.png)
Frameworks, core principles and top case studies for SaaS pricing, learnt and refined over 28+ years of SaaS-monetization experience.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
What Makes Commercial Real Estate AI Pricing so Incredibly Complex
Commercial real estate is an unusually hostile environment for simple AI pricing. A customer-service agent can charge when a ticket is resolved. A general productivity assistant can charge for each employee who uses it. A CRE system, by contrast, may screen an acquisition, extract a rent roll, compare a property with prior deals, monitor a portfolio, draft an investment memo and surface a lease risk that changes a decision months later. Dealpath and Prophia already span several of those workflows, while academic research has long shown that private commercial properties are heterogeneous, spatially dispersed, infrequently traded and priced without a central exchange.
Pricing mistakes therefore compound quickly. Charge by seat and successful automation can remove the very users that drive ARR. Charge by token or AI action and falling inference costs pull the customer’s price anchor towards a cost that technology keeps compressing. Charge only when a building sells, and revenue becomes hostage to deal timing, bargaining and market liquidity rather than the amount of work the AI actually performs. Research published in 2021 found that search frictions and bargaining can materially affect CRE transaction prices, while earlier work found that price information from public real-estate markets can take a year or more to transmit fully to private property markets.
Monetizely’s position is that commercial real estate AI should be sold through an annual enterprise commitment whose primary meter is the number of active properties being analysed or managed. Seats should govern access, compute should remain an internal cost, and deal, lease or document bands should protect the economics of unusually heavy workflows. Pure transaction-outcome pricing should remain a premium layer for the few outcomes that can be measured and attributed cleanly.
CRE combines durable assets with episodic decisions, so no conventional SaaS meter fits cleanly
The pricing problem starts before anybody sets a rate. Monetizely’s 5-Step Pricing Framework, developed further in Monetizing Agentic AI, forces the decisions into the right order. Goals and segmentation establish what the business needs pricing to achieve and which buyers it serves. Packaging decides what each segment actually buys. Pricing metric determines the unit that makes revenue move with customer value. Rate setting puts a price on that unit. Operationalisation makes the contract work in metering, entitlements, billing and renewal. The sequence matters here because choosing “tokens”, “users” or “deals” before deciding what kind of CRE work the product replaces almost guarantees that the meter will be wrong.
The public market already demonstrates how many plausible answers exist. As of 21 August 2026, leading B2B AI and CRE software vendors are charging against outcomes, actions, users, leads, tenants, properties and negotiated enterprise scope. Their choices are commercially coherent in isolation. Put them side by side, however, and the problem facing a CRE AI vendor becomes obvious.
| Vendor / product | Public pricing model | Primary published meter or contract driver | Public position as of 21 Aug 2026 | Source |
|---|---|---|---|---|
| Intercom Fin | Outcome, or hybrid with helpdesk seats | Fin outcome | $0.99 per outcome; Intercom annual plans start at $29 per seat/month | |
| Salesforce Agentforce | Consumption, per-user and flat-fee options | Actions, conversations or users | $500 per 100,000 Flex Credits; standard agent action uses 20 credits; $2 per conversation; $5 user licence plus credits; flat-fee access also offered at $125/user/month | |
| Microsoft 365 Copilot | Per seat | User | $30 per user/month, paid yearly on the US enterprise page | |
| 11x Alice | Packaged commitment with lead-volume envelope | Leads / prospects | Growth starts at $3,750/month billed annually, with up to five end users and 2,000 new prospects per month; 11x says it charges per lead rather than per send | |
| Reonomy | Seat plus export usage | User, with export allowance | Annual plans start at $400/month; pricing is typically per user; annual access includes 1,000 monthly exports | |
| Prophia Essentials | Annual CRE-specific subscription | Tenant | “Per Tenant - Annual Subscription”; unlimited users without per-licence fees; pricing is quoted | |
| Dealpath AI | Negotiated platform subscription | Enterprise scope, with user minimum | Dealpath says pricing is tailored to firm needs, usually with at least five users and no required deal volume; AI capabilities sit inside the deal platform | |
| CoStar / Matterport spatial platform | Fixed subscription with multiple rate drivers | Sites, users, properties, digital twins and other contract factors | CoStar’s Q2 2026 filing says it generally favours fixed monthly subscription charges over actual platform usage, while rates can reflect properties analysed, users, sites, geography, digital twins and other variables |
The source pages support each current pricing description above. -
The pattern matters more than any individual price. Fin can identify a billable resolution. Microsoft can identify the human receiving Copilot. Prophia can identify the tenant whose lease data is being maintained. CoStar, after decades of selling real-estate information, disclosed in its Q2 2026 10-Q that subscription pricing may incorporate the number of properties analysed alongside users, sites, organisation size, geography and several other factors.
CRE AI inherits all of those possibilities at once. An investment firm may have ten users but thousands of opportunities. An owner may have hundreds of buildings, thousands of leases and only a small asset-management team. A brokerage can process enormous listing volume while transaction revenue arrives irregularly. Charging one of those customers according to another customer’s economic unit creates predictable distortions.
Higher autonomy removes the user from the value equation before it removes human judgement
The Agentic Monetization Spectrum, or AMS, tells us which distortions matter most. It evaluates an AI product along three dimensions rather than asking whether it carries the fashionable label “agent”. Zero-human ability measures how much of the work the software completes without a person doing the production task. Operational domain measures whether it handles one task, an end-to-end workflow or work across several domains. Output/cost ratio asks whether customer value rises roughly with compute cost, begins to outrun it, or becomes dramatically larger than it. The higher a product moves across those dimensions, the weaker a human seat becomes as the pricing anchor and the stronger an output or business unit becomes.
For the scores below, Small means the human remains central, Medium means the system executes substantial work with human review, and Large means it can perform most of the work itself. For the output/cost dimension, Linear means value broadly follows delivery cost, Inflecting means value begins to pull materially ahead, and Exponential describes cases where the business output can be worth orders of magnitude more than the compute needed to produce it. These are Monetizely assessments of the products described by their vendors as of 21 August 2026, not vendor self-ratings. -
| Product | Zero-human ability | Operational domain | Output/cost | What the AMS says about its meter |
|---|---|---|---|---|
| Intercom Fin | Large | Medium | Exponential | Outcome pricing is credible because the agent can complete a bounded customer-service job |
| Salesforce Agentforce | Large | Large | Inflecting to Exponential | Consumption can bridge many workflows, but a generic action becomes less attractive as value per action diverges |
| Microsoft 365 Copilot | Small | Large | Inflecting | Per-seat remains logical while a human employee is the primary operator and beneficiary |
| 11x Alice | Large | Medium | Inflecting | Lead volume is closer to work delivered than human-seat count |
| Reonomy | Small | Medium | Inflecting | The human still performs the investment or brokerage decision, so conventional subscription pricing remains defensible |
| Prophia Essentials | Medium | Medium | Inflecting | Tenant-level pricing fits the lease workload better than user count, especially because expert review remains in the service |
| Dealpath AI | Medium | Medium | Inflecting | AI can execute screening and data work, but the investment professional still owns the capital decision |
| CoStar / Matterport AI spatial platform | Small | Medium | Inflecting | Property and digital-twin scale are stronger economic anchors than AI queries |
The AMS produces the non-obvious finding in CRE: the category is moving away from per-seat economics faster than it is becoming suitable for pure outcome pricing. Dealpath’s current AI suite can ingest offering memoranda, screen deals, surface comps and connect proprietary portfolio information into AI tools, yet its product page still describes human investment teams making the go/no-go, underwriting and portfolio decisions.
Prophia presents the same tension from another direction. Its Essentials product combines AI with expert human review and explicitly prices per tenant rather than per user; the company says organisations can give access to asset management, leasing, accounting, legal and outside partners without adding licence fees. As of 21 August 2026, that is a useful CRE precedent: when many people collaborate around the same underlying economic object, the user is not necessarily the thing worth monetising.
Real-estate outcomes are too delayed and negotiated to carry the core price
Pure outcome pricing sounds especially attractive in property because the outcomes are enormous. Help a fund avoid a bad acquisition, negotiate a stronger lease, improve occupancy or exit a property at a better price, and the economic value can dwarf any software bill. Yet large value does not automatically create a good billing meter.
The underlying market is the first obstacle. Devaney and Martinez Diaz wrote in the Journal of Property Research in 2011 that private CRE assets are heterogeneous and spatially dispersed, trade infrequently and lack a central marketplace where prices and cash flows are continuously observable. Their analysis also discusses well-known weaknesses in appraisal-based indices, including lag and smoothing concerns.
Price discovery creates a second problem. Barkham and Geltner found in Real Estate Economics in 1995 that information from securitised property markets did not fully transmit into private property markets for a year or more. A later 2021 Journal of Property Research model found that search and bargaining themselves can materially affect the equilibrium transaction price of commercial property.
Consider what that means for an acquisition agent. The system may screen 2,000 opportunities, identify 50 worth underwriting, help the team bid on 12 and contribute to three purchases. The final purchase price still reflects the seller, competing bidders, financing conditions, timing, due diligence and human investment judgement. Charging a percentage of “value created” requires somebody to prove which portion of the result came from the AI.
Customer service is structurally different. Intercom can define a Fin outcome inside one conversation and charge $0.99 when the agreed billing condition occurs. CRE outcomes can take months or years and may have several plausible causes. As of 21 August 2026, that difference makes Fin’s outcome architecture instructive without making it directly portable to real estate.
There are still cases where an outcome kicker makes sense. A vendor can share in a clearly measured reduction in lease-abstraction cost, a defined volume of completed due-diligence work, or another result with an agreed baseline and short measurement window. Monetizely’s position, however, is that the property transaction itself is usually too sparse, delayed and contested to serve as the primary revenue meter.
Compute prices fall faster than property value, so token-based pricing gives away the upside
The opposite temptation is to price close to technical consumption. Tokens and credits feel safe because AI has real variable cost. Salesforce’s Agentforce pricing page on 21 August 2026, for example, lists Flex Credits at $500 per 100,000 credits and specifies 20 credits for a standard Agentforce action, equivalent to $0.10 for that action before other relevant services.
That architecture can protect gross margin while usage is uncertain. It becomes much less attractive as the long-term value anchor, because the underlying cost is not stable.
OpenAI’s published prices demonstrate the speed of movement. GPT-4 launched in March 2023 at $30 per million input tokens and $60 per million output tokens. GPT-4.1 launched in April 2025 at $2 and $8 respectively; OpenAI explicitly attributed lower GPT-4.1 prices to inference-system efficiency and said median queries cost 26% less than GPT-4o. On 21 August 2026, its GPT-5.6 family ranged from $0.20/$1.20 per million input/output tokens for Luna to $5/$30 for Sol.
| OpenAI reference point | Date | Input per 1M tokens | Output per 1M tokens | Pricing implication |
|---|---|---|---|---|
| GPT-4 | Mar 2023 | $30.00 | $60.00 | Expensive model use could once look like a natural customer meter |
| GPT-4.1 | Apr 2025 | $2.00 | $8.00 | Input list price was about 93% below GPT-4’s 2023 launch rate; output was about 87% lower |
| GPT-4.1 nano | Apr 2025 | $0.10 | $0.40 | Routine work could be routed to a model priced far below the flagship |
| GPT-5.6 Luna | 21 Aug 2026 | $0.20 | $1.20 | Current low-cost tier |
| GPT-5.6 Sol | 21 Aug 2026 | $5.00 | $30.00 | Current premium tier, 25 times Luna’s input and output rates |
The historical and current prices come directly from OpenAI’s published research and API pages. -
The lesson is not that every future model will always be cheaper than every predecessor. OpenAI’s August 2026 range shows that more capable tiers can still command higher rates. The important commercial fact is that a vendor can increasingly route routine extraction, classification or summarisation to much cheaper models while reserving expensive reasoning for harder work. OpenAI itself offered GPT-4.1 nano at one-twentieth of GPT-4.1’s April 2025 input and output rates, and the August 2026 GPT-5.6 family spans a 25-fold price range between Luna and Sol.
A CRE vendor that marks up tokens therefore risks turning engineering progress into customer price cuts. Suppose a property analysis once needs $10 of inference and can later be produced for $2 through routing, caching and cheaper models. A cost-plus contract pressures the selling price down just as the vendor has improved its technology. The building being analysed did not become 80% less valuable because the model bill fell.
That is why compute belongs in the rate-setting model, not in the customer’s definition of value. Engineering should know cost per property, lease and deal. The buyer should not have to care which model processed the rent roll.
The active property is the strongest anchor for CRE AI pricing
A good CRE meter must survive several changes simultaneously: fewer human operators, more autonomous work, cheaper inference, uneven deal volumes and expanding portfolios. We can make that requirement explicit by scoring the main candidates from one, weak, to five, strong.
The scoring below is Monetizely’s analytical judgement rather than market data. The criteria reflect what the preceding evidence makes commercially important: proximity to customer value, budget predictability, resilience as AI replaces human work, resilience as inference cost falls, and the ability to grow with the customer’s CRE activity.
| Primary meter | Value alignment | Buyer predictability | Survives rising autonomy | Survives falling inference cost | Captures customer growth | Total / 25 |
|---|---|---|---|---|---|---|
| Seat | 2 | 5 | 1 | 5 | 2 | 15 |
| Token / AI action | 2 | 2 | 4 | 1 | 3 | 12 |
| Closed transaction / outcome | 5 | 1 | 5 | 5 | 2 | 18 |
| Active property | 4 | 4 | 5 | 5 | 5 | 23 |
The active property wins because it follows the durable economic object around which CRE work is organised. CoStar’s Q2 2026 filing gives this idea unusually strong market support: the company says its subscription rates can depend on the number of properties reported on or analysed and the number of digital twins hosted, alongside other account characteristics. Prophia’s current per-tenant model makes the same broader point from lease intelligence: CRE-native units can displace user licences when the underlying work naturally scales with the real estate itself.
“Active property” should be defined operationally. A property counts while it is being screened, underwritten, managed or reported through the product during the agreed billing period. High-volume lease products can then add tenant or lease bands, and unusually document-heavy acquisition teams can buy additional processing bands, without changing the primary meter.
A simple sensitivity model shows why the choice matters. Suppose four competing pricing structures are each calibrated to produce $360,000 of Year 1 ARR from the same institutional customer. The model below then subjects each to the pressure most likely to weaken it as AI matures.
| Modelled primary meter | Year 1 construction | Year 1 ARR | Year 3 pressure case | Year 3 ARR |
|---|---|---|---|---|
| Seat | 30 users × $12,000/year | $360,000 | Automation reduces paid operating users to 18 | $216,000 |
| AI consumption | 60,000 billable tasks × $6 | $360,000 | Tasks double to 120,000 while competitive unit rate falls to $2 | $240,000 |
| Closed deal | 12 transactions × $30,000 | $360,000 | Market slowdown produces six closings | $180,000 |
| Active property | 600 properties × $600/year | $360,000 | Portfolio / active evaluation base grows to 720 | $432,000 |
The model exposes the strategic choice. Seat pricing can punish the vendor for automating labour. Consumption pricing can punish it for lowering delivery cost. Closing-based pricing can punish it when the capital market freezes even if customers continue screening assets. Property-based revenue grows when the customer places more real estate under the system’s care.
That architecture should still have a committed annual floor. Large CRE customers require implementation, integrations, data controls, permissioning and support whether a portfolio is unusually busy that quarter or not. Dealpath, for example, says its 2026 plans include white-glove implementation and are tailored to the firm, while CoStar reported that the majority of its subscription agreements have terms of at least one year.
The resulting model is deliberately asymmetric: active properties drive price; usage limits protect margin; seats control access; outcomes provide selective upside. There is still one primary meter, and it is attached to the customer’s real-estate footprint rather than the vendor’s model architecture.
The winning contract will make the property count primary and everything else secondary
Commercial real estate makes AI pricing difficult because three economic systems collide. Traditional software teaches buyers to think in licences. AI introduces genuine variable compute cost and increasingly autonomous work. Real estate itself creates value through long-lived assets, negotiated transactions, leases and portfolio decisions whose outcomes arrive on very different clocks. The public evidence available by August 2026 shows all three systems operating at once. -
Trying to collapse that complexity into token pricing does not simplify the economics. It merely transfers a fast-changing engineering cost into the customer contract. Keeping per-seat pricing does not solve the problem either, because the AMS tells us that autonomous CRE products will increasingly perform work without a one-to-one relationship with human users.
Our recommendation is therefore not “use every meter”. The architecture needs a hierarchy. The property is the primary commercial unit because it remains economically relevant after the user count changes, after model costs change and between transactions. Secondary meters exist to handle real differences in workload rather than to avoid making a choice.
For operators making that choice now, four decisions matter most:
Decide whether the business intends to own a CRE workflow or merely assist one. A product that remains a research assistant can tolerate conventional SaaS pricing for longer. A system designed to screen, extract, monitor and act autonomously should build its commercial model for that destination now, rather than waiting until automation has already reduced the seat base.
Make portfolio expansion the revenue-growth mechanism. The strongest enterprise contract should let ARR rise because the customer trusts the system with more properties, markets and workflows, not because employees generate more prompts.
Treat inference optimisation as a margin programme, not a discount programme. Finance and engineering should track cost per property and per workflow internally while preserving a customer-facing price tied to CRE value. OpenAI’s 2023-2026 price history shows why exposing raw model economics to buyers creates an unstable anchor. -
Build the commercial system for the autonomy level the product is approaching, not the level it has today. When more screening, lease analysis or asset monitoring moves from human production to machine execution, the business should gain operating leverage rather than discover that its pricing model loses revenue every time the product improves.
Assumptions
AMS ratings and decision-matrix scores are Monetizely judgements based on public product and pricing information available on 21 August 2026, not ratings published by the vendors. The three-year sensitivity model is a pricing model, not a forecast or market benchmark: it assumes $360,000 of Year 1 ARR, 30 paid users, 60,000 consumption units, 12 closed deals or 600 active properties, then applies the Year 3 conditions shown in the table. For this analysis, an active property is a property being screened, underwritten, managed or reported through the system during the agreed billing period; lease or document bands can supplement that primary meter where workload varies sharply by property.
Footnotes
Monetizing Agentic AI. https://www.amazon.com/Monetizing-Agentic-AI-Handbook-Transformation/dp/B0H7Z13VKJ/
Intercom, official pricing and Fin outcome pricing, accessed 21 August 2026. https://www.intercom.com/pricing
Salesforce, official Agentforce pricing, accessed 21 August 2026. https://www.salesforce.com/in/agentforce/pricing/
Microsoft, Microsoft 365 Copilot enterprise pricing, accessed 21 August 2026. https://www.microsoft.com/en-us/microsoft-365/copilot/pricing/enterprise
11x, Alice official pricing, accessed 21 August 2026. https://www.11x.ai/products/alice/pricing
Reonomy, official pricing, accessed 21 August 2026. https://www.reonomy.com/pricing/
Prophia, official plans and pricing, accessed 21 August 2026. https://www.prophia.com/pricing
Dealpath, official plans and pricing information, accessed 21 August 2026. https://www.dealpath.com/plans/
Dealpath, official Dealpath AI product page, accessed 21 August 2026. https://www.dealpath.com/ai/
CoStar Group, Form 10-Q for the quarter ended 30 June 2026, U.S. Securities and Exchange Commission. https://www.sec.gov/Archives/edgar/data/1057352/000105735226000066/csgp-20260630.htm
Steven Devaney and Roberto Martinez Diaz, “Transaction based indices for the UK commercial real estate market: an exploration using IPD transaction data,” Journal of Property Research, published online 31 August 2011. https://www.tandfonline.com/doi/full/10.1080/09599916.2011.601317
Richard Barkham and David Geltner, “Price Discovery in American and British Property Markets,” Real Estate Economics, March 1995. https://onlinelibrary.wiley.com/doi/abs/10.1111/1540-6229.00656
Garrison Hongyu Song and Abeba Mussa, “Commercial Real Estate Market with Two-sided Random Search: Theory and Implications,” Journal of Property Research, Volume 38, 2021; published online 23 November 2020. https://www.tandfonline.com/doi/full/10.1080/09599916.2020.1844784
OpenAI, GPT-4 research and launch pricing, March 2023. https://openai.com/index/gpt-4-research/
OpenAI, “Introducing GPT-4.1 in the API,” April 2025. https://openai.com/index/gpt-4-1/
OpenAI, API Platform pricing, accessed 21 August 2026. https://openai.com/api/