From Data to Decisions: How AI Helps EDGE Develop Predictive Attributes

Brian Reshefsky – CEO

AI That Works for Lenders

Artificial intelligence is a powerful tool — but amid the noise, it’s tough to separate substance from spectacle. At EDGE, we focus on what matters to lenders: relevance, accountability, and compliance. That means building solutions with AI that deliver measurable predictive value, support transparent decisioning, and hold up under regulatory scrutiny.

Lenders are rightly cautious about opaque or unexplainable models. Regulations demand transparency and traceability — and so do sound risk practices. That’s why EDGE uses AI not as the product but as one of the enabling technologies behind sharper, more explainable insights into income, indebtedness, and ability to pay.

We don’t build black boxes. We build glass-box tools that are understandable, testable, and grounded in the operational realities of modern lending. One of the most impactful results of this work is the predictive attributes we produce for lenders and consumers nearly every minute of every day.

What Is a Predictive Attribute?

If models are the engines of modern credit and fraud decisioning, attributes are the fuel. Attributes are measurable properties or quantitative features distilled from raw credit, transaction, behavioral, or other data — each designed to illuminate a borrower’s risk, intent, or capacity.

Predictive value refers to how strongly correlated an attribute is with the outcome you're trying to forecast, either individually or within a model. In the case of income and risk attributes, we want to see future income and loan performance that tracks expectations grounded in sound data science. Just as higher octane fuel improves engine performance (up to a point), more predictive attributes improve model outcomes (also up to a point — see below).

For example, one of EDGE’s foundational income attributes translates individual deposits into historical semi-monthly income, normalized for volatility, frequency, and contextual signals. This allows lenders to better assess income even when W-2s or direct payroll access aren't available — and to instantly assess an applicant's ability to repay new debt without the risk of human processing error or the friction of manually gathering documentation.

Creating attributes that work — across products, geographies, and economic cycles — requires more than modeling or ivory tower data science. It demands a repeatable, disciplined process that blends advanced tooling with real-world pragmatism.

How EDGE Builds Attributes That Work

1) Pattern Discovery, Accelerated with AI

EDGE uses a combination of machine learning and generative AI to analyze high-dimensional data sets:

• Open banking transaction histories
• Time-series balance and cashflow behaviors
• Unstructured text and metadata
• Loan outcomes reported to EDGE by the lenders we serve

These machine learning and generative AI tools allow our data scientists to uncover subtle signals — patterns that ultimately correlate with borrower intent, repayment behavior, or early signs of risk. In this phase, AI helps:

• Spot patterns human analysts might miss
• Surface hypotheses worth testing
• Synthesize insights across disparate data types

2) Human-Validated, Regulator-Ready

From there, every EDGE attribute undergoes rigorous review. We don’t ship ideas — only verified insights that pass three core criteria:

• Predictiveness validated against real-world lending outcomes
• Consistency across borrower types, portfolios, and timeframes
• Explainability necessary for model governance and regulatory oversight

If an attribute doesn’t meet the bar, it doesn’t ship. In fact, many don’t — and that’s by design. We reject the myth that more attributes automatically lead to better models and find ourselves "SMH" when another analytics player in the industry announces thousands of newly released attributes.

In reality, overloading a model with marginal signals reduces performance — a well-known effect called the curse of dimensionality. One of EDGE's customers expressed this concept more tangibly: "Everyone is offering attributes and features coming out of their ears. How do I know where to start?"

Each attribute EDGE delivers was selected for its predictive value across relevant portfolios, validated for regulatory compliance, and clearly documented. However, our "built by lenders for lenders" ethos doesn't end with the build: we're continuously working with customers to optimize which, when, and how attributes are utilized in their underwriting and portfolio management.

Our hybrid AI-enabled, human-validated approach keeps us nimble without compromising rigor. It also enables EDGE to refine or evolve attributes quickly in response to changes in borrower behavior or lender strategy.

Select AI in Production: EDGE’s Income Detection Algorithm

While most of our AI lives in R&D, we do have limited applications in production where data and market realities require a technological solution — most notably, EDGE’s income detection algorithm.

When an applicant connects their financial accounts through EDGE, our income detection model analyzes account-level transaction patterns in real-time to identify and characterize recurring income, even when it's not clearly labeled.

For example, regular ATM deposits or transfers from a side hustle might go undetected by providers who rely solely on memo-matching or merchant tagging. These conventional approaches often misclassify income, especially when descriptions are vague or inconsistent — missing critical earning activity from non-traditional sources. EDGE’s algorithm looks for patterns in frequency, timing, and amount to identify underwritable income that would otherwise be missed.

This is AI in the most practical sense of "artificial intelligence" — codifying human experience and hard-won expertise in identifying income streams into repeatable operations with superior predictive value that scales across the next billion transactions.

The output is immediately visible to lenders, in summary form and at the individual transaction level, and can be validated with a calculator or spreadsheet. Our AI isn't hallucinating its way to a number you or the applicant would never come up with, rather the AI replicates the eyeballing and tallying that underwriters have done since the first proof of income. Just order-of-magnitude faster and more accurate than human intelligence.

The result of this select application of AI: greater visibility for lenders into borrowers' individual creditworthiness, less friction for applicants, decision support including CRA-related requests and dispute channels for consumers, more confident underwriting, and ultimately rightsized access to credit.

What's Different in EDGE’s Approach to AI

We don’t promise or aspire to AI that replaces underwriters. We use AI that make lending decisions smarter, faster, and more explainable. Three principles guide everything we deploy:

• Use AI to accelerate discovery, not replace expertise
• Ensure compliance and auditability from day one
• Focus on real-world performance, not theoretical edge

Over the coming months, we are going to share more examples of how EDGE's attribute engine supports lenders — from identity verification to credit expansion to borrower segmentation.

And on September 10th, I’m joining the Credit Union Times AI Town Hall alongside other fintech leaders to discuss how AI can be implemented thoughtfully, responsibly, and effectively in today’s lending environment.

Because at EDGE, we’re not an AI company. We’re a data company — using AI wisely — to make lending fairer, faster, and more effective.