Credit Risk Modeling Using Non-traditional Data: The Experience of Ping An OneConnect Bank Custom Case Solution & Analysis

1. Evidence Brief

Financial Metrics

The SME financing gap in Hong Kong is estimated at approximately HK$ 440 billion.
Traditional banks require 1 to 2 months for SME loan processing and demand physical collateral or audited financial statements.
PAOB is one of 8 virtual banks licensed by the Hong Kong Monetary Authority (HKMA) in 2019.
Tradelink Electronic Commerce Limited, the primary data partner, handles over 80% of the relevant trade declaration market in Hong Kong.
The 9-5-1 lending model targets: 9 seconds to apply, 5 minutes for approval, and 1 day for fund drawdown.

Operational Facts

PAOB utilizes alternative data including import/export volumes, frequency of trade, and payment history from Tradelink.
The bank employs Machine Learning (ML) and Artificial Intelligence (AI) to automate credit scoring, replacing manual underwriting.
Target customers are SMEs that have been in operation for at least 3 years and are active users of Tradelink services.
The digital-only infrastructure removes the need for physical branches, reducing fixed operational overhead.
Data integration is achieved via secure APIs, ensuring real-time or near real-time data flow for risk monitoring.

Stakeholder Positions

Ryan Fung (CEO, PAOB): Prioritizes financial inclusion for underserved SMEs and speed of execution as the primary competitive advantage.
Ping An Group: Views PAOB as a strategic extension of its OneConnect technology platform to prove the efficacy of its credit modules in a global financial hub.
HKMA (Regulator): Encourages virtual banking to promote innovation but maintains strict requirements on data privacy and risk management.
Tradelink: Acts as a critical data gatekeeper, seeking to add value to its own SME user base through integrated financial services.

Information Gaps

The specific default rates (NPL ratio) during a full economic contraction are not yet documented in the case.
The exact revenue-sharing agreement or data acquisition cost paid to Tradelink is not disclosed.
Long-term retention rates for SMEs once they grow large enough to qualify for traditional bank loans are unknown.

2. Strategic Analysis

Core Strategic Question

How can PAOB scale its SME lending volume and diversify its risk profile beyond the Tradelink ecosystem without compromising its 9-5-1 speed or increasing default rates?

Structural Analysis

The SME lending market in Hong Kong is characterized by high barriers to entry for traditional players due to information asymmetry. Using the Jobs-to-be-Done lens, SMEs do not just want a loan; they want liquidity that matches the pace of their trade cycles. Traditional banks fail because their cost-to-serve for a HK$ 1 million loan is nearly identical to a HK$ 100 million loan. PAOB breaks this via automation.

The competitive rivalry is increasing as other virtual banks (e.g., Mox, WeLab) and traditional incumbents (HSBC, Standard Chartered) digitize their processes. However, PAOB’s specific focus on trade-related alternative data creates a niche moat that is difficult to replicate without similar ecosystem partnerships.

Strategic Options

Option 1: Ecosystem Expansion. Form new partnerships with e-commerce platforms (e.g., Shopify, Amazon sellers in HK) and logistics providers to capture non-trade SME data.
Trade-offs: Increases data diversity but requires significant engineering resources for disparate API integrations.
Resource Requirements: High business development and data science capacity.

Option 2: Technology Export. Transition from a pure-play bank to a Banking-as-a-Service (BaaS) provider, white-labeling the credit scoring engine to traditional banks.
Trade-offs: Generates high-margin fee income but potentially empowers direct competitors.
Resource Requirements: Low capital, high legal and software documentation effort.

Preliminary Recommendation

PAOB should pursue Option 1. The bank must move beyond Tradelink to avoid concentration risk. By integrating with point-of-sale (POS) providers and utility companies, PAOB can build a 360-degree view of SME cash flow, making the credit model more resilient to sector-specific shocks in the trade industry.

3. Implementation Planning

Critical Path

Month 1-3: Secure data-sharing agreements with at least two non-trade partners (e.g., a major HK logistics firm and a digital payment aggregator).
Month 4-6: Develop and test the cross-sector credit model. The focus must be on normalizing data points from different sources into the existing AI engine.
Month 7-9: Launch a pilot lending program for SMEs in the service and retail sectors.
Month 10-12: Full-scale rollout and integration of real-time monitoring alerts to manage the expanded portfolio.

Key Constraints

Data Silos: Different partners use varying data standards. Normalization is the biggest technical hurdle.
Regulatory Compliance: HKMA’s Open API Framework requirements must be met for every new data stream to ensure customer consent and privacy.
Talent Scarcity: Competition for data scientists with both credit risk and ML expertise in Hong Kong is intense.

Risk-Adjusted Implementation Strategy

To mitigate the risk of model failure in new sectors, PAOB will implement a shadow-scoring phase for the first 90 days of any new partnership. During this period, loans will be capped at 50% of the standard limit. This allows the AI to calibrate against actual repayment behavior before full capital deployment. Contingency plans include a manual override protocol if the automated NPL triggers exceed 3% in any new segment.

4. Executive Review and BLUF

BLUF

PAOB must diversify its data sources immediately to sustain growth. The current reliance on Tradelink creates a strategic bottleneck and concentration risk. Success depends on the ability to ingest and normalize disparate data sets from logistics and retail sectors while maintaining the 9-5-1 delivery promise. The transition from a single-partner lender to a multi-platform credit engine is the only path to long-term viability in the virtual banking space. Speed of integration is the primary metric of success.

Dangerous Assumption

The analysis assumes that trade data from Tradelink is a perfect proxy for general SME creditworthiness. If the predictive power of trade frequency diminishes due to global supply chain shifts, the current model will fail across the entire portfolio simultaneously.

Unaddressed Risks

Adverse Selection: As traditional banks digitize, they may retain the highest quality SMEs, leaving virtual banks with a pool of borrowers that have been rejected elsewhere, despite what alternative data suggests.
Model Opacity: There is a risk that the AI model identifies correlations that do not imply causation, leading to systemic errors that are difficult to audit during a regulatory review.

Unconsidered Alternative

The team did not evaluate a pivot to consumer lending. While the SME gap is large, the customer acquisition cost for consumers is lower, and the data (credit cards, mobile bills) is more standardized than SME trade records. This could provide a faster path to profitability if SME scaling stalls.