AI for CRE Lease Comps: 4 Options Compared

By Jude Lee · September 8, 2026 · Comparison

CRE broker and analyst reviewing a lease document next to a laptop showing a comps spreadsheet

The data your brokerage already owns and can’t find

Every deal your team closes produces comps: base rent, term, escalations, TI allowance, free rent, expense structure, options. That data is more current and more specific than anything you can buy, because it’s yours. And in most shops it lives in signed PDFs on a shared drive, in an analyst’s spreadsheet, and in the head of the broker who did the deal.

The result is familiar. A broker prepping a proposal asks the team Slack channel “anyone have recent industrial comps in the north submarket?” and gets three answers, two of them stale. JLL’s piece on getting out of the slow lane in real estate AI makes a point worth repeating: the constraint is usually organizational, not technical. Comps are the clearest example — extraction technology has been adequate for a while; the discipline to capture consistently has not.

What AI comp abstraction actually does

Here is the honest version. Modern language models read a lease, LOI, or abstract and pull specified fields into structured output — tenant, premises, RSF, commencement, term, starting rent, escalation basis, free rent months, TI per foot, renewal options, expense treatment. They do this well on clean documents with standard language, and fast enough that abstraction stops being a scheduling problem.

They break in predictable places: scanned faxes and handwriting, amendments that restate rent schedules, rent expressed as an annual lump sum in one clause and per-foot in another, and anything requiring judgment about whether an unusual concession belongs in effective rent. An AI agent does not know your firm’s convention for net effective rent unless you write that convention down.

An AI can read a lease in seconds. It cannot decide what your firm counts as effective rent — that’s a policy you have to write, once, and enforce.

What makes this agentic rather than a chat trick is the extra steps: watching a folder or inbox for new executed documents, extracting fields, checking them against existing records, flagging conflicts, and writing the clean row into your CRM or comp table without a person copying and pasting. Same extraction, very different operational value. The pattern mirrors rent roll and lease abstraction automation, applied to comps instead of ownership analysis.

Option 1: An analyst types it in

Still the right answer for some firms. If you close a modest number of leases a quarter and one person already handles deal files, manual entry is cheap, auditable, and produces exactly the fields you want.

Best when: low volume, one clear owner, and you’d rather spend budget on data you can’t produce yourself.

Breaks down when: volume rises or the analyst gets busy. Comp entry is the first task dropped, so the database goes quiet for a quarter and never fully recovers.

Option 2: A subscription comps platform

CoStar, Crexi and CompStak all sell comp data in some form: CoStar as a broad subscription research database covering lease and sale comps across markets, Crexi as a marketplace whose listing and transaction activity feeds its comp product, and CompStak as an exchange where brokers contribute lease comps in return for access. Verify current coverage, contribution requirements and contract terms directly with each vendor’s official documentation and sales team — submarket depth varies enormously and is the only thing that matters. We compare the broader platforms in CoStar vs Crexi vs Reonomy.

Best when: your gap is breadth — new submarkets, asset types you don’t transact in, or sale comps you can’t source internally.

Breaks down when: you need your comps. A purchased database doesn’t contain the concession package you negotiated last month unless you or a counterparty put it there. Most firms end up needing both.

Option 3: A general AI assistant plus a fixed template

Upload the lease to Claude, ChatGPT, or Copilot with a standing instruction defining every field, its units, and the fallback when a term is absent. Paste the result into your comp sheet. This is the cheapest real improvement available, and it works on free or low-cost tiers.

The upgrade that costs nothing is turning the template into a skill — a written, reusable instruction set the assistant loads every time, so the abstraction comes out identically regardless of who runs it. See agent skills for CRE marketing for the same pattern elsewhere.

Best when: you want consistency this month, with no procurement and no engineering.

Breaks down when: dispatch and paste are still manual, nothing validates against existing records, and adoption depends on someone remembering.

Option 4: A custom agent over your own comp database

Here the assistant connects to your systems through MCP (the Model Context Protocol — an open standard for giving an AI governed access to specific tools and data). You expose a small set of operations: search comps, insert comp, flag duplicate, read the executed-lease folder. Then Claude or another MCP-capable assistant can both file new comps and answer questions against the set — “show me flex leases over 20,000 SF signed in the last 18 months with more than three months free rent.”

The engineering is contained if three things are already true: your comps sit in a structured table or CRM object, you have in-house or contract developer capacity, and authentication to that system is already solved. Then it is a server wrapping your database plus a handful of tool definitions — mechanics in connecting Claude to your CRE systems via MCP. It stops being contained when your comps exist only as PDFs, when field definitions differ by team, or when nobody owns conflict resolution — a query interface over bad data produces confident wrong answers faster.

Best when: proprietary comps are something you actually sell on, and one named person owns data quality.

Breaks down when: definitions are unsettled, volume is low, or the build has no internal maintainer after launch.

Buy a comps platform

Breadth immediately. No engineering. Predictable subscription. Coverage gaps in secondary submarkets. Your proprietary terms stay outside it unless you contribute them.

Build an agent over your own comps

Depth on the deals you actually did. Queryable in plain English. Requires field governance, a data owner, and ongoing maintenance. Worth it mainly if your comps are a differentiator.

Modeling the payback without making up numbers

Don’t take anyone’s ROI headline, including ours. Build it from your own inputs:

leases/quarter × min per abstract

Current abstraction load — measure it for two weeks

hours saved × loaded hourly rate

Recovered capacity, only if reallocated to revenue work

proposals won / lost on comp quality

The variable that usually dwarfs the hours

The recovered-hours math is the easy part and probably the smaller part. The bigger effect is proposals that go out with current, defensible comps instead of a broker’s recollection — harder to measure, likely more valuable. Model it as a judgment call, not a number. Fuller framework: the broker-ops payback model.

Two rules of thumb, handled honestly

The 2% rule — monthly rent at or above 2% of purchase price — comes from residential rental investing and circulates in CRE as shorthand. It is not an underwriting standard from any recognized industry body, and in most current markets it screens out nearly everything. Cap rate, effective rent and lease credit do the actual work.

The 30% rule in AI is not an established standard either. People use it to mean automate roughly a third of a task, or trust output about 30% of the way; there is no authority behind either reading. A better rule for comps: automate extraction fully, automate filing conditionally, and require human sign-off on any figure that reaches a BOV or proposal.

What the comp workflow says about broker replacement

AI can read every lease you’ve ever signed. It cannot call the landlord’s counsel, read the room on a tour, or know that a particular owner will trade rent for term. What it does displace is the unbilled evening spent typing lease terms into a spreadsheet — which isn’t brokerage, it’s overhead that happens to be performed by brokers. Firms that hold that distinction clearly tend to adopt faster and argue less.

A 30-day sequence to get comps under control

Write the field spec first

One page: every field, its unit, its source clause, and your convention for effective rent and concessions. No tooling decision until this exists.
Run 20 leases through a general assistant

Use the spec as a skill. Have your analyst score accuracy field by field. You now know which fields are safe to automate and which need eyes.
Decide buy vs build with real evidence

If coverage is your gap, subscribe. If your own closed deals are the gap, keep going.
Automate the boring half

Watch the executed-documents folder, extract, and stage rows for one-click approval. Staged-not-published is the right default for at least a quarter.
Only then open query access

Once the data is trustworthy, connect an assistant via MCP so brokers can ask questions directly — the principle behind AI search over your deal history.

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