How AI Is Transforming Credit Scoring and Lending

The Algorithm That Decides Who Gets Capital

When Mashreq Bank announced in March 2026 that its AI-driven credit engine had reduced loan default rates by 34% across its UAE retail portfolio, every chief risk officer in the Gulf paid attention. But the second number, buried deeper in the disclosure, was the one that mattered more: approval rates for first-time borrowers had simultaneously risen by 22%. The old trade-off between risk and access — the one every banker accepted as iron law — was being dismantled by machine learning models that see what traditional scorecards cannot.

Across the Gulf states, South-East Asia, and sub-Saharan Africa, artificial intelligence is rewriting the mechanics of credit assessment. This is not an incremental shift. It is a structural overhaul of how lenders evaluate borrowers, price risk, and deploy capital — with real consequences for private wealth managers, family offices, and institutional investors seeking exposure to consumer and SME lending.

Beyond FICO: The New Architecture of Creditworthiness

Traditional credit scoring relies on a narrow set of financial inputs — repayment history, outstanding debt, length of credit history. In markets where 1.4 billion adults remain unbanked, according to the World Bank's 2025 Global Findex update, these inputs simply do not exist. AI-powered models ingest thousands of alternative data points instead: mobile phone usage patterns, utility payment records, e-commerce transaction histories, even geolocation stability.

Saudi Arabia's Tamara, the buy-now-pay-later platform valued at $1.2 billion after its 2025 Series D round, processes over 8,000 data signals per application. Its proprietary scoring model — built on gradient-boosted decision trees and neural network ensembles — evaluates applicants in under 400 milliseconds. The company reported in its Q1 2026 earnings that net credit losses fell to 1.8% of gross merchandise value, down from 3.1% two years earlier. Chief executive Abdulmajeed Alsukhan told investors the model's predictive accuracy now exceeds 94% on 90-day delinquency forecasting.

In Nigeria, Carbon — formerly Paylater — has extended over $500 million in cumulative loans using an AI scoring system trained on mobile behavioural data from 4.5 million users. The platform's models assess keystroke patterns, app usage frequency, and contact list diversity as proxies for social stability and financial discipline. Carbon's non-performing loan ratio sits at 4.2%. For unsecured consumer lending in a market where conventional banks report NPL ratios above 10%, that number is remarkable.

The Gulf's Strategic Bet on Intelligent Lending

Gulf Cooperation Council states are pouring money into AI-augmented financial infrastructure, driven by national diversification agendas and ambitions to become global fintech hubs. The Abu Dhabi Global Market's 2026 Fintech Report identified AI credit scoring as the fastest-growing segment in the emirate's financial technology sector, with 17 licensed firms now operating in the space — up from four in 2023. That is a significant shift.

First Abu Dhabi Bank deployed an AI credit decisioning platform developed with Afiniti in January 2026, integrating it across personal loans, credit cards, and SME facilities. Processing time for SME loan applications dropped from 12 days to 47 minutes. The bank says it maintained approval accuracy within regulatory risk tolerances set by the Central Bank of the UAE.

Qatar's Dukhan Bank took a different route. It partnered with Singapore-based CredoLab to layer psychometric and device-intelligence scoring onto its existing framework. The hybrid model targets expatriate workers — a demographic historically underserved by Gulf banks due to thin credit files and transient residency. Few outside the region have noticed, but Dukhan reported a 40% increase in expatriate lending volumes in the first quarter of 2026 with no corresponding rise in early-stage delinquencies.

Private Wealth and Family Offices: A New Asset Class Emerges

For family offices and private wealth allocators across the Gulf and Asia, AI-scored lending portfolios are becoming an increasingly attractive alternative asset class. The logic holds up under scrutiny: machine learning models generate granular, real-time risk stratification that allows for precise portfolio construction with predictable yield profiles.

Mumtalakat, Bahrain's sovereign wealth fund, disclosed a $75 million allocation to Lendable, a London-based platform that uses AI to originate and manage consumer loan portfolios across Kenya, Nigeria, and South Africa. The fund's 2025 annual report noted that the allocation delivered a net return of 11.3% in dollar terms, with Sharpe ratios exceeding most liquid fixed-income alternatives.

Dubai-based Shorooq Partners, a venture capital firm backed by prominent Emirati and Saudi family offices, has invested in five AI lending startups since 2024, including Jordan's Liwwa and Egypt's Kashat. Managing partner Shane Shin has argued publicly that AI-scored credit portfolios in frontier markets offer "the risk-adjusted returns of private equity with the liquidity profile of fixed income" — a proposition that resonates with family offices hunting for yield without excessive lock-up periods.

Several single-family offices in Riyadh and Kuwait City have begun building dedicated fintech lending allocations of between $20 million and $100 million, according to three wealth advisors operating in the region who spoke on condition of anonymity. These allocations typically sit within alternative credit sleeves and are structured through special purpose vehicles that purchase tranches of AI-scored loan books.

Regulatory Reckoning and the Bias Question

The rapid adoption of AI scoring has drawn regulatory scrutiny. The Saudi Central Bank — SAMA — issued draft guidelines in February 2026 requiring all licensed lenders using algorithmic credit models to submit annual bias audits conducted by independent third parties. The guidelines, expected to be finalised by September, mandate that AI models demonstrate statistical parity across gender, nationality, and income brackets.

The concern is not theoretical. A 2025 study by the Brookings Institution found that AI credit models trained predominantly on male borrower data systematically underscored female applicants by 6 to 11 percentage points in markets across the Middle East and North Africa. Correcting these biases without degrading model performance demands sophisticated fairness-aware machine learning techniques that remain in relatively early stages of deployment. Nobody has fully cracked this yet.

The Central Bank of Bahrain has adopted a sandbox approach, granting conditional licences to AI lenders that agree to real-time model monitoring and quarterly recalibration. The European Union's AI Act, which took full effect in February 2026, classifies credit scoring as "high risk," imposing transparency obligations that are already shaping how Gulf-based firms with European exposure design their systems.

What is emerging from this collision of technology, capital, and regulation is a credit ecosystem that moves faster, reaches wider, and — if properly governed — operates more equitably than its predecessor. The institutions and investors who grasp the mechanics of these models, and the risks embedded within them, stand to capture returns that legacy lending infrastructure can no longer generate. Everyone else will be playing catch-up.