Why we re-weight instead of subtracting fraud

The standard workflow

Most MCA underwriting tools, the parsers, the categorizers, the spreadsheet templates, work the same way. They run through the bank statement, label some transactions as suspicious, subtract those from the deposit total, and report the remainder as revenue.

That looks reasonable until you ask: what about the deposits that aren't quite fraud, aren't quite customer revenue, and aren't easy to categorize either way?

The deposits the standard workflow misses

Real bank statements have deposits in the middle of the spectrum. A wire from an entity whose name half-matches the merchant's. A "MDEPOSIT" line item with no description and an unusual amount for the industry. A transfer described as "OTHER INCOME". A "BANKCARD" deposit at a business that doesn't accept cards.

None of these will trip a hard fraud filter. None of them are what an underwriter would confidently call customer revenue either. The standard workflow rounds them up to "not fraud, therefore revenue." The underwriting committee ends up funding against a number that includes them at face value.

When the merchant defaults, those middle-spectrum deposits are usually the ones that didn't sustain. The underwriter is left explaining a number that, in retrospect, included a lot of "I'm not sure" treated as "yes."

What we do instead

Vyaso doesn't binary-classify deposits. Every credit on the statement gets a confidence figure between zero and one. That figure represents how likely the credit is to be real customer revenue. The adjusted-revenue figure is the weighted aggregate of all deposits and all confidences.

Some examples of how this changes the math without changing the deposits:

• A deposit clearly described as a customer payment from a known marketplace platform earns near-full confidence.
• A deposit from a counterparty whose name closely matches the merchant earns near-zero confidence. That is the kiting signature.
• A deposit with an ambiguous description in a typical industry range earns moderate confidence. It is neither counted in full nor stripped to zero.

When the rule-based confidence falls in a middle band where the system is genuinely uncertain, an agentic synthesis step reviews each ambiguous deposit with full context (description, amount, counterparty, industry, the merchant's broader pattern) and resolves it.

Most bank-statement tools tell underwriters: this is fraud, that isn't. Vyaso tells them: here is how confident the engine is in every credit, and here is the aggregate.

Why this matters for the committee meeting

The conservative end of underwriting is where decisions survive collections. Funders that underwrite against a generous deposit total fund deals that default at higher rates than their pricing assumed. The middle-spectrum deposits the standard workflow rounds up are usually the ones that did not sustain.

Funders that underwrite against the adjusted-revenue figure fund deals priced for the actual cash flow. The default rate reflects the real risk, not an inflated risk.

The committee meeting changes too. When the analyst presents a confidence-weighted figure, the discussion shifts from "do we trust this number" to "do we want to fund this risk." Both are useful conversations, but the second one is the one underwriting committees should be having.

What this isn't

This isn't a probabilistic loss model. Vyaso is not estimating default probability or expected loss. The confidence figure is about the present. It describes how much of today's deposit total is real customer revenue. It does not forecast the future.

This isn't a black-box score either. Every deposit's confidence is traceable: which detection layers contributed, which signals raised or lowered the figure, which deposits were resolved by the synthesis step. The dashboard shows every step.

And this isn't an attempt to replace underwriter judgment. The adjusted-revenue figure is one input to the committee conversation. Files where the figure diverges materially from declared revenue, or where the flag list is unusual, are exactly the files where underwriter judgment matters most.

The category claim

Vyaso describes itself as the underwriting intelligence layer. The methodology choice in this post is what makes that claim real. Parsers tell underwriters what each transaction is. Vyaso tells them how confidently each transaction belongs in the revenue figure. The aggregate is a number underwriting committees can defend.

That's the layer.