AI Strategies for Real Estate Brokers: Unlocking Potential
Many firms struggle to fully harness AI due to confusion around key concepts like prompt engineering, retrieval-augmented generation (RAG), and fine-tuning. Understanding these distinctions and implementing effective AI strategies for real estate brokers is essential for survival in today's rapidly evolving market.
I encourage you to delve deeply into three these pillars of AI strategies for real estate brokers. By thoroughly understanding them, you'll realize that you can unlock the most powerful personal assistant you've ever seen with just a few tweaks in your workflow.
Caution: The ideas presented here are best suited for internal AI tools. Consider serious security concerns like "prompt injection attacks" and "exfiltration." I encourage you to research these topics and follow Simon Willison, who has extensively covered them.
Disclaimer: I'm not a broker. These insights are based on how we've utilized AI at my Mechanical, Electrical, and Plumbing (MEP) firm, and I hope they will be helpful to my broker friends. Shout out to Coy Davidson, Amy Calandrino, Brandon Avedikian, and Topher Stephenson, MBA for helping me understand their world.
The Three Pillars of AI Optimization for Brokers
Prompt Engineering
Retrieval-Augmented Generation (RAG)
Fine-Tuning
1. Prompt Engineering
We're all familiar with prompting at this point (ChatGPT). Prompt engineering involves carefully crafting instructions given to an AI model to produce relevant and targeted responses. It's like asking the right questions to get the best possible answers from a generic source.
Example for Brokers:
When using ChatGPT for preliminary client inquiries, you can:
Attach PDFs about the building, proposed leases, recent client social media posts, and other relevant documents.
Create a prompt requesting a summary of the property and critical lease terms.
Request advice on responding based on what the AI knows about the client and the building.
This method works best when dealing with straightforward data limited to what the pre-trained AI already knows and what you provide during the prompt.
2. Retrieval-Augmented Generation (RAG)
This is the first advanced topic, but it's so essential to your success that I need you to become an engineer briefly for this section.
RAG enhances AI responses by pulling in real-time data from your organization. Here's how it works:
Data Embedding: Convert your data into multi-dimensional vectors (embeddings) and store them in a vector database (like Pinecone).
Similarity Search: Your prompt is transformed into a vector and compared against your database vectors using the dot product (a mathematical operation that measures the similarity between two vectors). The closer the result is, the more relevant the match will be, making it an effective way to find database entries most similar to your input.
Contextual Responses: The results from this search are combined with your prompt to give your assistant real-time business context to help answer your prompt accurately.
Your objective is to connect as many data streams as possible from your company to your vector database.
Examples of Data Streams:
Internal chats
AI summaries from conference calls
Private market reports
Spreadsheets
Site photos and videos
Calendar events
Emails
Recent listings or lease agreements
Questions You Can Ask:
"What is the link for today's meeting?"
"What do I need to know for this meeting?"
"Did we get a signed contract?"
"Tell me what you know about this project."
"Where do we stand on lease terms as of now?"
"Based on what you know, what do we need to move this deal forward?"
Here are some screenshots of our team seeing the benefits of RAG after we spent two days connecting a few datastreams:
Proposed No-Code Tech Stack:
Here's what we use at PermitZIP to prototype ideas. Get your company's biggest nerd to understand these applications as soon as you can:
Make (formerly Integromat)
Pinecone (vector database)
OpenAI (LLM and vector embedding)
3. Fine-Tuning (Training)
Fine-tuning customizes an AI model to your firm's unique needs by retraining it on specific datasets related to your operations and language. This makes the model highly specialized for tasks unique to your business.
Common Misunderstanding: Many people assume that creating a custom GPT means training it, but that's not true. Instead, a custom GPT involves setting system-level instructions to guide its behavior and usually providing data to help it pull relevant information for responses (using RAG).
The Catch: You have to actually teach your robot!
Fine-tuning requires deliberate effort and is time-consuming:
Provide hundreds or thousands of examples to shape the AI's behavior.
Think of this as the same time investment you make for new employees.
Involve senior brokers to contribute their expertise.
Steps to Fine-Tune
Gather Data:
Provide senior brokers with diverse lease term PDFs.
Include market context, building details, proposed renovations, tenant information, occupancy classification, etc.
Structured Responses:
Have them write summary responses on behalf of clients.
Ensure they follow a consistent checklist or internal process.
Scale Up:
Repeat the process multiple times (100, then 1,000).
Regularly update and refine data to keep the AI current.
By doing this, your AI will start reviewing lease terms and projects as your senior brokers would, adhering to your firm's standards and processes.
The Transformation
Implementing RAG and fine-tuning turns your AI into a powerful "employee" who knows everything about your company. It continually improves as you train it to behave according to your preferences.
Why Understanding These Differences Matters
Most commercial real estate brokers are familiar with basic AI tools like ChatGPT. However, with an understanding of how RAG and fine-tuning differ from prompt engineering, brokers can benefit from AI's true potential to drive revenue and enhance client relationships.
The Power of RAG and Fine-Tuning:
Prompt Engineering Alone: Offers general statistics or basic renovation costs if you manually add data to each prompt.
With RAG, the AI accesses proprietary data, internal conversations, emails, meetings, photos, and real-time business data as they evolve over time.
With Fine-Tuning: The AI provides insights aligned with your strategy, such as the benefits of attracting specific tenant types based on historical success.
A Path Forward: Building an Effective AI Strategy
To fully benefit from AI, brokers must develop a strategy beyond surface-level automation. Implement the following steps from most accessible to most complex:
Educate Your Team:
Provide training on AI concepts.
Adjust company infrastructure and processes accordingly.
Start with Prompt Engineering:
Develop well-crafted prompts for basic queries and lead generation. Share with each other what works and what doesn't work.
Integrate RAG:
Connect your AI to internal and external databases.
Deliver responses based on real-time market data and proprietary information.
Invest in Fine-Tuning:
Tailor the AI to your firm's language and workflows.
Use historical data, past transactions, and nuanced industry knowledge.
With these strategies, your firm can leverage AI as a competitive asset that enhances client services, accelerates deal cycles, and provides deeper insights into complex transactions.
Conclusion
I understand why some in our industry associate large language models (LLMs) with "tech bros." The tech industry often needs to grasp the nuances of our field fully. However, many professionals in commercial real estate (CRE) or architecture, engineering, and construction (AEC) also lack an understanding of these essential AI distinctions. It's not just the tech industry holding us back; we must bridge this knowledge gap ourselves.
By embracing these three pillars—prompt engineering, retrieval-augmented generation, and fine-tuning—you position your firm at the forefront of AI innovation in the brokerage industry. The time to act is now.
Comments