# Financial spreading software: can you defend the numbers?

By Reza Esfahanian, Founder & CTO of FINKI

Financial spreading turns a borrower's financial statements into a consistent analytical format. Software can extract figures, map accounts and calculate ratios. The useful output is a set of numbers a credit officer can trace to the right company, reporting period and document.

A polished spread can still be wrong. If the software reads a group cash balance as money available to a project company, accurate extraction has produced misleading credit information.

## Separate extraction from credit judgement

Automated spreading is an established software category. nCino describes extraction, mapping and analysis, including review of unclear fields. That establishes what its product offers, not a performance benchmark for your documents. [nCino: Automated Spreading](https://www.ncino.com/solutions/automated-spreading)

Our recommended evaluation separates four operations: reading the document, assigning a financial category, calculating the result and approving its use. Record who reviewed each material adjustment. A model's confidence in reading a number should never count as approval of the accounting treatment.

For European bank buyers, the EBA's loan-origination guidelines provide relevant context for creditworthiness assessment and governance. They do not certify a particular spreading tool. [EBA: Guidelines on loan origination and monitoring](https://eba.europa.eu/activities/single-rulebook/regulatory-activities/credit-risk/guidelines-loan-origination-and-monitoring)

## Ask for the entity and period before the ratio

Give the vendor a small test pack containing a consolidated annual report, a project company's accounts and a later management statement. Include one document in EUR thousands and another in whole euros.

Require the output to retain these distinctions:

| Field | What the reviewer needs to see |
|---|---|
| Reporting entity | Legal name and group or standalone scope |
| Period | Start, end and whether the period is a full year |
| Units | Currency and multiplier used in the source |
| Document basis | Audited, management-prepared or forecast |
| Mapping | Original account label and selected category |
| Adjustment | Amount, rationale, evidence and reviewer |

Treat a missing field as a question to resolve. A software default should not decide which company owes the debt or whether six months of revenue represents a year.

## A small test that catches a large error

Consider this fictional test, with every figure in EUR. A statement labelled "EUR thousands" reports cash of 1,250. The correctly scaled amount is EUR 1,250,000. A note identifies EUR 300,000 as restricted cash. Under the test's assumed definition, unrestricted cash is EUR 950,000.

The output should retain all three amounts: the extracted 1,250, the scaled EUR 1,250,000 and the calculated EUR 950,000. It should link the restriction to the note rather than replace the reported cash figure.

Now suppose the cash belongs to the parent company and the borrower is a separate SPV. The test does not establish that any of it is available to the SPV. That requires evidence of the relevant arrangements. This is where a useful credit workflow stops and asks for an answer.

The arithmetic is simple. The value comes from preserving what the arithmetic means.

## Keep proposed adjustments visible

In another fictional test, reported EBITDA is EUR 2,000,000 and management proposes a EUR 400,000 add-back. Display EUR 2,400,000 as management-adjusted EBITDA, with the proposed adjustment beside it. Do not silently relabel it as reported EBITDA or the lender's accepted covenant measure.

Ask the supplier to reject the add-back. Every dependent ratio should then use the appropriate approved definition, while the rejected proposal remains available for review. Compare this behaviour with the way the system handles an amended statement: yesterday's reviewed spread should remain reproducible after today's document arrives.

## Measure correction time, not extraction theatre

Run a pilot on documents your team already understands. Count material errors separately from harmless formatting differences. Measure the time needed to reach a reviewed spread, including corrections, missing evidence and export to the lender's existing system.

Include scans, restated accounts and contradictory documents. Check whether an analyst can fix one mapping without silently changing previously reviewed borrowers. A learning feature needs approval and regression testing before a correction becomes a rule for other deals.

For project finance, also test the link to forward cash flow. A historical spread cannot, on its own, establish construction completion, future receipts or repayment capacity. A borrowing base calculation should also retain its own contractual definitions rather than inherit the spread's categories without review.

## Questions to settle before purchase

### Does financial spreading replace underwriting?

No. It prepares financial information for analysis. The lender still evaluates repayment, structure and risk, and retains its approval process.

### What should happen when the source is unreadable?

The system should flag the field and request a clearer source or a documented manual correction. Zero is a financial value, not a substitute for failed extraction.

### Should every analyst correction train the model?

No. Some corrections reflect one lender's policy or one borrower's circumstances. Preserve their scope and obtain approval before reusing them elsewhere.

## Bring a difficult statement to the evaluation

FINKI's design links financial evidence to the decisions and operating controls that depend on it. To evaluate that approach, [request a FINKI demonstration](https://finki.ai/request-demo) focused on your spreading and review process. Start with a redacted example agreed through a secure channel; do not attach confidential borrower records to the public enquiry form.

Canonical: https://finki.ai/insights/financial-spreading-software

Published: 2026-09-28