Two scores, two data sources, two different questions. A bureau score asks: “based on years of repayment history across registered credit accounts, how likely is this consumer to default?” A bank statement score asks: “based on three months of transaction data, do they have the income, capacity, and financial stability to meet a new obligation right now?” Neither question is optional. Neither answer is sufficient without the other.
For credit providers working under the NCA, understanding the distinction is not just conceptually useful. It has direct implications for Regulation 23A compliance, reckless lending exposure, and the quality of credit decisions across the full applicant population.
Two different questions, two different answers
The bureau and the bank statement operate on fundamentally different data and time horizons. Bureau data is backward-looking: it reflects what the consumer did with registered credit over months or years. Bank statement data is current: it reflects what the consumer’s financial life looks like over the past 90 days.
Both are legitimate risk signals. Both are incomplete on their own. The bureau cannot tell you what the consumer’s salary is, whether they have gambling habits, or whether they started stacking payday loans last month. The bank statement cannot tell you whether this consumer paid their accounts reliably for the past five years. A credit provider who uses only one has a partial picture.
The NCA’s affordability framework reflects this. Regulation 23A requires income verification from payslips or bank statements, not from bureau data, which carries no income signal at all. The bureau check satisfies a separate obligation (enumerating existing registered obligations). Together, they build toward a complete assessment. Separately, each leaves gaps that create both regulatory and credit risk exposure.
How bureau scoring works: the strengths and the gaps
A bureau score is calculated from the data that registered credit providers report to one or more of South Africa’s four credit bureaus: TransUnion, Experian, Compuscan, and XDS. The underlying data includes payment history on credit accounts (mortgages, vehicle finance, personal loans, credit cards, retail accounts), credit enquiries, account age, and credit utilisation.
The strengths are real. For consumers with substantial credit history, bureau data provides a stable, longitudinal view of how reliably they have managed credit obligations. A 780 bureau score on a 10-year file is meaningful. It reflects years of consistent repayment across multiple facilities and multiple lenders. That history is difficult to fake and correlated with future repayment behaviour on similar products.
The gaps are structural:
- Data lag. South African bureaus update 4–6 weeks after a payment event. A consumer who lost their job three weeks ago, or who has taken out three payday loans in the past month, may still show a clean bureau record. The score reflects who they were, not necessarily who they are.
- No income signal. Bureau data contains no income information. A consumer earning R8,000 per month and one earning R80,000 per month could have identical bureau scores. The score tells you nothing about affordability.
- Off-bureau obligations invisible. Informal lenders (mashonisas), unlicensed short-term credit providers, and BNPL providers who are not registered credit providers do not report to the bureaus. Real obligations that affect affordability are simply absent from the bureau file.
- Thin-file and no-file consumers unscored. TransUnion SA estimates that approximately 1.4 million consumers open new credit accounts each year with thin or no bureau history. For these consumers, the bureau cannot generate a reliable score, or generates no score at all. The bureau says “no data.” That is not the same as “high risk.”
How bank statement scoring works: what it adds
A bank statement score is derived from classifying and analysing the transactions in three consecutive months of bank statements. The core inputs are: income credits (salary, irregular income, transfers), debit orders (loan repayments, subscriptions, insurance), spending patterns, cash withdrawal behaviour, and balance dynamics.
What this approach adds over bureau data:
- Income verification. The statement shows what actually landed in the account each month. A payslip declares gross income; the statement shows net. Income stability, regularity of deposit dates, and variance across months are all measurable.
- Complete obligation picture. Debit orders to informal lenders, BNPL repayments, and stokvel contributions all appear as transactions even when they are absent from bureau data. Regulation 23A requires that “all monthly debt repayment obligations” be assessed; the statement is the only document that approaches that comprehensiveness.
- Behavioural signals. As explained in detail in our guide to behavioural credit scoring, signals like gambling spend, returned debit orders, cash withdrawal patterns, and low-balance days carry meaningful predictive weight. These are invisible to the bureau.
- Current-state accuracy. The statement is recent by definition. A three-month statement ending last week captures events that won’t appear in bureau data for another month or two.
The thin-file case: where bureau fails and statement succeeds
Consider a 27-year-old salaried employee who has been in formal employment for 14 months and has no registered credit history. No credit card, no personal loan, no retail account. The bureau returns “no data” or generates an indicative score based on thin information.
Their bank statement tells a different story. Salary of R18,500 lands on the 25th of every month from the same employer. The account maintains an average balance of R3,200 through the month. There are no returned debit orders. There are two debit orders (a funeral plan and a gym membership), both paying consistently. There is no gambling spend. Cash withdrawals are modest. The month-end balance has not gone below R800 in three months.
The bureau says “no data.” The statement says “low-risk.” For a credit provider who relies exclusively on the bureau, this consumer is either unserviceable or requires a risk premium that their actual profile does not justify. For a credit provider using statement-based analysis alongside the bureau, this consumer is a clear approval.
This is the population that credit-invisible consumers occupy. They are not unworthy of credit. They are simply unmeasurable using tools designed for consumers with credit history. The bank statement provides the signal that the bureau cannot.
The over-reliance case: where a good bureau score hides current stress
Now consider the opposite scenario. A consumer with a bureau score of 730: 10 years of clean repayment history, no defaults, no judgements. They look excellent on bureau.
What the bureau does not show: they were retrenched six weeks ago. Since then, they have taken out three payday loans from separate micro-lenders to bridge cash flow. Two new debit orders have appeared in their account from lenders not on their bureau file. Their salary credit has not appeared in the last two months. Their account balance has been below R500 for 15 of the past 30 days.
Six weeks ago, that 730 score reflected reality. Today, it reflects historical data that has not been updated. The bank statement, obtained today, tells a completely different story. A credit provider who approves this consumer based on their bureau score alone is not conducting a substantive affordability assessment. They are relying on stale data in a situation where the consumer’s position has materially deteriorated.
This is precisely the gap that the March 2026 court judgment on Section 81 of the NCA addressed. The court found that going through the motions of an assessment (pulling a bureau score, noting it is above threshold, approving) does not satisfy the substantive assessment obligation. The obligation requires genuine engagement with the consumer’s current financial position.
Using both: a practical framework for credit providers
The strongest origination process uses both signals in the roles they are suited for:
Bureau score: historical reliability filter. Has this consumer demonstrated the ability and willingness to service registered credit obligations over time? A clean bureau history with substantial tenure is evidence of reliability. A poor bureau history with multiple defaults is evidence of pattern risk. Use the bureau score as the first filter: does this consumer have a track record that supports the proposed product?
Bank statement score: current-state affordability and capacity. Does this consumer have the income, buffer, and behavioural profile to service an additional obligation right now? Is the income verified, stable, and sufficient? Are existing obligations (including off-bureau ones) enumerated? Are there behavioural red flags that suggest financial stress the bureau hasn’t yet registered? Use statement analysis to answer these questions and to satisfy the Reg 23A affordability documentation obligation.
The decision framework in practice: a consumer who passes both screens (reliable bureau history and sound current-state financials) is the strongest approval. A consumer who passes on bureau but shows current stress in the statement warrants additional scrutiny before decision. A consumer who fails on bureau but shows strong current-state financials in the statement (the thin-file case) is a candidate for a policy exception or a product designed for new-to-credit consumers. A consumer who fails on both is a decline.
Neither score replaces the other. AffyScore’s bank statement analysis is designed to sit alongside the bureau pull in the credit decision workflow, providing the current-state financial signal and the Reg 23A affordability documentation that the bureau cannot supply. The decision pack delivers both components: the affordability calculation and a behavioural score on the 300–850 scale, expressed alongside sub-factor reason codes that the underwriter can review and the audit file can contain.
What this means for your existing process
Most NCR-registered credit providers already obtain bank statements as part of their Reg 23A income verification process. The data is there. The question is whether it is being used as a risk signal or simply as a verification document that gets filed without systematic analysis.
Manual bank statement review can extract the income and declared obligations needed for a basic affordability calculation. What it typically misses: gambling spend patterns, informal obligation detection, cash withdrawal analysis, balance dynamics across the full three-month period, and income stability metrics. These signals require systematic transaction classification, which is what structured bank statement analysis provides.
The combination of a bureau pull (historical reliability) and structured statement analysis (current-state capacity) is not a new concept in credit risk practice. It is increasingly the standard for lenders who have experienced the limitations of bureau-only origination in a market where 4–6 weeks of data lag can mean the difference between a performing loan and a default.
