Workshop #4 - Give AI the CRE Intelligence It Needs [REPLAY]

Adventures in CRE (A.CRE)
Adventures in CRE (A.CRE)Apr 15, 2026

Why It Matters

Without reliable CRE data and tool integrations, AI-generated analyses risk hallucinations, potentially misguiding investment decisions; curated data platforms are essential for trustworthy AI adoption in real‑estate finance.

Key Takeaways

  • AI output quality hinges on data, not just prompts.
  • Tools enable AI to access external real‑estate datasets.
  • Knowledge layer—up‑to‑date CRE databases—is CRE’s biggest weakness.
  • Demonstration shows raw LLM hallucinations without proper data connectors.
  • CRA Agent offers pre‑built catalog tasks and custom data engines.

Summary

The fourth AI Edge workshop, led by Spencer Burton of CRA Agent, tackled the core challenge of equipping artificial intelligence with real‑estate intelligence. While many participants are seasoned prompt engineers, Burton emphasized that instructions are only one leg of the optimal‑output framework; the other two—tools and knowledge—determine whether AI delivers reliable commercial‑real‑estate (CRE) insights.

Burton broke down the framework: "instructions" (the prompt), "tools" (the external capabilities such as web search, coding environments, or data‑engine connectors), and "knowledge" (the underlying, constantly refreshed CRE datasets). He argued that the industry’s weakest link is knowledge: most AI models rely on stale or generic sources, leading to the classic "garbage‑in, garbage‑out" problem. To illustrate, he walked through a hypothetical Kent Valley Logistics Center offering memorandum, first running Claude without any data connectors.

The raw‑L​LM session produced a polished one‑page investment proposal, but its data provenance was dubious. The model scraped Reddit‑style forums, generic market reports, and even fabricated a 10‑mile radius demographic—information that does not exist in free public feeds. When the AI attempted to pull debt curves, it hit login‑protected sources and fell back to spot rates, further highlighting the hallucination risk. By contrast, connecting the AI to CRA Agent’s curated CRE databases yielded accurate comps, rent validation, and reliable debt metrics.

The takeaway for CRE professionals is clear: without dedicated data pipelines and tool integrations, AI remains a glossy but unreliable analyst. Platforms like CRA Agent that provide pre‑built catalog tasks, custom data engines, and seamless connectors can turn AI from a novelty into a trustworthy decision‑support tool, reshaping acquisition, underwriting, and advisory workflows.

Original Description

In this workshop, Spencer Burton walks through the data problem in CRE AI and how to fix it. He explains why raw AI sounds confident but produces unreliable outputs when it lacks access to real data, demonstrates what happens when Claude analyzes a real offering memorandum with no data connections, and then fixes it step by step using a live Airtable comp database connected via MCP and the A.CRE Intelligence Hub.
Along the way, he covers the Optimal Output Framework (instructions, tools, and knowledge), how to build and connect a production-grade comp database without writing code, what MCP is and why it matters, and what the A.CRE Intelligence Hub delivers for CRE professionals today.
The session closes with a side-by-side comparison of raw AI output versus data-connected AI output on the same deal.
00:00 - Introduction and workshop overview
01:32 - The Optimal Output Framework: instructions, tools, and knowledge
03:57 - Why data is the biggest weakness in CRE AI today
07:25 - The hypothetical deal: Kent Valley Logistics Center
09:07 - Demo: raw Claude analyzes the OM with no data connections
17:45 - AI without data: confident, but often wrong
18:52 - Fix #1: building an Airtable comp database
24:07 - Connecting Airtable to Claude via MCP
26:09 - Demo: Claude queries the comp database and updates its analysis
30:43 - How AI can write back to your database
33:08 - What MCP is and how it differs from an API
33:48 - Fix #2: the A.CRE Intelligence Hub
39:31 - Live data feeds: rates, demographics, employment, and permits
41:28 - Demo: pulling radius demographics and the SOFR curve from the hub
44:11 - The side-by-side: raw AI vs. data-connected AI
46:39 - Q&A: non-disclosure states, Copilot, CoStar, Hello Data, data governance
To learn more and access the related written resource, visit the accompanying blog post: https://www.adventuresincre.com/workshop-4-give-ai-the-cre-intelligence-it-needs
Join AI.Edge free, largest community of CRE professionals leaning into AI: https://aiedge.ac/join
Get the power of AI without having to learn AI: https://creagents.com

Comments

Want to join the conversation?

Loading comments...