Spencer Dobbs | Senior Analytics Engineer

Beyond the Hype

Everyone's talking about Gen AI, but in enterprise settings, the question isn't "is this cool?" but "is this useful?" After months of experimentation, I've found several high-value applications in real estate analytics.

Document Processing

Real estate generates mountains of documents—leases, inspection reports, maintenance logs, vendor contracts. LLMs excel at extracting structured data from these unstructured sources.

We've built pipelines that can:

Extract key terms from lease agreements

Summarize inspection reports into standardized formats

Flag anomalies in vendor invoices

Natural Language Analytics

Giving non-technical stakeholders the ability to query data in natural language has been transformative. Instead of waiting for the analytics team to run a report, portfolio managers can ask questions directly.

Predictive Narratives

One unexpected application: using LLMs to generate narrative explanations of our predictive models. When a property is flagged as high-risk, the system can explain why in plain English, making our models more trustworthy and actionable.

The Challenges

It's not all sunshine. We've had to carefully manage:

Hallucination risks (especially with financial data)

Cost optimization for high-volume use cases

Integration with existing security frameworks

What's Next

I'm particularly excited about multi-modal applications—combining property images, documents, and structured data into unified models. The potential for more nuanced property understanding is enormous.