Underwriting has always been at the heart of real estate finance. Every investment decision ultimately reflects the quality, speed, and accuracy of the underwriting process behind it.
With the rapid rise of artificial intelligence over the past 18–24 months, a key question has emerged:
Can AI truly speed up underwriting—and if so, how?
In a recent Cambridge Finance webinar, Sherry Yi Xiang Xu (University of Manchester) explored exactly this. The takeaway is clear: AI is already delivering meaningful efficiency gains—but not in the way many expect.
AI in Underwriting: Augmentation, Not Replacement
Despite the headlines, AI is not replacing underwriters.
Instead, it is acting as a powerful assistant.
Today’s AI tools can:
- Read and process large volumes of documents
- Extract and structure key data
- Pre-screen and prioritise submissions
- Draft rationales, exclusions, and follow-up queries
- Flag inconsistencies and potential risks
What they cannot do:
- Make final underwriting decisions
- Handle complex edge cases independently
- Replace human judgement and experience
- Operate without oversight or governance
The reality is simple:
AI handles up to 70% of the repetitive workload—freeing underwriters to focus on the 30% that truly matters.
Understanding the 3 Layers of AI in Underwriting
To use AI effectively, it’s important to understand how these systems are built. Most underwriting tools combine three key layers:
1. Rule-Based Automation
This is the foundation.
- Works on “if–then” logic
- Fully auditable and transparent
- Ideal for standardised decisions (e.g. LTV thresholds, escalation rules)
While not technically AI, it’s often the highest-impact starting point.
2. Machine Learning Models
This layer identifies patterns in historical data.
- Detects risk combinations humans may miss
- Suggests pricing and risk bands
- Improves consistency across decisions
Importantly, these models support decisions—they don’t make them.
3. Generative & Agentic AI
This is the newest layer.
- Reads emails and attachments
- Extracts structured data automatically
- Drafts underwriting reports and communications
- Acts as a workflow assistant
What was experimental two years ago is now production-ready for many tasks.
Where AI Delivers the Most Value: The 4 Stages of Underwriting
Underwriting typically follows four stages. AI enhances each—but not equally.
Stage 1: Intake & Triage (Highest Impact)
This is where AI delivers the biggest immediate gains.
AI can:
- Read emails and attachments
- Extract key data fields
- Check submissions against underwriting criteria
- Draft follow-up emails for missing information
Impact:
Up to 30–50% of underwriters’ time—previously spent on admin—can be significantly reduced.
Human role remains critical in:
- Setting underwriting policies
- Reviewing low-confidence data extractions
- Managing broker relationships
Stage 2: Data Aggregation & Enrichment
AI consolidates data from multiple sources into a single, structured view.
- Pulls credit, property, and risk data
- Identifies inconsistencies across sources
- Maps geolocation risks (flood, wildfire, etc.)
- Benchmarks against historical performance
Impact:
Manual research drops from hours to minutes for standard cases.
Humans still decide:
- Which data sources to trust
- How to resolve conflicting information
Stage 3: Risk Assessment & Scoring
This is where human judgement remains dominant.
AI can:
- Suggest risk ratings
- Recommend pricing adjustments
- Provide confidence levels
But:
The model suggests. The underwriter decides.
This is also where the real value of human expertise comes into play—especially in complex or non-standard deals.
Stage 4: Decision Support & Documentation
Generative AI shines here.
It can:
- Draft underwriting rationales in plain English
- Generate documentation and reports
- Suggest terms, exclusions, and conditions
- Create a detailed audit trail
Impact:
- Faster turnaround times
- Stronger, more consistent documentation
- Improved auditability and compliance
Humans remain responsible for:
- Final wording and approvals
- Negotiation strategy
- Quality control
The Real Business Impact
When implemented effectively, AI in underwriting can deliver:
- 50–70% reduction in processing time (for standard risks)
- 2x increase in underwriter productivity
- Improved consistency across teams and regions
- Stronger audit trails and regulatory compliance
However, results vary widely depending on:
- Data quality
- Implementation approach
- Change management
Governance: The Non-Negotiable Foundation
AI adoption must be paired with strong governance.
Key considerations include:
- Explainability of decisions
- Data quality and bias control
- Clear ownership of models vs. decisions
- Continuous monitoring and retraining
Regulatory frameworks (FCA, PRA, EU AI Act) are rapidly evolving—making this an essential priority from day one.
What’s Next: The Future of AI in Underwriting
Looking ahead, the next 12–18 months will bring:
Agentic AI Assistants
AI that actively supports workflows—not just responds to prompts.
Conversational Systems
AI interacting with brokers and internal teams in real time.
Continuous Underwriting
A major shift:
- Moving from point-in-time decisions
- To ongoing risk monitoring
AI will track:
- Weather patterns
- Asset performance
- ESG data
- Maintenance signals
And trigger action before losses occur.
Key Takeaways
- AI enhances underwriters—it doesn’t replace them
- Start with data-heavy, repetitive stages (intake & aggregation)
- Focus on clean data and clear processes
- Build governance frameworks early
- Think of AI as a long-term capability, not a quick fix
As the industry evolves, those who adopt AI thoughtfully—balancing efficiency with judgement—will gain a clear competitive advantage.
The question is no longer whether to use AI in underwriting.
It’s how quickly and effectively you can integrate it into your workflow.
