incentives, leadership pipelines, training cultures, and algorithmic infrastructure have all been allowed to encode the same hierarchy that India's constitution formally forbids.
"India has constructed a two-tier credit system: marginalised borrowers are too risky for the bank and too easy to exploit for the loan shark. This is not a market failure. It is a market design."
The question of whether a truly inclusive financial system is possible without first dismantling caste is, at one level, unanswerable. At another level, it has already been answered. India has had decades of financial deepening, policy reform, digital infrastructure build-out, and rhetoric about reaching the last mile. The Dalit borrower is still at 66.25% of rejected applications. The Scheduled Tribe applicant is still being turned away at 77%. The senior Dalit officer is still leaving an eight-page note.
The algorithm has not fixed this. It has hidden it. And hiding is not the same as solving.
Comments
Post a Comment