Integrating payment transaction data into lending decisions raises predictive accuracy in small-business credit by about 7 percentage points. It also delivers a 5 percent lift in return on investment when lenders use it to select the top 20 percent of loans. This represents a shift towards combining sources rather than choosing between them: static records of past repayment behaviour alongside high-frequency indicators of current economic activity.
Behind the aggregate gain, borrowers fare very differently.
Around 60 percent of firms see better risk assessment. Firms with strong credit histories gain most often. The more interesting winners are firms with weak credit histories but strong payment flows. Payment data picks up current performance their bureau record misses. Firms with both poor credit histories and thin payment records face sharper exclusion. Precision cuts both ways, and this tension is central to how interoperable financial data should be governed.
From Static Histories to Real-Time Signals
Traditional credit systems rely on bureau scores, numerical summaries of past repayment behaviour. These scores update infrequently and reflect long-term financial discipline. Payment data captures something different: ongoing business activity through sales volumes, transaction frequency, and volatility. These high-frequency signals map directly to a firm's current repayment capacity.
The empirical gains follow from this difference. Payment data alone matches credit bureau models on overall ranking ability. It does meaningfully better at flagging the rare delinquent cases that matter most for lenders. Combining the two improves both. In credit markets where margins are thin, this matters.
These gains arise because the two sources capture distinct dimensions of borrower behaviour. Payment data reflects capacity: whether a firm currently generates enough cash to service debt. Credit bureau data reflects intent: whether a borrower has historically chosen to repay. Each captures a different source of repayment risk.
A natural experiment makes the mechanism visible. For fintech loans made to the same borrowers, where repayments are deducted automatically from sales, borrower discretion is removed by design. Payment data retains its predictive power in this setting. Credit bureau data, once payment data is in the model, adds nothing. Bureau records earn their keep in conventional lending by measuring willingness to repay, the dimension that automated repayment shuts off.
Monitoring, Not Just Screening
Most discussion of payment data focuses on the lending decision itself. Once a loan is disbursed, payment flows continue to arrive. They carry information that traditional sources cannot match in frequency or timeliness.
The pattern is clear. Within 120 days of disbursal, post-loan payment data adds as much predictive power as pre-loan payment data did at screening. The most recent observations carry the strongest signal, and predictive power decays as data ages. Emerging distress shows up in the payment stream while bureau records, which update monthly at best, lag behind.
A loan becomes less a one-time screening decision and more an ongoing assessment. Lenders can intervene as conditions shift rather than after repayment has already broken down. This shapes what Open Banking should ask of data access: continuous flow over the life of the loan, not a single pull at origination.
Constraints Across the System
Continuous flow is the ambition. Several constraints stand between the ambition and its realisation.
On the economic front, interoperability requires investment in data standardisation and infrastructure. Granular transaction data must be harmonised across institutions through standardised APIs. These investments are non-trivial, especially for smaller lenders.
The data themselves cover only part of the picture. Payment records capture inflows but not outflows, expenses, or balance sheet conditions. The predictive gains documented here arise despite this partial coverage, which suggests the lower bound on payment data's value is already meaningful. Where firms operate across multiple payment channels, any single stream offers a partial view.
Institutionally, banks have been slow to adopt even existing data tools. Organisational silos, legacy IT systems, and regulatory uncertainty limit how payment data feeds into credit decisions. The constraint is not only the availability of data, but its incorporation into institutional processes.
The social consequences run alongside. Borrowers with thin digital footprints, often less formal enterprises, face worsening assessments. Those outside digital payment networks risk becoming statistically invisible, even when they are economically viable.
Distributional Effects and Market Reconfiguration
Statistical invisibility is one face of the broader pattern. The other is a reordering of credit access, not a uniform improvement.
Firms with strong payment flows but weak credit histories gain recognition and fit the classic definition an invisible prime. Their current performance, picked up by payment data, overturns the assessment their thin or damaged credit history would have produced. Firms with the opposite combination, weak credit history and weak payment activity, face sharper exclusion. Greater granularity in payment data reinforces both effects.
This reallocation reflects a transition from broad-based lending to finely segmented risk pricing. Credit becomes more efficient but less forgiving. As uncertainty is priced more tightly, the system reduces its tolerance for incomplete or informal economic activity, with direct consequences for inclusion.
From Data Access to Data Architecture
Pricing uncertainty tightly is one design choice. Whether the system can also extend access is another, and it depends on how the underlying architecture is built. The policy challenge is to move beyond enabling data sharing towards designing coherent data ecosystems.
A natural extension is broader integration. Payment data could be linked with other economic signals such as tax filings, supply chain transactions, and alternative credit indicators. This is a direction policy could take rather than a finding from current evidence.
Baseline inclusion matters in parallel. Mechanisms to bring cash-based or low-digital-activity firms into the data ecosystem are essential. Without this, interoperability risks reinforcing exclusion rather than expanding access.
Regulatory clarity is another lever. Clear protocols for data sharing, combined with safeguards on usage and privacy, can reduce institutional hesitation and support broader adoption.
Finally, data access rules should match the structure of the information. The strongest gains from payment data emerge over time, as borrower conditions are tracked across the lending lifecycle. Open Banking frameworks built around one-time access at origination capture only part of the value.
The Institutional Shift Ahead
Three design choices follow. Payment and bureau data work best as a joint system, since each measures what the other cannot. Open Banking rules should run across the lending lifecycle rather than stopping at origination. And interoperability needs companion mechanisms for firms with thin digital footprints, who remain most vulnerable to exclusion.