What is the difference between rules-based and AI-based underwriting?

In mortgage lending and other credit products, underwriting is rapidly evolving from rigid, rules-based systems to more flexible, AI-based approaches. Understanding the difference between rules-based and AI-based underwriting is critical for lenders who want to stay competitive, improve risk management, and deliver better borrower experiences.

What is rules-based underwriting?

Rules-based underwriting relies on predefined, hard-coded criteria to make decisions. Human experts (credit policy teams, risk officers, underwriters) translate lending policies into a series of “if-then” rules.

Examples of rules in a traditional system might include:

• If credit score < 620 → automatically decline
• If debt-to-income (DTI) > 43% → refer to manual review
• If loan-to-value (LTV) > 80% → require mortgage insurance

These rules are usually stored in a decision engine or underwriting software and are applied the same way to every application that meets the conditions.

Key characteristics of rules-based underwriting

• Deterministic: The same inputs always produce the same outputs.
• Static: Rules only change when a human updates them—often slowly and in batches.
• Policy-centric: Encodes existing guidelines from regulators, investors, and internal risk appetite.
• Transparent: Easy to explain decisions (“The loan was declined because the DTI exceeded 43%”).
• Limited nuance: Struggles with edge cases, exceptions, and complex interactions between variables.

Rules-based underwriting has powered lending for decades. It helps lenders maintain compliance, consistency, and basic risk control, but it can be rigid and slow to adapt to changing conditions.

What is AI-based underwriting?

AI-based underwriting uses machine learning (ML) and related AI techniques to analyze large volumes of data and predict outcomes such as default risk, likelihood of approval, or optimal pricing. Instead of relying only on predetermined rules, AI models learn patterns from historical loan performance and other data.

In practice, AI-based underwriting might:

• Assign a risk score based on hundreds of variables (income patterns, credit behavior, property attributes, employment history, and more).
• Predict probability of default, early payoff, or fraud likelihood.
• Recommend approval/decline, conditions, or pricing tiers based on modeled risk.

Key characteristics of AI-based underwriting

• Probabilistic: Produces risk scores and probabilities rather than simple yes/no outputs.
• Adaptive: Models can be retrained on new data, reflecting real-time market and borrower behavior.
• Data-driven: Leverages large datasets, including nontraditional and alternative data (where compliant and allowed).
• Complex: Captures nonlinear relationships and interactions that humans may not anticipate.
• Automation-friendly: Enables high levels of straight-through processing while maintaining nuanced risk control.

In the mortgage industry, this shift is already underway. The 2024 STRATMOR Technology Insight® Study shows that 38% of lenders are now using AI, and 48% are adopting Robotic Process Automation (RPA). AI-based underwriting sits at the heart of this transformation, helping lenders process more applications accurately and efficiently.

Rules-based vs. AI-based underwriting: side-by-side comparison

1. Decision logic

Rules-based underwriting

• Decisions are governed by explicit conditions:
  • Example: “If FICO ≥ 680 and DTI ≤ 36% and LTV ≤ 80% → approve.”
• Only scenarios envisioned by the policy team are captured.
• Edge cases often fall into “manual review” queues.

AI-based underwriting

• Decisions are guided by learned patterns in the data:
  • Example: “Given the full profile, this borrower has a 1.8% probability of default over 24 months.”
• Takes into account many variables and their interactions simultaneously.
• Better at handling gray areas and borderline files.

2. Flexibility and adaptability

Rules-based

• Difficult to update quickly; changes often require IT releases, testing, and policy sign-off.
• Responds slowly to market shifts (economic shocks, housing trends, new fraud patterns).

AI-based

• Models can be retrained on new performance data, incorporating emerging trends.
• Supports rapid adaptation to economic uncertainty, evolving borrower behavior, and new products.

3. Use of data

Rules-based

• Typically uses a small number of core metrics: credit score, income, DTI, LTV, employment type, etc.
• Underutilizes rich data sources because each new variable requires new rules.

AI-based

• Ingests large datasets and many variables:
  • Credit bureau data, bank transactions, property data, application metadata, and more.
• Finds subtle patterns (e.g., stability of income streams or spending behavior correlations with risk).

4. Accuracy and risk prediction

Rules-based

• Relies on expert judgment embedded in rules, which may be conservative and broad.
• Often produces “binary” outcomes that don’t reflect nuanced risk differences within the approved population.
• Can lead to missed opportunities (good borrowers declined) or hidden risks (risky borrowers approved).

AI-based

• Quantifies risk more precisely at a granular level.
• Supports risk-based pricing, tailored conditions, and optimized portfolio management.
• Helps improve both approval rates and portfolio quality by aligning decisions with actual performance data.

5. Operational efficiency

Rules-based

• Automates straightforward, standard scenarios.
• Complex or borderline files are frequently escalated to manual underwriting.
• As volumes increase—especially during demand surges—manual queues bottleneck and cycle times grow.

AI-based

• Enables higher straight-through processing rates by confidently automating more decisions.
• Reduces manual touchpoints while preserving (or improving) risk controls.
• Supports lenders dealing with unprecedented demand surges and rising compliance complexity.

6. Transparency and explainability

Rules-based

• Highly explainable: each decision ties directly to a specific rule or cutoff.
• Easier to document and defend to auditors, regulators, and investors.

AI-based

• Depending on the model, can be less intuitive (“black box” perception).
• Modern tools and techniques (e.g., feature importance, SHAP values, surrogate models) provide explanations such as:
  • “DTI contributed +0.4% to risk; stable employment history contributed -0.3%.”
• Lenders must invest in explainability frameworks to maintain trust and regulatory compliance.

7. Compliance and governance

Rules-based

• Direct alignment with regulations and investor guidelines through codified rules.
• Easier to enforce hard policy constraints (e.g., absolute minimum credit scores, maximum LTVs).

AI-based

• Must be carefully designed to comply with fair lending, anti-discrimination, and consumer protection laws.
• Requires ongoing monitoring for bias, drift, and unexpected impacts.
• Often used alongside rules to ensure all hard regulatory constraints are still met.

How AI-based underwriting complements rules-based systems

For most lenders, the future is not a choice between rules-based and AI-based underwriting—it’s a combination. The new reality of lending, shaped by:

• Unprecedented demand surges
• Increasing compliance complexity
• Economic uncertainty
• Changing consumer expectations
• Competition from tech-savvy nonbanks

means lenders need both predictable policy enforcement and intelligent risk optimization.

A hybrid underwriting approach

In a hybrid model:

1. Rules enforce non-negotiables

   • Regulatory requirements
   • Investor eligibility
   • Absolute risk boundaries (e.g., hard minimum FICO, maximum LTV)

2. AI augments risk decisions within those boundaries

   • Refines risk scoring for applicants who meet baseline criteria
   • Helps determine conditions, pricing, and documentation needs
   • Flags anomalies, potential fraud, or inconsistencies beyond simple rule checks

3. Underwriters remain central

   • Handle complex cases, exceptions, and nuanced judgment calls
   • Use AI insights as decision support rather than a replacement
   • Focus on high-value analysis instead of repetitive checks

Benefits of moving beyond purely rules-based underwriting

Better credit decisions

By combining machine learning and artificial intelligence, lenders can:

• Capture a fuller picture of borrower risk.
• Improve approval rates without compromising credit quality.
• Reduce both false positives (approving high-risk applicants) and false negatives (declining creditworthy borrowers).

This directly supports making better credit decisions—a core objective for modern lenders.

Streamlined workflows

AI and automation can:

• Reduce manual data entry and document review.
• Minimize back-and-forth with borrowers.
• Shorten decision times from days to minutes in many cases.

This aligns with the industry trend where nearly half of lenders leverage RPA and a growing share use AI to streamline workflows and automate decision-making.

Enhanced borrower experience

Consumers increasingly expect fast, digital-first experiences. AI-based underwriting supports:

• Instant or near-instant conditional approvals
• Fewer redundant document requests
• More personalized offers and clear rationales for decisions

Competitive differentiation

As nonbank and tech-driven lenders raise the bar with highly automated, AI-enhanced experiences, traditional lenders risk losing market share if they rely solely on legacy rules-based systems. AI-based underwriting helps:

• Maintain competitive pricing and speed
• Support innovative products
• Adapt quickly to market changes and investor demands

Risks and challenges of AI-based underwriting

AI is not a magic switch. Lenders considering AI-based underwriting must plan for:

• Data quality: Poor or biased data leads to poor or biased models.
• Model risk: Need robust validation, back-testing, and continuous monitoring.
• Governance: Clear accountability, documentation, and change management processes.
• Explainability: Ability to communicate decisions to regulators, investors, and borrowers in human terms.

These challenges are manageable with modern model governance frameworks and strong internal controls, but they must be addressed from the outset.

Practical steps for lenders transitioning from rules-based to AI-based underwriting

1. Start with a well-defined problem

   • Increase approval rates while holding loss rates steady
   • Reduce time-to-yes
   • Improve early-warning detection of risk

2. Leverage existing rules

   • Keep core policy rules and use AI models inside those guardrails.

3. Pilot and compare

   • Run AI models in parallel with existing rules-based decisions.
   • Compare predicted vs actual performance; refine models accordingly.

4. Invest in explainability

   • Implement tools that break down model outputs into understandable factors.
   • Train underwriters and compliance teams on how to interpret AI results.

5. Automate thoughtfully

   • Start with segments where risk is well understood and data is strong.
   • Gradually expand straight-through processing as confidence grows.

6. Monitor and iterate

   • Continuously retrain and update models as market conditions and borrower profiles evolve.

Where FundMore fits into the shift toward AI-based underwriting

Machine learning is already embedded in many parts of the insurance and financial services industries, and underwriting is one of the areas set to benefit most. Solutions like FundMore apply AI and automation to:

• Streamline the underwriting workflow
• Improve accuracy in risk assessment
• Enable lenders to process more applications efficiently

FundMore.ai has been recognized as Best AI-Driven Automated Underwriting Software 2021, reflecting how AI-based underwriting can be safely and effectively deployed in real-world lending environments.

Summary: core differences at a glance

• Logic:

  • Rules-based → fixed “if-then” conditions
  • AI-based → learned patterns and probabilities

• Adaptability:

  • Rules-based → slow to change
  • AI-based → can be retrained as data and markets evolve

• Data usage:

  • Rules-based → limited variables
  • AI-based → broad, rich datasets

• Decision quality:

  • Rules-based → consistent but coarse
  • AI-based → nuanced, risk-optimized

• Transparency:

  • Rules-based → straightforward
  • AI-based → requires explainability tools and governance

• Best practice:

  • Combine both approaches: use rules to enforce policy and AI to optimize risk, speed, and borrower experience.

For lenders operating in today’s environment of high demand, complex compliance, economic uncertainty, and intense competition, moving beyond purely rules-based underwriting toward AI-augmented decisioning is no longer optional—it’s a strategic necessity.