AI Pricing Models for CRE Brokerages: 4 Compared

By Jude Lee · · Comparison

Brokerage operations manager and broker reviewing an AI vendor contract and cost spreadsheet in a conference room

Why agentic work breaks the per-seat habit

For fifteen years, brokerage software has been priced per user per month. That made sense when the software was a place humans typed things. An AI agent isn’t a place humans type things — it’s a worker that runs tasks. A single abstraction agent might process 400 leases in a week and touch nobody’s login. Per-seat pricing can’t see that work, so some vendors are reaching for units that can.

The clearest evidence is at the infrastructure layer: the model providers themselves bill by consumption, and Anthropic and OpenAI both publish per-token rates in their own pricing documentation. Document-extraction vendors commonly meter per page or per document. Further up the stack, at least some vendors are experimenting with per-deal or per-matter pricing that bundles agent work with licensed human review rather than selling seats. Treat that as a live experiment rather than a settled shift — but if you sign brokerage vendor contracts, expect to be quoted in units other than seats more often than you were two years ago.

Per-seat pricing assumes work is done by people at desks. Agentic work isn’t, and that mismatch is now a line item in your budget.

Model 1: AI features bundled into the platform you already pay for

This is the default for most brokerages as of 2026. Your CRM, listing platform, or data provider adds AI drafting, summarization, or search, either inside an existing tier or as a per-seat add-on. Pricing is predictable and procurement is trivial — it’s a line change on a renewal you were already signing.

The honest trade: you get the vendor’s interpretation of the workflow, over the vendor’s data, with limited ability to change the output format. If your marketing team has a house style for property one-pagers, a bundled generator that ignores it creates editing work rather than removing it. Feature availability and pricing move fast here, so check the vendor’s own documentation rather than any blog post — including this one.

Watch the utilization math. If you’re paying for an AI add-on across 40 broker seats and six people actually log in, that’s the first line to audit at renewal. Our related breakdown of what a CRE brokerage tech stack actually costs is worth reading alongside it.

Model 2: Usage-based — per document, per run, per token

Document-extraction vendors, API access to models, and most agent platforms price by consumption: per lease abstracted, per page processed, per agent run, or per million tokens. This matches lumpy CRE work well. A 180-unit rent roll and a 60-lease due diligence batch cost real money in the month they happen and nothing in the month they don’t.

The failure mode is cost opacity. Agents that loop — retrying, re-reading long documents, calling tools repeatedly — can consume far more than a naive per-document estimate suggests. Before committing, ask for a metered pilot on your own files, not a demo on the vendor’s.

Per-seat AI add-on
Predictable monthly cost. Easy to budget and approve. Cost is tied to headcount, not output — so it punishes you for hiring and rewards nothing for volume. Best when many people use a little AI, continuously.
Usage-based AI
Cost tracks work actually done. Scales to zero in slow months and spikes during due diligence. Requires caps, monitoring, and someone who understands the meter. Best when a few workflows do heavy, bursty volume.

Model 3: Outcome and per-deal pricing, where you’re buying a deliverable

Here you aren’t buying software at all. You’re buying a finished artifact — an abstracted lease set, a reviewed title package, a legal workstream — from a firm that happens to run AI agents behind the counter with humans reviewing. Pricing is per deal, per package, or per matter.

For brokerages, this is often the cleanest entry point: no implementation, no prompt engineering, and a named party accountable for the output. The trade is that you build no internal capability and your unit cost stays fixed while model costs fall. It also demands a precisely defined deliverable. “Review the lease” is not a specification; “return these 22 fields plus exceptions flagged against our checklist, within two business days” is.

Verify the accountability rather than assuming it. Ask who signs off on the deliverable by name, under what professional license or credential, and what the firm’s errors-and-omissions coverage actually covers — ask for the certificate or policy summary, not a verbal assurance, and check whether AI-assisted work product is excluded. Anything with legal or regulatory stakes — lease interpretation, title exceptions, commission compliance — should keep a qualified professional in the loop regardless of pricing model. See our comparison of AI legal review options for CRE deals for where that line usually sits.

Model 4: Build and run your own agent

A custom build means connecting an AI assistant such as Claude to your own systems — CRM, email, document storage, accounting — typically via MCP (the Model Context Protocol, an open standard for giving an AI governed access to tools and data), optionally with your own MCP server over your listing and deal data, plus packaged skills that make the agent do a job the same way every time.

The cost structure is genuinely different: a defined build cost up front, then ongoing model usage, hosting, and maintenance — with no per-seat tax as you grow. The risk is also genuinely different. You own the breakage. Model updates, a CRM API change, schema drift in your rent roll exports: that’s your afternoon now.

One cost line that rarely appears on a proposal is time. A custom agent earns nothing during the build and stabilization window — the weeks between kickoff and the first run you’d trust unattended. That lag belongs in your payback math alongside build, usage, hosting, and maintenance, because it delays the point where the other four start paying for themselves. Custom wins when the workflow is specific to how your firm operates and stable enough to be worth encoding; it loses when you’re automating something every brokerage does identically. Our longer treatment of custom versus off-the-shelf CRE software walks that decision properly.

Matching the model to the shape of your work

A rough heuristic — our opinion, not a measured finding:

Why there is no single best AI tool for CRE

People look for a single winner. There isn’t one, and any article that names one is guessing on your behalf. The useful reframing: the best tool for a job is the one whose pricing unit matches that job’s volume pattern, whose output format matches what you already send to clients, and whose failure mode you can catch. A general assistant like Claude or ChatGPT is excellent at drafting and reasoning over documents you paste in; a CRE platform’s built-in AI is better at things that need your listing data already loaded; a custom agent wins when the answer requires three of your systems at once. Map it by task first — our task-by-task map of AI in commercial real estate is built for exactly this.

The payback math, without an invented headline

There’s a whole genre of content promising a specific monthly return from AI. Ignore the number; keep the structure. For a brokerage, AI doesn’t generate revenue directly — it converts admin hours into pursuit hours, and pursuit hours into commission probability. Model it with your own inputs:

(hrs saved/mo × loaded hourly cost)
Recovered capacity value
Worked example — use your own figures
(extra pursuits/mo × win rate × avg fee)
Revenue upside, probability-weighted
Worked example — use your own figures
(license + usage + review time + build/maintenance)
True all-in cost
Worked example — use your own figures
(months to first trusted run × monthly value forgone)
Time-to-value lag
Worked example — use your own figures

Two honesty checks most vendor ROI decks skip. First, recovered hours only count if they get reallocated — an hour saved by a broker who then takes a longer lunch is worth zero on the P&L. Second, human review time is a real cost line: if an agent drafts in four minutes and a human verifies in eleven, your savings are measured against the verified total, not the draft. The CRE automation payback calculator walks this through with more structure.

There is no standard “30% rule” for AI spend

If you’ve seen a “30% rule in AI” cited confidently, be careful: there is no single, authoritative definition of it. The phrase gets used for at least three different things — a rough claim about what share of knowledge work AI can absorb, a budgeting heuristic about capping AI spend as a share of a tech budget, and an accuracy-threshold rule of thumb. None of those is a standard published by a recognized body, and we’re not going to pretend otherwise.

A plainer heuristic we’d defend as opinion: if an AI step can’t get a task roughly right most of the time on your own real documents during a pilot, the review burden will erase the savings. Measure that on your files, not a vendor average.

A procurement stress test before you sign

  1. Ask what unit the price actually tracks

    Seat, document, page, run, token, or deal — and what happens at the boundaries. Get the overage rate in writing.

  2. Baseline your own volume for 90 days

    Count the leases, OMs, inquiries, or invoices you really processed last quarter. Vendor estimates are built on someone else’s volume curve.

  3. Price the floor and the ceiling

    Model a dead quarter and your busiest closing month. If the ceiling scares you, you need a cap, not a different vendor.

  4. Interrogate data handling

    Is your client and deal data used to train models, and can you opt out contractually? Where is it stored and processed, what is the retention period, who at the vendor can access it, and what are the breach-notification terms and timelines? Get these in the agreement, not the marketing page — and have counsel review them where client confidentiality obligations apply.

  5. Cost the human review loop

    Who checks the output, how long it takes, and at what loaded rate. Add it to the vendor price — that’s your real unit cost.

  6. Check the exit

    Can you export the structured data, the prompts, and the skill definitions? Who owns the configurations you build? Ask before, not during, a renegotiation.

A last note on “free.” Free tiers of general assistants are genuinely useful for drafting, summarizing, and thinking out loud, and plenty of brokers get real value without spending a dollar. What free doesn’t cover is anything that needs governed access to your firm’s data, an audit trail, or a guarantee the work runs when nobody’s watching. That’s the boundary where paid models — any of the four above — start to earn their keep.

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

Related guides