SME credit decisioning has run on the same dataset for two decades: filed accounts, bureau data, and where available, current-account transaction history. The dataset is structurally lagging. Filed accounts are 12 to 18 months stale by the time they reach the lender. Bureau data captures defaults after they have happened. Transaction history is partial. Payment behaviour observed across an AP network gives lenders a forward-looking signal the existing dataset cannot. The question is how to use it.
The current SME credit dataset
Three data sources dominate SME credit decisioning in the UK.
Filed accounts. Statutory accounts filed at Companies House, typically 9 to 18 months old by the time the lender consumes them. They cover the period before the credit request, not the period during which the lender will be exposed.
Bureau data. Credit bureaux aggregate defaults, CCJs and similar adverse events. The data is reliable but reactive. By the time a business shows up on the bureau as distressed, the underlying behaviour has been deteriorating for months or years.
Transaction history. Where the lender has visibility into the borrower's current account through open banking or a banking relationship, transaction history adds operational colour. Useful but not predictive on its own.
None of these forecast forward. The lender's view of the borrower is structurally backwards-looking. We have made the broader case for behavioural data in payment reputation as collateral.
Why payment behaviour is a strong forward indicator
Payment behaviour from the AP side, observed across multiple counterparties, is one of the strongest leading indicators of SME financial distress.
The pattern that signals distress is consistent. Payment timeliness against terms slips first, often by days rather than weeks. Bank-detail change frequency rises. Suppliers report longer chase cycles. Exception rates on invoices increase. The cumulative pattern is visible in the data 60 to 120 days before the same business shows up in bureau records.
The same pattern in reverse signals recovery. A business that has been late and starts paying cleanly again is materially less risky than its bureau record would suggest. The bureau will catch up eventually, but the operational data leads.
The data shape that makes this work is network-level. A single buyer's view of a supplier's payment behaviour is too thin. Cross-counterparty observation, the same shape as the supplier identity graph, is the source of the signal.
The data shape lenders need
Lenders consuming payment behaviour data need three things.
Consent. The borrower has consented to the lender accessing the behavioural data, with the consent recorded and contestable. The consent model parallels open banking, with adjustments for the B2B context.
Verifiability. The data carries provenance. The lender can trace each data point to its source counterparty and date. The data is auditable in the same way bureau data is auditable. This is the discipline laid out in payment behaviour should be underwritten, not promised.
Longitudinality. The behavioural data covers a meaningful time series, not a snapshot. Trend matters more than current state. Six to twelve months of history is the operational minimum.
Where these three are in place, the data is admissible alongside the existing dataset. Where they are not, the data is operationally interesting but cannot be embedded in a credit decision that has to be defensible to a regulator or to the borrower.
Two emerging product patterns
Behaviour-tiered terms. The lender offers different terms to borrowers based on their payment behaviour across the network. A borrower with strong, stable behaviour receives lower rates, higher limits or longer terms. A borrower with deteriorating behaviour receives a price that reflects the forward risk. The mechanism is transparent and contestable, which distinguishes it from black-box credit scoring. The transparent-scoring discipline is the same one in supplier trust scoring.
Behaviour-conditional credit lines. Credit lines that scale with continued behavioural performance. Where the borrower's payment behaviour remains within a defined range, the line is extended. Where it deteriorates, the line is reviewed and potentially reduced. The product structure is well-suited to revolving credit, particularly receivables-backed facilities.
Both patterns require the data to be live, not periodic. Lenders need current observation, not last quarter's report.
The governance question
Who can use the data, under what consent, for what purposes? The question is not theoretical. It is the gating issue for the lending application of payment behaviour data.
The cooperative model addresses this directly. The borrower retains consent rights. The data is shared with the lender for the defined credit purpose, with contestation mechanisms in place. The lender's use is bounded. Secondary uses (resale, model training, derivative products) require separate consent.
Without the governance frame, the lending application risks two failure modes. Over-collection, where the lender consumes more data than the credit purpose requires. And drift, where the data flows into uses the borrower did not anticipate. Both are addressable in the design, but neither happens by default.
Where this is going
The pathway from operational AP data to financial-market-grade lending data is well-defined but not finished. The next twelve to eighteen months will see the first integrated products, with consent and governance built into the design. Lenders that build now have a structural data advantage that bureau-only lenders will struggle to match. The category-level case sits in the network effect in accounts payable.
The shift is similar in shape to the move from filed accounts to bureau data in the 1980s, and from bureau data to open banking data in the late 2010s. Each shift unlocked a class of product that the prior dataset could not support. Payment behaviour data is the next dataset in the sequence.
.jpg)
