Case Studies

Robo-Underwriter

June 17, 2020

Results

  • 28% increase in portfolio profitability
  • 19% decrease in write-offs
  • 12% increase in quarterly volume growth
  • 74% increase in online STP
  • 32% reduction in customer acquisition costs
  • 9% improvement in customer retention rate
  • 31% improvement in online conversion rate

Client

A major, North American, consumer finance company with a substantial store network focused on the subprime and near-prime market.

The client’s products include

Sub-prime lending products ranged from APR 42%-116%, with a term up to 5 years. Loan amounts ranged from $500-$5,000. Near-prime APR ranged from 26% – 38%, with a 3-5-year term, with loan amounts from $5,000 to $15,000.

The client had accumulated a substantial amount of predictive data. However, risk management, underwriting, customer lifecycle management had been built around the offline business. So operations were predominantly manual.

Operational controls were state of the art, with mature KYC validation and verification systems and procedures. Customer satisfaction was high due to a strong relationship management approach and culture.

Due to changes in the regulatory environment and other market pressures, the client, wished to,

The client wanted to rapidly implement intelligent automation and, at the same time, accelerate the transformation of its consumer financial products from offline to digital.

Zoral software used

Zoral decision engineZoral DE
Zoral model libraryZoral ML
Zoral loan managementZoral LM
Zoral analytical data workbenchZoral ADW
Zoral behavioral data warehouseZoral BDW

Zoral MVP Implementation Approach

The implementation approach was based on Zoral‘s Minimum Viable Product (MVP) implementation methodology, followed by phased productionalization, system training and monitoring.

Zoral analyzed the offline and online loan portfolio. Zoral ML standard models were used to baseline the current business KPI’s and estimate the potential uplift possible by applying the Zoral robo-underwriter.

Case study. Offline loans performance Case study. Vintage analysis

Zoral ML default models, based purely on client application data for new and returning offline customers, were applied to the portfolio. This highlighted underperforming segments in the portfolio and inefficiencies in the existing underwriting process.

Case study. XML default model for new offline clients Case study. ZML default model for returning offline clients

Note: each segment represents 5% of the portfolio population.

Case study. ZML default model for online clientsInitially, the offline portfolio was significantly larger than online. Zoral ML online default model was applied to the online portfolio, exposing underperforming segments.

After the baseline KPIs were calculated, portfolio data was enhanced. Credit bureau and bank statement data was added to the application data. This allowed the Zoral ML model capabilities to be used more fully and achieved a significant uplift in predictive quality.

Case study. ZML default models with additional data, offline clients Case study. ZML default models with additional data, online clients

Additional fraud was discovered in both the offline and online portfolios. Zoral ML fraud models were deployed, reducing fraud. This was important as it not only reduced fraud costs, but also provided a reliable digital platform, suitable for high volume growth.

Case study. XML fraud scorecard for online clients Case study. XML fraud scorecard for offline clients

Zoral ML fraud models uncovered further fraudulent CSR activity at the stores. To address this, automated underwriting processes were introduced, substantially removing CSR’s from loan approval decisioning.

Case study. Selected store locations with highest fraud rate

Affordability controls and models were also introduced using Zoral ML and Zoral DE. This was necessary in order to improve the quality of underwriting decisions, reduce and control frauds, and comply with the regulatory framework. This was applied to both the online and offline business.

Zoral case study. Affordability estimation

Zoral ML limit management models were also deployed for all offline and online financial products. These calculated the maximum, affordable monthly limit for each lending transaction.

Zoral DE and Zoral ML robo-underwriter

Zoral ML default, and fraud models were used as inputs to robo-underwriter to rank and segment clients based on profitability. Overall portfolio profitability was an important goal.

As well as optimizing profitability, the client wanted to keep acceptance levels as high as possible. To achieve an optimum balance, a range of Zoral ML robo-underwriter inputs were used. These included Zoral ML default, fraud, affordability control, and limit management models for offline and online product lines. Uplifts generated were significant, as can be seen from the graphs below.

Case study. Revenue and average write-off after robo-underwriter optimization

Productionalization

Zoral platform components

Phase I – optimize offline

Phase II – scale online

Phase III – tune and optimize results

During Phase III online revenue for certain products began to exceed the offline revenue.

This, in turn, allowed the client evolve other aspects of its business more rapidly, e.g. product innovation and differentiation. Operating costs were now decoupled from portfolio size growth. So the client was able to expand market share whilst containing costs and growing profitability.

The client continues to innovate, periodically introducing additional Zoral ML models and increasing automation using Zoral DE A/I intelligent workflows.

Results

The results achieved are summarized below:

Case study. New clients, dynamic of default rate Case study. Returning clients, dynamic of default rate
Zoral case study. Customer application journey, conversion rate dynamics