What a good underwriting report refuses to tell you
An analysis tool is usually judged on the numbers it prints. It is worth judging on the ones it will not. Every figure in a report is a claim about a business, and a claim the evidence cannot carry does not become careful by being printed — it becomes confidently wrong. Here are the four places our report stops short, what it shows in place of the missing figure, and what to ask the merchant for when it does.
- Four gates, each with a reason printed in place of the number
- Errors are not symmetric — missing rows understate burden
- Everything the broken evidence does not touch still prints
A number you cannot check is worse than no number
An underwriter reading a burden percentage is not reading a calculation. They are reading a claim: that this business earns roughly this much, that roughly this much of it is already committed, and that the difference is what a new advance would have to live on. The arithmetic is trivial. Everything that makes the claim true or false happens before it — in whether the rows extracted from the statement are the rows the statement actually contains.
When that breaks, printing the number anyway does not produce a cautious answer. It produces a precise one that happens to be wrong, and precision is what gets trusted. Worse, the failure is not even-handed. Rows go missing far more often than they are invented, missing rows are usually debits, and missing debits make a file look lighter than it is. The error runs in the direction that funds deals which should not be funded.
So the report carries gates. Each one asks a single question — does the evidence support this figure? — and when the answer is no, the figure does not appear. What appears is the reason.
Four places the report stops short
Each gate withholds the same set of ratios, so a figure can never be withheld in one place and printed in another.
The statement disagrees with itself
A statement prints its own totals: deposits in, withdrawals out. Those totals are ground truth. If the extracted rows do not sum to them within tolerance, rows are missing or duplicated, and every ratio built on them is quietly biased. Burden, safe ceiling and payer concentration are withheld, and the report says the totals do not reconcile.
There is not enough statement to divide by
A ratio over a handful of days, or over a few thousand dollars of trade revenue, is arithmetic performed on noise: one large deposit moves it by tens of points. Below those floors the ratios are withheld as limited statement data, and the underlying figures — deposits, debits, positions — are left to speak for themselves.
The account is salary-funded
When wages arriving into the account exceed the business revenue in it, the denominator does not describe what services the debt. The ratios are withheld, and the withholding is itself the finding: this is probably a personal account, or the wrong one of several, and the operating account is the file you actually want.
The balance ledger cannot be vouched for
Where a statement omits per-row balances, the daily ledger is rebuilt from the opening figure and the transactions. That reconstruction is published only when it lands where the statement says it should — on the printed closing balance, or on the bank’s own daily snapshots. When it lands somewhere else, the rebuilt rows are dropped and every figure derived from them goes with them: low-balance days, negative days, average and lowest balance.
A drifting ledger is the most convincing kind of wrong
The balance gate deserves its own paragraph, because of how quietly it fails. A rebuilt ledger is a running sum: start at the opening balance, add each day’s movement, carry the result forward. One row missing near the start of the month, or one row counted twice, shifts every later day by exactly that amount and by nothing else.
What comes out the other side looks perfect. The series is smooth. It moves with payroll and with the daily debits. It has no spikes, no gaps, nothing an eye would catch — it simply sits at the wrong level. If the drift is downward, a merchant who runs a comfortable float can be reported as closing almost every day below the low-balance threshold, and the report will say, with a straight face, that this business lives on the edge.
No amount of looking at the series reveals that. The only test that catches it is arithmetic against something the statement itself printed: does the chain land on the closing balance, or on the snapshots the bank published along the way? If it does, the reconstruction has been checked against the bank’s own answer and can be published. If it does not, it is not a slightly imperfect ledger. It is a different month.
A blank that says why
A withheld figure is not a dash. In its place the report prints the reason it is missing — the totals do not reconcile, limited statement data, the account appears salary-funded — because “more data needed” tells an underwriter nothing they can act on. Each reason points at a different next step: ask for the missing pages, ask for more months, ask for the operating account.
And the withholding is narrow on purpose. Only the figures that actually depend on the broken evidence step back. A statement whose totals do not reconcile still shows its positions, their cadence, the funders behind them, the deposits and the transaction rows themselves. The parts of the file that remain observable stay fully readable; what stops is the arithmetic that would have quietly inherited the defect.
One decision, every surface
Each gate is decided once, for the whole analysis, and every surface reads the same decision. The report on screen, the merchant view, the multi-month trend, the PDF and the exported file cannot disagree about whether a number exists — which matters more than it sounds, because the figure that ends up in a credit memo is usually the one from the export.
There is a third possibility between a number and a blank, and it gets used where it belongs. When an obligation is real but structurally invisible — an advance repaid out of card settlement, where the holdback never touches the account — the report publishes a range with its assumption stated rather than a confident figure or an empty cell. A bounded estimate you can argue with is honest. A single number resting on an undeclared assumption is not.
None of this makes a report more impressive to look at. It makes the numbers that do appear mean something, which is the only property an underwriting file really needs.
Common questions
Why does my report show no burden percentage?
Because one of three conditions made the ratio unsafe to compute: the extracted rows do not match the statement’s printed totals, the period or the revenue is too small for a ratio to mean anything, or the account is salary-funded so the denominator is not business revenue. The report names which one.
What does “statement totals don’t reconcile” mean?
The deposits and withdrawals printed in the statement’s own summary do not match the sum of the transactions extracted from it. Usually a page is missing from the upload, or a section was listed twice. Re-uploading the complete statement normally resolves it.
Why are low-balance days missing on some statements?
When a statement carries no running balance column, the daily ledger has to be rebuilt from the opening figure and the transactions. If that rebuilt chain does not land on the closing balance the bank printed, it is discarded rather than published, and the day counts derived from it are withheld with it.
Does a withheld figure mean the analysis failed?
No. Only the figures that depend on the broken evidence are withheld. Positions, payment cadence, funders, deposits and the transaction rows are all still there, and for many decisions they are the part that matters.
Can I turn the gates off?
No, and that is deliberate — a number that can be unlocked by a setting is not evidence. The way to get the figure is to supply what it needs: the missing pages, more months, or the operating account.
Related
See which numbers your statements can support
Upload one file and read the positions, the cadence, and the burden — with the reasoning attached.