LLMs for Lease Administrators

Less time looking for information. More time acting on it.

8 min read  •  Part 1 of 2

How much of our day do we spend looking for information? A clause in a lease, a question about whether the landlord or the tenant is responsible for a cost or maintenance item, an amendment that changed the terms, or a portfolio-wide request from an executive that needs an answer today. All of that takes time, and it keeps us from spending our time where it makes the most impact.

This is one area where many of us already use AI. We use it to locate lease language, review drafts, analyze data, and put our findings into a spreadsheet or presentation. This is the first of a two-part series. In this post, I cover the AI tools most of us use today and share how I get reliable results. In Part 2, I look at where I see the biggest opportunity: AI agents connected to our lease management systems.

Using AI in Our Daily Work

Large language models, or LLMs, are behind many of the AI tools we use today. Examples include ChatGPT, Claude, Grok, Google Gemini, and Microsoft Copilot. You ask a question, provide information, and work through the response.

Before uploading a lease or any internal document, though, we need to know which tools our company has approved and what information we can share with them. Leases contain confidential business terms, financial commitments, and information about our operations. The same goes for the spreadsheets, reports, and presentations we use every day.

Use your company’s approved enterprise AI environment, through the account and access method your company has authorized. A personal account with the same provider is not necessarily covered by the same protections, settings, or company agreements. A familiar product name or a paid subscription does not make a tool appropriate for company information.

Approval can also have limits. Follow your company’s rules for document uploads, sensitive data, connected applications, and any actions an AI tool can take on your behalf. If the requirements are unclear, ask the appropriate internal team before you share anything. This will vary by company.

Choose the Tool and Model for the Task

Once you understand those boundaries, get familiar with the available options. Your enterprise environment may offer several tools and different models within each tool. You may even need to request a different or more robust access level for your role. Some basic plans limit the number of file uploads, which is inconvenient when you are working with a lease that has seven amendments.

These choices make a difference. Rewriting an email is very different from analyzing a complicated Excel workbook or building a presentation from several sources. Some tasks benefit from a faster model, while others need stronger reasoning and more time to work through the details. The AI tools that I used at Oracle vs. Kenco vs. Meta perform differently.

In my experience, Claude works best for lease analysis, spreadsheets, and slide presentations, while I get better results from ChatGPT for renderings and visualizations. Those are my observations. Your results may differ depending on the task, your prompts, and the features available to you.

Even within one tool, choosing a model is part of the process. Claude, for example, offers models such as Haiku, Sonnet, Opus, and Fable, depending on your plan. Each balances capability, speed, and cost differently, and those differences shift as new versions are released.

A straightforward task may not need the same horsepower as a workbook with interconnected calculations or a review that draws on multiple documents. Enterprise versions may read low to high vs. the specific models. Read the model descriptions and your company’s usage guidance, then compare the approved options on work you already understand well. They say practice makes perfect, and I can attest that learning to use LLMs effectively and efficiently takes some trial and error.

If a result falls short, look at the instructions you provided, too. Did you explain what you needed? Does the tool have the right files? Did you pick a model suited to the assignment?

Finding and Reviewing Lease Language

Think about reviewing a lease for audit rights. We need to find the provision, understand the deadline, identify any requirements for disputing a statement, and determine whether an amendment changed those rights. An LLM can pull that information together and give us a starting point.

Instead of asking for a general lease summary, we could ask:

Then comes the step we cannot skip: verification.

I learned this firsthand when I asked an LLM to find a specific clause in a lease I had uploaded. It gave me a convincing answer, complete with a section number. The problem… that section did not exist. After some back and forth, it admitted it had made up the reference, explaining that the provision “should” have been in the lease. Well, what should be in a lease and what is actually in the lease are two very different things. I remember thinking how glad I was that I double-checked the results. I also revised my prompt going forward.

Go back to the source. Read the section, confirm the wording, and review the surrounding language, definitions, and amendments. If the tool says a provision is missing, verify that too. “I could not find it” and “it does not exist” are not the same thing.

The same approach works when reviewing a draft lease. Lease administration knows where an agreement becomes difficult to administer. Do we have audit rights, and are the timing requirements favorable? Can we get the supporting documentation we need? How are occupancy costs controlled? Can we administer the lease as written? I cover this in detail in a previous post, Get a Seat at the Table.

We can ask AI to review a draft against those questions or an approved checklist and organize the provisions that need attention. That gives us a focused starting point for the conversation with Legal and the transaction team, while we still have time to change the language.

Working With Data and Communicating the Results

The value goes beyond reading leases. AI tools with the right capabilities can analyze data, build spreadsheets, develop formulas, and turn findings into a slide presentation. Those capabilities vary by tool and company setup.

For us, that might mean comparing rent schedules, organizing reconciliation data, flagging unusual expense changes, or preparing a portfolio update. Sometimes the hardest part is getting started and AI can help by providing a workable first draft instead of a blank spreadsheet or slide.

One use I found especially helpful this past year was asking an LLM to review an operating expense reconciliation against the lease language and flag potential issues, ranked from high to low risk. Then I took it a step further and asked it to compare the current year’s reconciliation with the prior year’s and update its assessment. That told me where to focus my time and energy first. Where would I get the biggest return on my time?

Here is a version of the prompt I used:

This does not replace the knowledge and experience required to audit lease expenses. You still need to understand the terminology and the differences between lease structures, from base year arrangements in modified gross leases to the various types of net leases. You need to understand how controllable expense caps and gross-up provisions work, and you need to verify that the analysis reflects what the lease actually says. Nothing replaces years of working with institutional landlords.

But that initial assessment saved me significant time. It pointed me to the issues with the greatest recovery potential first. With AI supporting the initial review, I reviewed reconciliations and recovered overcharges more than twice as fast as in previous years, and I would say audit recovery increased slightly.

Verification matters here as well.

Build a Prompt Library

When an approach works for a repeatable task, save your prompts. That might be one prompt or a sequence that takes you from the initial review through follow-up questions and verification. Note which tool and model you used, what information it needs, and how you checked the results.

For lease administration, that could become a shared prompt library for reviewing audit provisions, comparing amendments, or analyzing reconciliation data. Update the inputs for each assignment and refine the prompts as you learn. Share them with your team so no one is reinventing the wheel. More people using the same prompts also means more people finding ways to improve them.