As generative AI (GenAI) gains traction in the rental housing sector, some larger organizations are asking a big question:

Should we build our own custom large language model (LLM)?

The idea of a purpose-built AI system designed specifically for your business is enticing. After all, who wouldn’t want a tool that deeply understands your operations, resident needs, compliance requirements, investment history, and development pipeline?

But building a custom LLM is no small feat. It’s a complex decision that involves significant trade-offs in terms of cost, expertise, security, and long-term value.

Some Organizations Are Evaluating Custom LLMs

For many organizations, the idea of building a custom LLM stems from the desire for precision, control, and sustained competitive advantage. Our experiences today point to a glaring problem: generic GenAI struggles to fully grasp the nuances, data, and processes of rental housing operations, like compliance with specific regional regulations or tailored lease terms. Custom LLMs promise a model trained exclusively on your organization’s proprietary data, delivering outputs that reflect your unique standards, policies, and workflows.

Additionally, concerns about data security, privacy, and model training are driving interest in custom solutions. With a custom LLM hosted in-house, sensitive operational and financial data remains within your control - addressing potential vulnerabilities tied to third-party providers. For organizations looking to differentiate themselves in a competitive market, owning a proprietary AI system may also serve as a long-term strategic asset, positioning them as leaders in innovation. Simply put: if an organization is investing in the 10X opportunities created with AI, it may be a better idea to minimize the risks & maximize the benefits that more generic AI tools can't provide.

Is There a Middle Ground?

A promising middle ground for the rental housing sector is the idea of an industry-level custom LLM. Instead of individual companies shouldering the burden of building and maintaining their own models, a real estate association or consortium could make the investment to develop a purpose-built LLM for the industry. This shared resource could be trained on a mix of public real estate data, aggregated best practices, and common operational workflows, offering significant value to participating businesses.

Organizations could then license access to the model, gaining the benefits of a tailored, industry-specific AI without the prohibitive costs or technical challenges of building one from scratch. This approach would also enable collaboration across the sector, driving innovation while ensuring that smaller players have access to cutting-edge tools.

(Note: Our advisors at The Strategic Edge are evaluating this path. If you're interested to learn more or get involved, book a call with our team to discuss.)

The Pros of Building a Custom LLM

1. Tailored to Your Business Needs
   A custom LLM can be trained on your organization’s unique data flowing across your organization - communications, policies, documentation, operational workflows, financial reporting, pipelines, and more. This ensures the model understands the nuances of your business, delivering outputs that are more accurate and relevant than a generic AI.

   Example: Instead of using a generic AI to draft lease agreements or calculate potential future revenue, a custom LLM could generate contracts preloaded with your organization's specific terms, local regulations, and resident preferences.
   
   

2. Enhanced Data Security
   With a custom LLM built and hosted internally, sensitive or privileged data never leaves your organization’s environment. This can significantly reduce the risk of data breaches and ensure compliance with privacy regulations.

   Example: Organizations handling subsidized housing or sensitive investor data could benefit from tighter control over their AI's data handling.

   
   

3. Long-Term Strategic Advantage
   Investing in a proprietary LLM positions your organization as a leader in AI innovation. It allows you to build capabilities that competitors using off-the-shelf tools may struggle to replicate.

   Example: A custom LLM that anticipates resident needs or optimizes maintenance schedules could become a core differentiator in how you deliver value to residents - or other stakeholders.

   
   

4. Integration with Internal Systems
   A custom LLM can be deeply integrated with your existing software ecosystem, from property management platforms to CRM systems. This enables seamless workflows and AI-driven insights tailored to your operations. Generic AI options have very limited integration points specific for the real estate industry.

The Cons of Building a Custom LLM

1. High Initial Costs
   Developing a custom LLM is expensive. Beyond the cost of infrastructure, you’ll need to hire or contract AI experts to design, train, and maintain the model. Training LLMs also requires significant processing & storage resources - big dollars if you’re starting from scratch.

   Alternative: Pre-trained LLMs from providers like OpenAI or Anthropic allow you to leverage powerful AI without the heavy lifting of building your own.
   
   

2. Maintenance and Updates
   AI models aren’t static; they require constant updates to stay relevant and effective. Maintaining a custom LLM means regularly retraining it with new data, updating it to reflect the evolution of your business, and monitoring for bias or performance issues.

   Reality Check: If your team isn’t prepared for ongoing investment, the model’s performance could degrade quickly, making it more of a liability than an asset.
   
   

3. Expertise Gap
   Building and running a custom LLM requires specialized skills in AI development, machine learning, and data engineering. Many rental housing organizations don’t have this expertise in-house, which means relying on external vendors or hiring new talent—both of which add cost and complexity.

   Consideration: If tech isn’t your core competency or differentiator, it might be better to work with a vendor that specializes in GenAI.
   
   

4. Security Isn’t Automatic
   While keeping an LLM “in-house” may seem more secure, it also makes your organization solely responsible for protecting it. That means implementing robust data governance policies, setting up secure environments for training and hosting, and continuously monitoring for vulnerabilities.

   Pitfall: Without dedicated resources for security, an in-house model could be more exposed to risks than a third-party solution.
   
   

5. Missed Opportunities with Pre-Trained Models
   General-purpose models like OpenAI’s GPT-4 or Anthropic’s Claude are continuously improved by their developers, benefiting from cutting-edge research and broader datasets. By going custom, you might miss out on these advancements and have to rely solely on your internal team to keep the model humming smoothly.

Key Factors to Consider

If your organization is evaluating whether to build a custom LLM, ask yourself these questions:

1. What’s the Scale of Our Operations?
   Custom LLMs are best suited for large organizations with complex, high-volume operations that justify the investment. Smaller organizations may find pre-trained models more cost-effective and flexible.
   
   

2. Do We Have the Expertise or Resources?
   If your team lacks deep AI expertise or the budget to outsource it, a custom LLM may be out of reach. Consider whether your resources are better spent customizing pre-trained models.
   
   

3. What’s the Use Case?
   For niche, high-value applications—like advanced compliance analysis or tenant experience optimization—a custom LLM might make sense. But for general workflows, pre-trained models are often sufficient.
   
   

4. Can We Sustain This Over Time?
   The investment doesn’t end at launch. Make sure you’re prepared for ongoing costs, from retraining to security updates to system integration.

Final Thoughts: A Balanced Approach

For most rental housing organizations, the sweet spot lies between fully custom LLMs and generic off-the-shelf tools. Many real estate-focused AI providers now offer pre-trained models that can be fine-tuned with your data—giving you the best of both worlds: tailored performance without the heavy lift of building from scratch.

The key is to focus on the right tool for the right job. If the task at hand is complex, high-value, and central to your competitive advantage, a custom LLM may be worth the investment. But if your goals are more general, or if you're testing the AI waters before jumping in too deep, a fine-tuned, pre-trained model will likely deliver 10X value at a fraction of the cost and effort.

As with any major decision, it comes down to alignment with your strategy, budget, and long-term goals. AI isn’t just a tool: it’s a lever for transformation. Make sure you’re pulling it in the right direction.