Seeing the Invisible: How Trust Science® Scores the 53% of Canadians Missed by Traditional Bureaus

Trust Science’s Canadian Installment Six°Score™ provides industry-leading accuracy tailored to the non-prime market. Leveraging the Credit Bureau+™ platform, the Canadian Automotive Six°Score™ harnesses the power of machine learning and explainable artificial intelligence for unparalleled predictiveness on more applicants. Clients get vastly improved business performance, double-digit ROI, and more financially inclusive lending, backed by data-driven insights that enable you to swap out bad deals while swapping in good deals.
The Canadian Automotive Six°Score™ was built on hundreds of thousands of records in the Canadian subprime market, creating a highly focused and highly predictive model.
The Canadian Automotive Six­°Score™ offers strong statistical performance across all predictiveness and good/bad separation metrics. Validation completed on an analysis of thousands of booked subprime loans.
Auto Case Study Canada

Kolmogorov–Smirnov (KS) Test Results

KS represents a model’s ability to differentiate good and bad populations. Six°Score™ returned a KS result of 31.43, highlighting its power at its most predictive.

AUC and ROC Results

ROC signifies the diagnostic ability of a model. Six°Score™ returned an AUC and ROC result of 0.7127, highlighting its predictive power across its score range.
Auto Case Study Canada
Auto Case Study Canada

AUC and ROC Results

ROC signifies the diagnostic ability of a model. Six°Score™ returned an AUC and ROC result of 0.7127, highlighting its predictive power across its score range.
Auto Case Study Canada

Gini Coefficient and Lorenz Curve

The Gini Coefficient and Lorenz Curve measure a model’s ability to capture bad loans in lower score tiers. Six°Score had a Gini coefficient of 0.5048, proving its bad capture abilities.
Trust Science prepared a validation study on all clients that ran the Canadian Automotive Six°Score™ in parallel with a conventional bureau score or re-scored historical applications. Across all metrics of performance, the Six°Score™ vastly outperformed the bureau score. The test window ran on over 500,000 subprime loans booked from Q1 2022 to Q1 2023

Statistical Lift

64.5%

KS Lift

Six°Score™ differentiates between good and bad loans with significantly greater accuracy vs. a common bureau score.

36.2%

Gini Lift

Six°Score™ rank orders better, capturing more bad loans in low score bands vs. a common bureau score.

12.7%

ROC Lift

Six°Score™ has overall superior good and bad classification, with less error vs. a common bureau score.

Regression & Distribution

In a dual-score matrix generated from the analysis of thousands of approved loans, scored by both the Six ̊ScoreTM, and a traditional credit bureau. Scores from both sources were divided into quintiles based on the population, allowing for a comparison of performance.

Six°Score™

Auto Case Study Canada

Bureau Score

Auto Case Study Canada
While the bureau score had minimal differentiation power between good and bad loans, Trust Science offered successful risk rank-ordering even in this tier by providing a 360° view of the consumer. Six°Score™ enables you to approve more underserved borrowers in this range with confidence. Six°Score™ vastly outperformed the bureau score with:
0 %
KS Lift
0 %
Gini Lift
0 %
ROC Lift

In a dual-score matrix generated from the analysis of thousands of approved loans, scored by both the Six°Score™,
and a traditional credit bureau. Scores from both sources were divided into quintiles based on the population, allowing for a comparison of performance.

Auto Case Study Canada

Default Rate Matrix

Six°Score™ reveals Invisible Prime™ borrowers missed by conventional bureaus,  and correctly identifies bad borrowers that would typically otherwise be approved by the bureaus.

Loans scored in the bottom 20% of the Bureau, yet scored in the upper 40% of Six°Score™ exhibited a lower rate of default. Conversely, those scored in the upper 20% of the Bureau Score but in the lower 20% of Six°Score™ exhibited a substantially higher rate of default. Additionally, there is a benefit in applying a dual-score approach for further lift.

Distribution Matrix

Six°Score™ is able to identify significant swap sets, providing material opportunities to boost return on assets and improve overall portfolio performance.

Auto Case Study Canada

Return on Investment: Client A

Trust Science was tasked with increasing one client’s profitability while retaining the same approval rate. By providing a more comprehensive view of the consumer’s risk (on the same sample evaluated above), Trust Science was successfully able to identify and reduce more defaults than the traditional bureau.

Auto Case Study Canada

After approving the top 40% of loans scored by both Six°Score™ and the traditional bureau, Trust Science generated a 5.1% lift on the client’s earnings while also reducing the bad rate of their loans by 27.8% and, when keeping the bad rate of the sample at 4.0%, Trust Science was able to approve 79.52% more loans than the traditional bureau.

Return on Investment: Client B

Trust Science was tasked with reducing another client’s defaults. While running the Six°Score™ in parallel with the conventional bureau score, Trust Science found that two tiers of loans were being approved that were unprofitable.

Six°Score™ Successfully Identified Unprofitable Tiers of Loans to Swap Out

By rejecting the bottom 25% of loans, as adjudicated by the Six°Score™, Trust Science could eliminate 44.6% of bad loans. As a result of rejecting these bad tiers, Trust Science increased earnings by 31.9% and improved return on capital by 75.9%.

Auto Case Study Canada

How Can Trust Science Help Your

Auto Lending?