In our last article, we explored the dangers of using GenAI as a decision-maker in FinTech In our last article, we explored the dangers of using GenAI as a decision-maker in FinTech workflows. But here’s the good news: GenAI can still be incredibly valuable in automating processes including Fintech and home lending – if we understand what it’s good at and build around what it’s not.
The key is to stop thinking of GenAI as the engine and start thinking of it as the interface. It shouldn’t make final decisions, but it can help humans and intelligent systems make better and much faster decisions while minimizing human interaction.
Where GenAI Shines
GenAI excels at:
- Language Translation and Summarization: Perfect for digesting complex guidelines into plain English or other languages.
- Drafting and Communication: Drafting borrower explanation letters, loan conditions, or follow-up emails.
- Contextual Pattern Matching: Identifying likely document types or summarizing borrower narratives based on past examples.
These are all high-leverage tasks that don’t require perfect reasoning, just fluency, consistency, and speed.
Designing for Safety: The “Behind Glass” Model
We advocate a “GenAI Behind Glass” architecture – a design pattern that allows GenAI to assist without making unsupervised decisions. This supervision can come in the form of either a human-in-the-loop (HITL) or satisfying a set of logical “proof of correctness” conditions.
Here’s what that looks like:
This model ensures GenAI contributes value, not liability.
Eventually, as the art and science of AI evolve, we will develop progressively more autonomous AI-based solutions. As these solutions grow more complex, the key isn’t that a human must always be “in-the-loop” – but rather, that some degree of supervision is required. That oversight can take various forms, including logical checks, proof-of-correctness methods, etc.
5 Best Practices for FinTech Teams Deploying GenAI
To build with confidence, follow these principles:
- Use for Explanation, Not Unsupervised Judgment Let GenAI help users understand concepts, but don’t let it determine unsupervised outcomes or eligibility.
- Embed Human-in-the-Loop or Supervised Expert Logic Review Every GenAI-generated output tied to compliance, disclosure, or underwriting should be supervised, either by a qualified human or with expert system logic.
- Offload Core Calculations to Deterministic Systems Don’t ask GenAI to compute DTI or LTV. Use trusted loan origination software and validated rules engines for that.
- Incorporate Retrieval-Augmented Generation (RAG) Connect GenAI to verified source documents so it can ground its answers in real policy – not hallucinated patterns. Verify the correctness of any RAG-based outputs.
- Monitor and Audit Regularly Track GenAI’s outputs in production. Use regression testing or spot-check for hallucinations, bias, and noncompliance – especially in customer-facing use cases.
What “Good” Looks Like
An ideal GenAI-enhanced lending platform will:
- Use GenAI to pre-fill explanations, draft messages, and surface guidance
- Validate calculations through core systems
- Route compliance decisions through legal or QC
- Keep a clear audit trail of every GenAI-assisted interaction
- Continually test GenAI responses against source truth
This isn’t theoretical. FinTech leaders are already applying this approach – strongly boosting productivity while maintaining trust, transparency, and compliance.
TRUE and Provable Correctness
GenAI can be a powerful automation tool but requires a well-honed design pattern to be effective and minimize risk. At TRUE, we’ve perfected the art and science of utilizing both classic AI and GenAI technology, all while minimizing any risk of false data, lack of reliable confidence metrics, hallucinations, etc.
At the heart of what we’re uniquely pioneering at TRUE is automating home lending using the concept of “provable correctness” – that is, knowing with certainty that data, documents, and related logic are all correct before taking action.
Conclusion: Smarter and Faster, Without the Risk
The future of mortgage automation isn’t about handing the wheel to GenAI – it’s about using it like a high-performance co-pilot. When placed behind glass, with the right system controls, GenAI becomes a force multiplier that accelerates operations, enriches customer experience, and empowers your team.
Join the Conversation!
Subscribe to our biweekly newsletter for a deep dive into where AI technology is going for mortgage lenders, specific use cases, and discuss a “smarter, not harder” approach to innovation.