AI for CRE LOIs and Lease Proposals: 4 Options

By Jude Lee · August 26, 2026 · Comparison

Commercial real estate brokers comparing printed lease proposals and LOIs in a conference room

The short version

Drafting letters of intent and comparing landlord counter-proposals is one of the most repetitive, highest-stakes jobs in brokerage — and one of the least automated. You have four realistic paths: a disciplined template and clause bank with no AI at all; a general AI assistant (Claude, ChatGPT, Copilot) driven by a written drafting skill; AI features inside CRE or contract software you already pay for; or a custom agent wired to your deal folders and CRM through MCP. This piece compares them on setup effort, where each breaks, and what must stay with a human. Short version: skills-driven drafting on a general assistant is the fastest first win for most teams; a custom agent earns its keep once you’re comparing multi-party term sheets often enough that the spreadsheet is the bottleneck.

An LOI is a negotiation artifact, not a form

The reason LOI production resists automation isn’t formatting. It’s that the document encodes a position: which concessions you lead with, how you phrase the renewal option, what you deliberately leave silent. A junior broker handed a blank template produces something legally fine and strategically useless.

That distinction matters for choosing tooling. Anything that only does layout — merge fields, saved templates — solves the smaller half of the problem. What eats broker hours is the second half: taking the client’s parameters, the comp set, the landlord’s last counter, and the market’s current concession norms, and turning them into a coherent set of terms. Then doing it again when four landlords counter in four different formats.

Option 1: A template library and clause bank, no AI

The unglamorous baseline. A well-maintained set of LOI templates by property type and side of the table, plus a clause bank of pre-approved alternates (three versions of the renewal option, four of the TI allowance language), plus a checklist for what has to be filled in before it leaves the office.

If your firm does not have this, build it before you buy anything. Every AI option below gets dramatically better when it has approved language to draw on, and dramatically worse when it’s inventing clauses from general training data. For a broker sending a handful of standard LOIs a month, honestly, this may be the whole answer.

Breaks when: volume rises, product types multiply, or you need side-by-side comparison of counters. Version control degrades fast when five people edit templates.

Option 2: A general AI assistant driven by a written skill

A skill is a packaged, reusable set of instructions that teaches an AI assistant to do one job the same way every time — the drafting standard, the clause bank, the tone, the guardrails, all written down once and invoked on demand. It’s the same pattern we’ve described for standardized property marketing and follow-up, applied to a document with real money attached.

A solid LOI skill contains: your approved template text; the clause bank with rules for when each alternate applies; a required-inputs list (tenant entity, premises, RSF, term, base rent, escalations, TI, free rent, options, brokerage disclosure); an explicit instruction to leave [CONFIRM] markers rather than guess; and a hard rule never to introduce clause language not in the bank.

Setup is measured in hours, not sprints. The assistant reads the client’s requirement summary and last counter, produces a draft in your house format, and lists what it couldn’t determine. That last part is the feature — a draft that admits its gaps is more useful than one that quietly invents a security deposit figure.

Settle the confidentiality question first. Landlord counters, client requirement summaries and deal economics are frequently covered by an NDA, a client engagement term, or your firm’s own data policy — and this, more than any technical limitation, is what stops an Option 2 rollout mid-pilot. Before anyone pastes a counter into a chat window, confirm two things: whether the specific assistant tier you’re on trains on customer inputs (consumer and business/enterprise tiers differ, and the terms change — read the vendor’s current terms, not a summary of them), and whether whoever owns data policy at your firm has approved that tier for client material. Get it in writing once and the rest is easy.

Breaks when: the model doesn’t have access to your systems. You’re still pasting in the comp set, still saving the output manually, still updating the CRM by hand.

A draft that flags what it doesn’t know beats a polished draft that guessed.

Option 3: AI features inside the software you already run

CRE CRMs, e-signature and contract-lifecycle tools, and lease-data platforms have all been adding generative features. Vendor generative features have changed materially within single release cycles, so check current product documentation rather than a conference demo.

Three capability categories are worth testing, and each has a cheap test you can run during a trial:

Clause-level search across executed leases. Pick ten executed leases you already know well — including two you’re certain contain a fixed-rate renewal option and one that doesn’t. Run the search and count misses and false hits. If it can’t find what you already know is there, it won’t find what you don’t.
Key-term extraction into structured fields. Feed it documents your team has already abstracted by hand and compare field by field. Pay particular attention to escalations and option notice windows, where an off-by-one is expensive and invisible.
Redline or counter summarization. Have the person who actually negotiated that deal read the summary and mark what it left out. Omissions matter more than errors here.

The honest tradeoff: bolt-on features are cheap to try, already inside your security perimeter, and require no engineering. They’re also generic by design. They will not know that your firm never grants a fixed-rate renewal option under a seven-year term, because that rule lives in a partner’s head, not in the vendor’s product. The same build-vs-buy logic we lay out in our guide to custom versus off-the-shelf CRE software applies here directly.

Option 4: A custom agent connected through MCP

MCP — the Model Context Protocol — is an open standard for giving an AI assistant secure, governed access to specific data and tools. A custom MCP server exposes your systems: the deal record in the CRM, the folder of PDFs for this requirement, the comp set, the approved clause bank. The assistant can then read and act rather than wait to be spoon-fed. We walk through the architecture in our guide to connecting Claude to your CRE systems via MCP.

For this workflow, the payoff isn’t drafting — it’s comparison. Five landlord counters arrive as five PDFs with five layouts. An agent with document access can extract terms into a normalized grid, compute net effective rent on your firm’s convention, flag where a proposal is silent on something your client requires, and draft the summary email. The extraction half is the same machinery covered in our document review comparison — reliable on clean text, less reliable on scans and handwritten margin notes. Plan the fallback up front: route scanned or handwritten counters to manual entry, and add a flagged-source column to the comparison grid so every field shows whether it was machine-extracted or keyed by a person.

General assistant + skill

Hours to set up. No engineering. Works on any deal immediately. You supply context by hand every time, and nothing writes back to the CRM. Best first move for nearly every brokerage.

Custom MCP agent

Weeks of build plus ongoing ownership. Reads deal data and documents directly, produces normalized comparison grids, updates records. Justified by volume and by comparison work, not by drafting alone.

So which approach is actually right for your desk?

There is no single best tool, and any article that names one without asking about your volume is selling something. A defensible way to decide, stated plainly as our opinion rather than measured fact:

Low LOI volume, standard product types: templates plus a general assistant with a drafting skill. Stop there.
Heavy tenant-rep or landlord-rep flow with multi-party counters: the comparison step is your bottleneck; that’s where a custom agent pays.
You already own a CRE CRM with proposal features: use them for layout and delivery, and let the assistant handle judgment-heavy drafting.

hours saved × loaded hourly cost

The only ROI formula worth trusting here

Fill in with your own figures

× monthly LOI volume

Count your own from the last three months

Your CRM or sent-mail archive

To model it, don’t guess your volume — pull the last three months of sent LOIs and proposals from your CRM or sent-mail folder and use that number. Then time one broker producing an LOI end to end today, time the same task with a drafting skill, multiply the difference by that monthly volume and your loaded hourly cost. Add the part most people skip: what the recovered hours get reallocated to. An hour returned to prospecting is worth more than an hour returned to inbox triage, but only if someone actually redirects it. McKinsey’s ongoing State of AI survey research tracks precisely this gap — organizations reporting AI use versus organizations reporting bottom-line impact from it. Pilots are easy; workflow change is not.

Standing up the proposal-comparison workflow

Freeze the clause bank

Have counsel approve the template and every alternate clause. Nothing downstream is safe until this exists as a file, not a habit.
Write the drafting skill

Required inputs, house style, when each alternate applies, and an absolute rule to mark unknowns rather than fill them. Test it on five closed deals where you know the right answer.
Define your comparison grid

Decide the exact fields and the exact net effective rent convention your firm uses. An agent cannot normalize what you haven’t standardized.
Run it in shadow mode with a defined exit

For a month, have the agent produce grids alongside the analyst’s spreadsheet. Log discrepancies by field and by source type. Exit shadow mode only when clean-text extraction runs discrepancy-free across a full month and every scanned or handwritten counter is being routed to manual entry with the source column flagged.
Connect systems only after the manual version works

MCP access to the CRM and deal folders is worth building when the process is proven, not before.

What stays human, permanently

Strategy and delivery. What you concede first, what you hold, when to call the landlord’s broker instead of sending anything — none of that belongs to an agent. Neither does the final read of an outbound document. Treat AI output as a first draft prepared by a fast, tireless, occasionally confident-and-wrong analyst, and route anything non-standard to counsel.

Two heuristics worth deflating

The 2% rule originates in residential rental investing — the idea that monthly gross rent should equal roughly 2% of purchase price — and it does not translate cleanly to commercial assets, which are valued off net operating income and cap rates. Treat it as a rough screening heuristic, not an underwriting standard. Defensible screening logic belongs in your underwriting automation, tied to real comps.

The “30% rule” in AI is not a recognized industry standard, despite how often it circulates. Different people use it to mean a share of tasks to automate, a failure-rate heuristic, or a confidence threshold. If a vendor cites it at you as established, ask which primary source they mean. Our practical version: automate the steps where a wrong answer is cheap and visible, keep humans on the steps where a wrong answer is expensive and quiet. LOI strategy is firmly in the second category.

Not sure where to start?

Get a free automation audit: we map your deal pipeline, marketing, and back-office workflows and show you what's worth automating — before you spend a dollar.

Get a free automation audit