Why Credit Scores Miss Repayment Risk That Bank Transaction Data Can See

A credit score summarizes how someone has repaid in the past, while bank transaction data shows how much they can repay now, which is why the strongest lending decisions use both.

In short

Credit scores and bank transaction data answer different questions. A score summarizes repayment history. Transaction data shows repayment capacity as it stands today.

• Bureau scores compress past credit behavior into one number. They are valuable, but they look backward, update periodically and say little about new-to-credit borrowers.
• Bank transaction data shows income timing, balance cushions, obligations and how all three are trending, which are direct drivers of whether the next payment can be made.
• Two applicants with the same score can have very different cash flow, and a thin-file applicant can have a strong one.
• Cashflow Score returns a 1 to 100 score, where higher means lower risk, built from five behavioral areas: Velocity, Liquidity, Stability, Leverage and Resilience.
• Transaction data is not always the better signal. The strongest approach is usually to add it to the bureau score and validate the result on your own outcomes.

What does a credit score actually measure?

A credit score is a summary of the information in a person’s credit file. That file is built from what lenders report to the bureaus, such as payment history on existing accounts, balances against limits, the age of accounts and recent applications. The score turns that history into a single number that ranks borrowers by relative risk.

It does that job well when a file is long and conventional. A score is also fast, consistent and widely understood, which is why it sits at the center of so many lending processes.

It is worth being clear about what a score is not. A score is a ranking of past credit behavior. It is not a measurement of how much a person can afford to repay today.

Where do credit scores fall short?

Scores have well-known limits, and they matter most at the edges of a lender’s book:

• They look backward. A score reflects what has already been reported. It reacts slowly to a job loss, a pay cut or a new obligation.
• They update on a schedule. The file a lender sees may be weeks behind the borrower’s real situation.
• They miss what is not reported. Rent, utilities, subscriptions and many other regular commitments often sit outside the file.
• They are thin for some borrowers. People who are new to credit, or who use little of it, leave few traces to score.
• They do not measure affordability. A score ranks risk. It does not say whether this borrower, at this income, can take on this payment.

For these reasons, a traditional score is not always the most predictive signal of repayment, even though it remains a useful one.

What does bank transaction data show that a score cannot?

With a customer’s permission, an account history shows how money actually moves. It can reveal:

• Income. How much arrives, how often, how regularly and from what kind of source.
• Cushion. Typical and minimum balances, and how often an account runs close to zero.
• Obligations. Rent, loan repayments and other commitments, and when they fall relative to payday.
• Volatility. How much income and spending vary from month to month.
• Direction. Whether deposits are growing or shrinking, and whether cushions are building or eroding.

Together these describe capacity to repay, and they update as fast as the account does. Raw spending categories are only a small part of the picture. What matters is the behavior built from them, which we cover in why spend categories rarely move the needle and what actually does.

How can two applicants with the same score look so different?

Consider two applicants, both with the same bureau score. The example is illustrative, not drawn from any lender’s data.

Applicant A has income that lands on the same days each month, a balance that stays above a comfortable level, and commitments that are modest and paid on time. Applicant B has income that arrives late and unevenly, a balance that drops close to zero before each pay date, and commitments that have been rising.

The score treats them alike. Their capacity to take on a new payment is very different. The reverse also happens. A thin-file applicant with a low or missing score can show steady income and a healthy cushion that a bureau file does not capture.

How does a cash flow score describe behavior?

Alternative scoring like Carrington Labs’ Cashflow Score offers a standardized, explainable score built solely from account and transaction data. Carrington Labs’ Cashflow Score runs from 1 to 100, and a higher score means lower risk. It groups behavior into five areas: Velocity, Liquidity, Stability, Leverage and Resilience. Together they describe how money moves through an account, how much cushion it holds, how steady it is, how much is already committed and how the account holds up under pressure.

Each score comes with score drivers, SHAP contributions and plain-English explanations, so a credit analyst can see what raised or lowered it. It does not require personally identifiable information. It also does not verify income. It assesses income patterns and stability.

When is a credit score still the better tool?

Transaction data is not always the better signal. A bureau score can offer a strong choice when, for example:

• The borrower has a long, rich file and a stable profile.
• The borrower has not connected a bank account or cannot do so.
• Speed and convention matter more than additional precision for a given product.

Even then, a score tells a lender how someone has handled credit in the past. It does not show what they can afford today. Transaction data fills that gap, because it shows current income, balance cushion and obligations. Neither source sees everything the other does.

Using both signals can give a lender a fuller view of each borrower than relying on either one alone. A borrower with a strong credit history and a shrinking cash cushion carries a risk that the score alone would miss. A borrower with a thin file, steady income and healthy balances shows a capacity that a bureau file cannot. How much the two add together depends on the lender’s population and product, so the benefit should be measured, not assumed.

Carrington Labs’ approach to a custom Credit Risk Model is to combine sources. It can use any appropriate combination of bureau, application, transaction and internal performance data, and no single source is required. Each source fills in what the others miss, so a model that draws on more than one can weigh more of the borrower’s picture. The aim is to include the signals that add value on your own population, and to prove it with your own outcomes.

How do lenders use both without disrupting their process?

Most lenders do not replace the bureau score. They add a second view alongside it. A practical sequence looks like this:

1. Keep the existing score, rules and decision engine in place.
2. Start where the bureau score is weakest, such as thin-file and borderline applicants.
3. Score applicants with the cash flow signal in shadow first, without changing any decision.
4. Compare against repayment outcomes, then bring the signal into policy or into a model that takes both inputs.

The Add Cash Flow Underwriting use case describes this approach. Our comparison of cash flow underwriting and traditional credit scores and the article on whether to use a cash flow score and a bureau score together go deeper on the differences. For a short introduction, see a 5-minute guide to cash flow underwriting.

What should lenders measure to know it is working?

The test is incremental value on the lender’s own book, not a general claim. Useful measures include:

• The extra separation between good and bad borrowers when the cash flow signal is added to the bureau score.
• The bad rate in approved bands at a fixed approval rate, or the approval rate at a fixed loss rate.
• Calibration, meaning whether predicted risk matches observed outcomes.
• Stability of the signal over time and across segments.

Results vary by lender, product and population, and early shadow results need time to mature before firm conclusions are drawn.

How can lenders add cash flow signals to a bureau-based process?

For consumer lenders, the Add Cash Flow Underwriting use case and Cashflow Score are built for this step. The score complements your bureau score, rules and existing models rather than replacing them, and the lender keeps control of cutoffs and final decisions.

If you want to see how a cash flow signal would sit alongside your bureau score on your own applicants, book a demo with the Carrington Labs team.