Book diagnostics.

Why did arrears, revenue or approvals change? Luminnovate splits any movement exactly into who you lent to and how they performed, drills down to where it happened, and says when a change is only noise.

A rise in arrears prompts a tighter scorecard, but often the borrowers did not get worse: the mix changed. Acting on the wrong reading costs good customers.

Book diagnosticsEarly arrears, Q2 against Q1

Early arrears rose from 2.10% to 3.35%

Who we lent to+0.90

How they performed+0.35

Percentage points of the change. The two parts add up exactly.

Where it happened

Loans $20k to $50k	+0.62
Trades and construction	+0.41
Everything else	+0.22

Illustrative example.

What Luminnovate does

Splits every change exactlyInto mix and performance, with the parts adding up to the whole.
Finds where it happenedDown to the segment, region or product that moved.
Knows when not to actChanges within normal variation are reported as such.

Two cases

Results from Luminnovate's work, labelled by source. Unlike the illustration above, these figures are not invented.

88%Real lending data from an open research dataset

Risk or mix?

of the rise in early default sat in one loan size band

What was observed: Early default rose from 4.56% to 7.04% between two years of lending.
What a generic approach says: Borrowers are getting riskier: tighten the scorecard.
What the engine read: 88% of the rise sat in loans of $4,000 to $10,000, which grew from 36.4% to 47.9% of the book. Within them, borrowers scoring 660 to 699 grew from 18.5% to 31.9%.
What it meant for the decision: The rise was mostly a shift in mix. The response belongs in who is lent to.

Read the full case

15%Real retail data from an open research dataset

A fall that was noise

fall in revenue that was within normal variation

What was observed: Revenue fell from £1,749,039 to £1,479,878 between the same months of two years.
What a generic approach says: A 15% fall: investigate and act.
What the engine read: The fall was within normal variation for this business, and returning customers carried 76% of it.
What it meant for the decision: Knowing when not to act is part of the decision.

Read the full case

Where to set the cut off

Impact Sandbox chart of the estimated bad rate of declined applicants by score band against the breakeven line, and the proposal to buy the evidence first by approving a small random share of declines. — From Luminnovate's Impact Sandbox. Bad rates estimated from real lending data from an open research dataset, with a measured correction added; dollar figures from illustrative assumptions, not a client's.

Where it is used: Banks and lenders
One case: Risk or mix? (Real lending data from an open research dataset)
Trust: How explanations are made and tested

Start with last quarter's movement that nobody could explain. We split it on your own data.

Book a conversation