Blog

Blog

Patrick Reily

Patrick Reily is CEO and co-founder of Verde International. He launched Verde International in 2006. He is a 25-year financial services veteran, having served in executive and P&L ownership roles with some of the largest US and multinational financial institutions. His expertise includes business startup and turnaround, credit risk and underwriting, marketing science, mergers and acquisitions (M&A), automated decision systems, financial performance and regional economics. Since the late 1980's, he has led the use of unstructured and transactional data for predictive modeling in applications for underwriting, utilization, retention, optimization, offer management, cross-selling, customer service and fraud prevention. His work has been used to predict macroeconomic expansion and contraction by the Federal Reserve. He holds a MBA and MS in Economics from Wright State University in Dayton, Ohio.

Choosing the Right AI Underwriting System – 4 Steps to Success

Sep 18, 2017
Choosing the Right AI Underwriting System – 4 Steps to Success

As SME lenders, we face the challenge of satisfying simultaneous objectivesshareholder return, borrower empowerment, economic development and financial inclusionall within regulatory compliance. Even the most capable underwriters struggle to do all of these well, especially all at the same time.

The top Fintech/AI underwriting systems consistently outperform manual processes, achieving higher performance at greater scale with higher reliability, inclusion and compliance; but how do you find the right system for you? How do you find a system you can trust?

I’ve been building AI driven decision systems for leading global lenders for almost 30 years. The perfect system is unique for each institution, but the following guide holds true for all.

  1. Think Systematically

People and computers solve problems in different ways, and systems need to be designed accordingly. For example, a great digital loan app designed for human underwriters will typically do a poor job of supporting AI decision systems. At each stage in the processfrom application, data collection, decision and pricing, all the way to disbursementsystem designers need to know when the system will rely on people, AI, or both. The highest performing solutions are cohesive systems, not an assembly of disparate parts.

  1. Different Thinking Needs Different Data

The best data to support human decisions is different than the best data to support AI decisions. People prefer to think about just a few things at a time, so the best decision makers tend to simplify the process while preserving accuracy. AI, however, doesn’t struggle with complexity or detail. AI decisions actually improve as data complexity, granularity and diversity increase. Rather than thinking “how do I make the best decision with only mobile data or a credit report or financials”, the best AI systems adapt and optimize the decision process based on all the information available for each individual decision.

  1. Pick Smart AI, Not Dumb AI

It’s easy to see that rocket scientists and bricklayers think on different levels. The same is true for AIsome are brilliant, some aren’t. Great AI systems are more than smart, though. They’re also: 

  • Transparent about how they make decisions
  • Explicit about the confidence of each decision
  • Humble when a decision is beyond their expertise
  • Comprehensive rather than narrow problem solvers
  1. Expert Systems Only Come From Expert Authors

It’s obvious that software and mathematical modeling expertise are important, but it’s equally critical that AI architects have deep understanding of underwriting, loan management and banking. 

Brilliant, reliable AI, requires category expertise regarding the decision being made. It’s impossible to teach an AI system how to make the most of each lending opportunity if you don’t have experience creating lending opportunities. It’s vital that designers have expertise in the following areas:

  • Lending & Economic Development

    • Risk, Pricing, Profitability and Capital Management
    • Regulation, Compliance and Governance
    • Economic Development, Poverty Abatement, Inclusion and Fairness
  • Modeling, Analytics and AI Systems

    • Behavioral and Financial Modeling
    • Econometrics, Statistics and Data Valuation
    • Artificial Intelligence & Machine Learning
  • Software Development

    • Data Interfaces and Warehousing
    • Agile Development because each situation is unique
    • Statistical & Modeling Systems Integration

When I built my first intelligent lending systems in the 1980s, it was difficult to imagine making an AI system affordable for community size lenders. In 2007, we delivered our first AI lending system to a community lender and they are still using our systems today. Advances in technology make it possible to teach intelligent modeling systems how to build intelligent lending solution that empower global, national and community lenders. Within the next five years, most lending around the world will utilize AI driven decision processes. Some AI systems will be much smarter than others. Choose wisely.


Have questions for Patrick? He is available to answer questions from SME Finance Forum Members via our Members App.

Patrick and Verde International will be speaking on SME Credit Risk Management and Collection Strategies at the Global SME Finance Forum 2017

Credit Risk & Scoring