Figures below are from a representative worked case, not a client Discuss a pilot
The worked case

What a categoriser sees, and what the business is doing.

One SME file, read two ways. The transactions are identical. The conclusion a lender would reach from them is not.

Run this on your own book
The gap

Inflow is not revenue.

A merchant categoriser counts what arrived. It has no view on where the money came from or what it obliges the business to repay, so loan proceeds, owner injections and internal transfers all read as trading income.

SIX MONTHS OF INFLOWS, COLOURED BY WHAT EACH ONE DOES Jan Jun What a categoriser totals $778,800 What customers actually paid $528,800 Not revenue $250,000 Every transaction was present, correctly dated and correctly valued. What had never been established was what each one does.
Customer payments Loan proceeds Owner injection Internal transfers Refunds and tax

Net surplus reads $516,820 naively and $174,200 once economic role is applied. A serviceability assessment built on the first figure overstates capacity by roughly three times.

Nothing in the data was missing and nothing was wrong. Every transaction was present, correctly dated and correctly valued. What had never been established was what each flow does.

Operating model calibration

The same number means different things in different businesses.

A threshold set across a whole portfolio has to be wrong somewhere. Read against how a business actually operates, three of the most common red flags resolve without a phone call.

A generic threshold saysRead against the operating model

34 negative balance days in 180

A generic threshold says

Scores 25 out of 100

Read against the operating model

Normal for a contractor waiting on progress claims. Alarming for a cafe settling daily.

Receipts $1.41m against declared revenue $2.4m

A generic threshold says

Flag a discrepancy, chase the borrower

Read against the operating model

5% retention is withheld on every progress claim. Explained, no action.

Trading collapses across December

A generic threshold says

Sharp deterioration

Read against the operating model

Industry wide construction shutdown. Expected.

How it would be adopted

The downside is bounded at zero.

Your scorecard keeps running. Ours runs alongside, and the more conservative view wins.

Luminnovate agreesLuminnovate declines
You approved
Funded, as before

Nothing changes.

The only decision that moves

You already wrote this loan, so the outcome is recorded. It defaulted, or it repaid.

You declined
Still declined, either way

The more conservative view wins, so the overlay can never approve someone you rejected. These applicants were never funded, so no outcome was ever observed. This is the row we make no claims about.

Downside bounded at zero

One cell can change, and only in the conservative direction.

Measured in dollars

Losses avoided against margin forgone, on loans already written. No simulation, no waiting.

Start with a real decision

Bring a file you already funded.

We will read it the way this case was read, and you can compare the result with what your own process concluded at the time.

Start a pilot