Global Equipment Finance Scorecard + Credit Bureau Bridge
I put nine business credit bureaus side by side this week and sorted them by what tense they speak in.
Past — 2 of 9. Present — 1 of 9. Future — 6 of 9.
Nobody sells you all three. Most credit files hand you one tense and let you assume it covers the other two.
That matters, because a scorecard has to capture how a customer paid, how they are holding up now, and whether they can keep paying you next year. Those are three different questions and the market answers them with three different products that all arrive looking like “a business credit score.”
Here is the same business, at the same credit quality, read off all nine:
PAYDEX 80 · Cortera CPR 710 · D&B Delinquency Predictor 527 · NACM 515 · Intelliscore 61 · Creditsafe 72 · Global Database 71 · CRIF class 4 · Rapid Ratings FHR 63
Four of those are numbered 1–100. They read 80, 61, 72 and 71. Same company.
And the ranges collide in ways nothing on the report warns you about:
- D&B’s Delinquency Predictor runs 101–670. NACM’s National Trade Credit Report runs 101–662. Different vendors, different models, different horizons, near-identical ranges.
- Experian’s Intelliscore Plus V3 runs 300–850 — the consumer FICO range, on a business.
- Cortera’s CPR reads the trailing three months. PAYDEX reads the trailing twenty-four. Same construct, very different memory.
So how much does the substitution actually cost you?
Get Suneel Reddy’s stories in your inbox
Join Medium for free to get updates from this writer.
I simulated it — 400,000 paired draws, rank correlation to PAYDEX set explicitly per bureau. Translating a score from one bureau onto bins fitted on another lands in the correct bin:
- 47% of the time for the closest substitute (Cortera — same construct, different panel)
- 41% for NACM
- 37% for Creditsafe
- 31% for Rapid Ratings, which reads filed accounts and never sees a trade line
Measured bias sits inside ±0.06 points everywhere. So the cost of a substitute bureau is variance, not drift. It won’t systematically flatter or punish your book. It will just be wrong in both directions, roughly ±10 points on a 465–810 scale, which at a fixed cutoff moves a real slice of applicants across the line.
Note the ranking follows the data source, not the vendor’s size. NACM — a members’ cooperative where creditors contribute their own receivables ageing — beats both large commercial forecasts below it. Trade lines are what payment behaviour is made of.
Three things I’d tell anyone choosing a data partner:
1. Coverage is a revenue decision, not just a risk one. A decent data partner gets you to yes faster on the customers who will actually pay. That is a sales asset, not a compliance line item. The flip side is real too — in roughly half the world’s markets there is one business bureau or none, and there you take the construct you are given.
2. Before you sign, run the partner’s data through your model on customers you have already sold to. You know how those accounts performed. That is a free back-test, and it is the only number that matters. I simulated my correlations under stated assumptions — on your own closed book you can measure them. Any vendor confident in their data will let you do this. Pay attention to the ones who won’t.
3. Design the model to capture past, present and future. Payment history, current financial position, forward capacity. If your file only carries a 12-month failure forecast, you have bought one tense and are underwriting as though you own three.
The full scorecard is live — 14 markets, three ticket segments, nine business bureaus, every attribute with a “not available” option that carries its own weight instead of an imputed guess:
suneelreddy.com/calculators/equipment-finance-scorecard.html
Pick your bureau, enter the score you actually have, and it shows you the translation and what the translation cost.
Built on synthetic data — the 0.63 Gini flatters. The method transfers; the number doesn’t.
One correction worth flagging, since it happened while building this: I had Cortera’s CPR on a 0–100 scale and D&B’s Delinquency Predictor on a percentile. Both wrong. If I can get two scales wrong while writing a page about scales, it is worth checking what your own system thinks it is receiving.
What tense does your business credit file speak in — and did you choose it, or did it come with the contract?
#CreditRisk #EquipmentFinance #TradeCredit #Underwriting #CommercialCredit #RiskModelling #DataStrategy
Originally published at https://www.linkedin.com.
