Session Summary: AI and the Future of Work

From the MBA Annual Conference held in Las Vegas, NV, in October 2025

Speaker: Geoff Kramer, Financial Services Engineering Leader, Google (Head of Customer Engineering, Enterprise Financial Services North)

Geoff Kramer’s session provided an in-depth look at how AI—particularly generative AI—is reshaping workflows across financial services, with a focus on the mortgage industry.

Since joining Google in 2020, Kramer noted that 90% of his initial work centered on document processing for lenders and SaaS providers. What began as routine automation for FHA, Fannie Mae, and Freddie Mac documentation has evolved rapidly: in just one year, the mortgage industry has shifted from skepticism about chatbots to deploying cutting-edge GenAI solutions.

Key Trends and Data Points

• Acceleration of GenAI Adoption:
  • 74% of companies adopting GenAI report ROI within the first year.
  • 53% see measurable revenue gains.
  • 51% move use cases into production within 3–6 months.
• Mortgage Industry Context:
  • Average time to close a conventional loan: 42 days.
  • Average cost to originate: $11,000+.
  • Strong focus on using technology to streamline and reduce costs.

Major Use Cases in Mortgage Automation

1. Underwriting Productivity

Kramer traced the evolution from OCR to pre-trained parsers, noting that these approaches reached diminishing returns. Even with 99% accuracy per data point, the overall page accuracy fell to ~90%, making the “long tail” of document diversity a persistent challenge.

Today, teams are leveraging LLMs and Gemini-based custom processors to improve accuracy and adaptability. To mitigate hallucinations, Google applies machine learning layers to verify extracted data against trusted sources—creating an auditable, transparent process.

ROI Impact: Document extraction remains the single largest and fastest-returning GenAI use case, delivering 3–6x productivity gains.

2. Regulatory Compliance

Generative AI is being deployed for automated policy comparisons—analyzing old vs. new regulations, summarizing amendments, and assessing loan-level impacts. Loan officers also use these tools to compare loan products and identify eligibility shifts in real time.

3. Loan Review and Approval

Kramer highlighted how GenAI streamlines pre-approval by identifying missing documents, dramatically reducing underwriter cycle time and improving speed-to-close. At the end of the cycle, it helps teams adjust quickly to regulatory changes that occur mid-origination.
He personally uses NotebookLM to generate audio overviews and summaries from uploaded materials—an example of GenAI’s assistive potential.

4. Lead Generation & Servicing Support

Conversational AI is now in production across multiple financial institutions. Kramer outlined four key use cases:

• Virtual Agent: AI handles customer conversations under human supervision.
• Call Assistant: Real-time, dual-channel listening that surfaces customer data, tracks dialogue, and assesses sales performance.
• Trusted Advisor: A safe, rule-bound chatbot for basic inquiries (e.g., rates, PII-safe questions).
• Digital Twin: An AI replica capable of full audio-to-audio conversations within defined boundaries.

Kramer demonstrated a Gemini Live 2.5 Flash Preview (Native Audio, Sept 2025)—showing a real-time AI agent conducting a natural, audio conversation as a customer service representative. The system generates its own prompts and operates within pre-set guardrails.

Key Takeaways

• The mortgage sector’s adoption curve for GenAI has dramatically accelerated in just one year.
• The biggest near-term ROI remains in document extraction and underwriting support automation.
• Regulatory and compliance automation is emerging as a high-impact secondary area.
• Conversational AI is moving from experimentation to full production—especially for customer engagement.
• Trust, auditability, and data-grounded outputs are essential to overcoming hallucinations and building user confidence.

Kramer closed by emphasizing that LLMs are inherently non-deterministic, and the key to enterprise adoption lies in setting the right system instructions, verification layers, and human oversight—turning GenAI from an experimental tool into a trusted, productivity-driving partner.

Source: MBA Annual Convention Session – AI and the Future of Work