Nova Credit: cash-flow information, customer access and lending economics

The first job is to make financial information usable

Cash Atlas is presented as a way to turn permissioned account information into cash-flow reports and attributes. NovaScore is a separate proprietary scoring output. A report can help staff interpret an applicant’s finances even when a score does not determine the outcome. The vendor materials support those product distinctions; they do not establish a universal improvement in access or profitability. [1][2][3]

Analysis: the customer may be trying to explain variable earnings, complete a purchase or obtain an appropriate payment amount. The institution needs evidence that arrives in time and can be understood consistently. Reducing document collection can be valuable, but repeated connection attempts can move the workload to the customer rather than remove it.

The commercial question therefore begins before model performance: who can supply usable data, how much work is required, and what decision changes as a result? Customer access, staff effort and product fit belong alongside the predictive and affordability questions below. A successful connection is an information event, not a lending outcome.

What the public documentation establishes

Nova Credit describes Cash Atlas as a product that analyzes bank transactions and supplies cash-flow attributes, reports and scores for underwriting. Its NovaScore Cash Flow page describes a credit-risk score built from bank-account information, accompanied by adverse-action factors. These are distinct outputs. A lender may use transaction-derived attributes in its own policy or evaluate a proprietary score; purchasing the report does not resolve which approach is appropriate.

The reviewed public pages identify intended uses but do not disclose a complete model specification, training population or independently reproducible performance study. Accordingly, this article distinguishes documented interfaces from vendor claims about predictive improvement. It does not assume that Cash Atlas, NovaScore and Nova Credit’s other products are interchangeable, or that an adoption announcement establishes a measured loss benefit.

Adoption evidence: three different credit workflows

These original 2025 announcements provide customer context for the September 29, 2026 review. They establish selections or intended integrations, not a current utilization percentage. None should be read as proof that the customer uses every Nova Credit product or that the announced scope is identical to an institution-wide production rollout.

Scroll horizontally to see all columns.

Customer commentary: right-sized credit is a different claim from more credit

In the September 3, 2025 announcement, Chris Reagan, then identified as President of Chase Branded Cards, described an objective to “approve customers with right-sized lines of credit.” That is a management statement of intended use, not a published measurement of incremental approvals or subsequent losses. [4]

The distinction matters analytically. Cash-flow information might support a different line, term, repayment schedule or request for evidence without changing the approve/decline result. A lender evaluating value should identify which decision actually changed. An improvement in reported income coverage should not automatically be credited as an improvement in creditworthiness or affordability.

What to ask these adoption examples to prove

For a card portfolio, test whether the additional data improves initial line assignment and later performance at comparable utilization and pricing. For a short installment product, examine the cash available around each payment date and whether obligations elsewhere are visible. For a co-branded program, analyze the effect of merchant mix and application channel before transferring results between programs. These are proposed evaluation methods, not claims about the named customers’ internal policies.

Ask for the exact module and version deployed. A Cash Atlas customer can use attributes or reports without relying on NovaScore as its decisive score. A customer reference should identify the application segment, connection-completion rate, fallback for missing accounts and the period over which outcomes matured. No announcement above supplies all of those details.

The practical status is a vendor with named relationships across banking and fintech, while the strength of a specific lending case still depends on implementation evidence. Preserve that distinction when comparing Nova Credit with bureau scores, data aggregators or document-analysis services. The market overview on cash-flow underwriting supplies the broader provider map; this profile focuses on the decision contract and the evidence needed to evaluate this product.

Preserve the meaning of the information

The Cash Atlas V1 API reference exposes details useful to implementation and financial interpretation: source identifiers, income-model versions, recent and annual income fields, confidence measures and report-related status information. Its documentation distinguishes three complete months of recent income from a twelve-month annual measure. Those definitions matter when comparing a report with an applicant’s stated annual earnings or a lender’s existing debt-to-income calculation.

The implementation lesson is to preserve meaning through the decision system. A missing value must not silently become zero. An annualized recent observation is not a full year of actual income, and a confidence measure about a derived field is not the borrower’s probability of repayment. The bank should document units, observation windows and null handling for each attribute it uses, then test that the downstream decision engine interprets the response correctly.

Why cash flow can add information

A repayment history describes how a consumer has managed previously reported credit. Bank transactions can add evidence about current inflows, recurring obligations and the timing of liquidity. The practical opportunity is to distinguish applicants who look similar on conventional variables but have different capacity to absorb a new payment. That is an analytical rationale, not proof that every transaction-derived feature improves a specific portfolio.

A linked account can also present an incomplete picture. The customer may use several banks, split payroll deposits, move funds between owned accounts or pay major obligations in cash. A transfer can look like income if classification is wrong. Those limitations are particularly consequential when the model’s apparent precision encourages a lender to stop asking what the account does not show.

A hypothetical seasonal-income example

Assume an applicant receives $18,000 of net income during the latest three complete months and $48,000 during the latest twelve complete months. Annualizing the recent period produces $72,000; the twelve-month measure remains $48,000. Both calculations can be arithmetically correct while answering different questions. If the recent months reflect a seasonal peak, using $72,000 as durable annual capacity could materially overstate what supports a fixed installment.

Suppose the proposed loan requires $450 monthly and the applicant’s typical off-season cash remaining after existing expenses is $300. A strong recent inflow measure cannot eliminate the $150 monthly gap. The lender could consider a smaller obligation, additional evidence or another decision according to its policy. These are hypothetical figures, not a NovaScore forecast, product recommendation or claim about any customer outcome.

Separate score validation from policy validation

A proprietary score can rank applicants well and still produce poor decisions when combined with inappropriate cutoffs, amounts or terms. Recommended testing first assesses discrimination and calibration against the lender’s relevant outcome. Then evaluate the full policy, including income rules, fraud checks, overrides and the offer the borrower actually receives. The objective is to understand the incremental contribution of the score rather than attribute every policy change to it.

Use a later validation period and report results by account-history length, income variability, product and data completeness. Compare performance at similar approval rates and approvals at similar expected loss. Include connection abandonment and unusable reports in the business case. Restricting analysis to applicants with clean, complete transaction histories can overstate the value available in production.

Reason codes, corrections and operational ownership

Nova Credit’s materials describe adverse-action factors associated with the score. A bank should test whether those factors remain understandable and accurate in its own decision process, particularly when several models and policy rules contribute. A list of reasons why a score is not higher is not automatically a complete account of why a particular application was denied or offered different terms.

Operational controls should identify who handles disputed inputs, corrected reports and reconsideration. Retain the report version, relevant attributes and policy version used at the time. A later refresh may change the transaction history or a derived estimate; it should not erase the evidence needed to explain the original decision. The lender also needs clear access, retention and permitted-use controls for the underlying financial information.

Evaluate the whole connection process

Hypothetical: 10,000 applicants enter an account-connection step, 8,000 finish it and 6,000 produce reports usable for the intended policy. If the service costs $60,000 across that population, the effective cost is $10 per usable report, even though it is $6 per applicant entering the step. These assumptions are not Nova Credit pricing or completion rates.

Suppose each usable report removes 12 minutes of document-review work. That releases 1,200 hours, worth $48,000 at an assumed $40 an hour, before integration and exception costs. The direct capacity benefit alone would not cover the assumed service cost. Additional value might come from better decisions or faster completion, but it must be demonstrated separately rather than added as an unsupported multiplier.

A useful evaluation also follows applicants who cannot connect or supply complete histories. Otherwise, the business case can appear strong only because it excludes the hardest customer journeys. Preserve the distinction between better information, a changed offer and a realized financial benefit.

Costs, evidence gaps and the adoption decision

Cost includes report retrieval, integration, validation, customer support and exception processing. A more complete report can save manual work, but only if it arrives early enough and is usable by the existing process. Repeated account connections or unexplained data gaps can increase customer burden. A pilot should therefore measure funded accounts and total operating effort, not just model lift among completed reports.

Evidence that would strengthen the case includes matured performance on the lender’s own applicants, robust treatment of incomplete accounts and stable explanations across versions. The case would weaken if gains disappeared after controlling for connection completion or if recent income systematically overstated durable capacity. Sources reviewed September 29, 2026 support evaluating Cash Atlas as a structured cash-flow input and NovaScore as a separate proprietary risk assessment. They do not establish a universal approval increase or remove the need for product-specific lending judgment.

Sources

1. Nova Credit: Cash Atlas; undated product page, reviewed September 29, 2026SourceBack to text: ↑
2. Nova Credit: Cash Atlas V1 API Reference; undated documentation, reviewed September 29, 2026SourceBack to text: ↑
3. Nova Credit: NovaScore Cash Flow; undated product page, reviewed September 29, 2026SourceBack to text: ↑
4. Nova Credit: Chase selects Cash Atlas and Credit Passport; September 3, 2025SourceBack to text: ↑1↑2
5. Nova Credit: PayPal selects Cash Atlas; September 4, 2025SourceBack to text: ↑
6. Nova Credit: Imprint integration through Alloy; June 25, 2025SourceBack to text: ↑