Quick Answer
AI in commercial real estate is not inherently expensive. It gets expensive when it runs without governance. Most runaway bills trace back to four operational patterns, all of which are fixable without reducing how much AI the team uses. This article explains how the pricing works, what those patterns are, and what practical controls bring costs back in line.
One of the more common conversations happening inside CRE firms right now starts something like this: everything was working, the team was getting value from the AI tools, and then a bill showed up that nobody expected.
“ A firm connected AI straight to their systems, their CRM tool, SharePoint, their Microsoft Teams, Outlook, and it worked really well for them. They were able to get access to all this information that was in disparate systems, all surfaced in their AI tool. Great efficiencies were gained, better insights. All was good until the bill showed up. About forty thousand dollars to be exact. That's a true story and it's our world and it is very, very common these days. ”
Wes Snow, CEO and Co-Founder of Ascendix
The pricing model behind most AI tools runs on tokens, small units of text the model reads and generates, and most organizations don’t look closely at that meter until something goes wrong. Seventy-three percent of enterprises say their AI costs have exceeded original projections (The FinOps Foundation’s 2026 State of FinOps report).
In commercial real estate, where AI is increasingly connected to CRMs, email, listing data, and document workflows, the exposure is higher than in most industries because the workflows tend to involve high volumes and live data.
How AI Pricing Works
AI tools charge per token. A token is roughly three to four characters of text. When someone sends a message, the model reads the input and generates a response. Reading the input costs less. Generating the response costs more.
Analysis of 2.4 billion enterprise API calls shows that the per-token price has been falling. Between Q1 2025 and Q1 2026, blended AI costs dropped 67%, from $18.40 to $6.07 per million tokens. But enterprise AI bills have been rising, because usage is scaling faster than the price is falling, and the types of tasks organizations are running AI on have changed.
A simple question costs very little. An automated workflow, where an AI agent reads a dataset, reasons over it, generates output, and loops through hundreds of records, uses anywhere from 5 to 30 times more tokens per task than a conversational query. That multiplier applied to a data-heavy CRE operation at volume adds up quickly, often without anyone tracking it.
Four Patterns That Drive Unexpected Costs in CRE
Most runaway AI bills in commercial real estate trace back to one of four situations or their combination.
Automated Jobs with No Scope Limit
An AI agent connected to a live data source without a ceiling on how much it processes scales cost with data volume. This is common in lead enrichment, email processing, and listing management. An agent that handles twenty inbound records one week may handle two hundred the next, depending on market activity, with no change in your setup and no alert that usage has increased.
The further risk is that these agents typically run in the background. By the time the volume spike is visible, it shows up on the invoice rather than in a dashboard. In commercial real estate, listing activity can fluctuate sharply. So for a realistic exposure, agents are often set to run overnight or on schedules.
“ In the commercial real estate world, maybe it's like an inbound listing. And that listing could go viral. There could be hundreds of responses. And AI enriched and researched all of those inbound listings, and suddenly the bill's much higher than what you'd suspect. ”
Todd Terry, CTO and Co-Founder of Ascendix
The practical fix: Add an explicit record cap to any recurring agent job. Telling an agent to process a maximum of 50 records per run and stop converts an open-ended cost into a predictable line item. The agent may not finish all available work in one run, but the exposure is controlled.
Defaulting to the Most Capable Model for Everything
AI models come in tiers. High-capability models cost several times more per token than mid-tier alternatives. The instinct is to default to the most powerful model available, because it produces the best output and that feels like the prudent choice. In practice, the majority of routine CRE tasks (drafting emails, parsing standard documents, summarizing a meeting) are well within the capability of mid-tier models. And the cost difference between tiers, applied across a full team’s workload, compounds significantly.
“ We don't want to compromise the quality of what AI is doing on our behalf. And so if you don't know what certain models can do and what their appropriate usage is, then most will just default to the best and damn the consequences on the back end, and that ends up manifesting in a really large bill. We're all going to be having to think more prudently about when we use AI and how do we use it responsibly. ”
Wes Snow, CEO and Co-Founder of Ascendix
Most platforms allow default models to be set at the team member or task type level. Pinning different roles to different tiers removes the decision from individual users, prevents the default-to-top pattern, and has a predictable, ongoing effect on spend.
Context That Grows Without Being Reset
Most people think of an AI query as a direct exchange: you type a question, you get a response. The actual transaction includes more than just the question. Every message the model receives includes the full conversation history, a description of every tool it has access to, and any background instructions it has been given. The longer a session runs and the more tools are connected, the more tokens each new message costs, regardless of how simple the question is.
“ It's not just your question that gets sent to the model. It's everything in the context, every MCP tool that's been connected, a description of every skill... If you're the type of user who's connected everything and sees everything as a way to use AI, then there's a lot of baggage that's getting sent in your request. ”
Todd Terry, CTO and Co-Founder of Ascendix
The same dynamic affects agents running on automated schedules. As they process work, the context fills with intermediate data, outputs, and records from previous steps. Each subsequent action costs more than the last, because the model is sending and receiving a larger context window with every turn.
Tip: If a session starts producing vague or less precise responses than it did earlier in the same conversation, the context is likely filling up. Starting a new session at that point is faster than trying to redirect the existing one.
Rework From Unclear Initial Prompts
Every exchange with an AI model is billed. An unclear prompt produces an incomplete or misaligned response, which leads to a follow-up message, which produces another response, and so on. In a team of twenty or fifty people, those extra round trips accumulate into a substantial share of overall spend, largely invisibly because no individual conversation looks expensive.
A UT Austin and KPMG research project analyzed 1.4 million workplace AI interactions over eight months. The finding: 90% of users were actively using AI, but only about 5% were using it in a way that consistently produced high-quality outputs. The difference was how specifically people briefed the model at the start, and how willing they were to correct an inadequate first response rather than accepting it.
“ That's not a technology problem, that's a technique problem. And the thing is, every one of those wasted round trips was metered. A broker that takes the minute to brief it properly, gets two passes. The ones who don't, take eight. Same question, four times the bill, and probably getting a worse answer at the end of it. ”
Todd Terry, CTO and Co-Founder of Ascendix
Brokers don’t need to learn prompt engineering. Giving the model the right context at the start (what the task is, what constraints apply, what a useful output looks like) and being willing to say “that’s not right, adjust it this way” produces better results in fewer exchanges. The research found that the highest-performing users wrote more casually, not more technically. They also were more specific.
Another term worth noting for CRE leadership is a shadow spend when individuals paying for AI subscriptions out of pocket and expensing them. And it is almost always higher than leadership estimates and sits outside any governance structure. Then when the firm transfers all users to one company-billed LLM subscription, the cost comes unexpectedly.
Five Ways to Manage AI Costs Without Reducing Usage
None of the following require technical expertise or significant changes to how a team works.
Match the Model to the Task
Mid-tier models handle routine CRE work (drafting, parsing, summarizing) at a fraction of the cost of top-tier models. Most platforms allow default models to be set by team member or task type. This removes the model selection decision from individual users, prevents the default-to-top pattern, and has a consistent, ongoing effect on spend. Reserve high-end models for tasks where the quality difference is noticeable.
The same Q1 2026 analysis of 2.4 billion enterprise API calls found that organizations that use tiered model architectures, matching different model tiers to different task types, achieve a median cost of $2.31 per million tokens, compared to $18.40 for single-model deployments. Getting governance in place before scaling is consistently more cost-effective than retrofitting it after a surprise bill.
Ask the Model to Delegate Subtasks
When working with a high-capability model on a complex task, you can ask it to hand off portions of the work to less expensive sub-models. No configuration is required. Adding a phrase like “keep tokenomics in mind and delegate where it makes sense” to a prompt is enough.
“ Just ask your favorite model, “Is there any of this you can delegate?”The agent has been built to understand how to delegate and when to delegate. When the model's doing the delegation, that gas meter [meaning usage] is moving more slowly. ”
Todd Terry, CTO and Co-Founder of Ascendix
Set Spend Alerts Rather Than Hard Cutoffs
Most platforms allow firm-level spend limits. A hard cap that cuts off access mid-task creates disruption and doesn’t produce useful information about what drove the spike. An alert that flags unusual usage and prompts a conversation is more practical. The goal is visibility, not restriction.
“ Nobody caps the electricity on the building, but you get an alert, you get to find out who left the lights on, and you have a five minute conversation about it ”
Todd Terry, CTO and Co-Founder of Ascendix
Measure One User Before Rolling Out to Fifty
Before expanding AI access across a brokerage, run a two-to-three week pilot with two or three team members, with usage monitoring enabled. This gives a real per-user cost figure based on your actual workflows before you scale. During that period, also survey how many people at the firm are already paying for AI tools out of pocket to understand the shadow spend.
Cap Recurring Agent Jobs and Reset Context Regularly
Any agent task running on a schedule should have an explicit limit on how many records it processes per run. This is a single instruction in the agent’s setup and prevents most runaway billing from unattended automation. On the context side, start a new conversation when you switch topics. The simplest version of context management is just opening a new chat window.
Ready to Put the Right AI Governance in Place for Your Brokerage?
Ascendix helps CRE firms build practical AI operating models, from model selection and spend controls to agent architecture.
Why Rreducing AI Usage is Not the Answer to Cost Control
The figure that 73% of enterprises have exceeded their AI budget projections tends to show up in arguments for cutting back on usage. But the case for pulling back carries its own costs.
“ Everything we've talked about today is really about not wasting money. But honestly, the most expensive position in this whole thing is the firm that got spooked, capped everything, and just stopped. Because the people who aren't using it, they're burning money too. They just never get an invoice for it.
If you tried something last year and it didn't work, you should go try it again. These tools have moved twice since then. And I bet half or more of the AI can't-do-that list at most brokerages is out of date at this point. ”
Todd Terry, CTO and Co-Founder of Ascendix
A JLL survey found that 92% of CRE companies are now running AI pilots, up from 5% three years ago. Only 5% report achieving most of their program goals. The gap between adoption and results is largely a governance question. The tools have matured substantially in the past two years. Workflows that didn’t deliver usable results in 2024 often do now.
Firms seeing a return on AI are not necessarily the ones spending the most. They’re the ones who know what each workflow costs, match the model to the task, and have basic controls in place so that nothing runs unchecked. Operational discipline is available to any size firm.
How Ascendix Helps CRE Firms Get the AI ROI Right
Ascendix has been building software for commercial real estate for nearly 30 years. Our AI advisory work is grounded in direct operational experience, not just consulting frameworks.
Because Ascendix builds CRE software and uses AI across its own operations, we have a direct view of what specific CRE workflows cost to run on AI. Those experiences shaped how Ascendix approaches AI governance with clients, not as a theoretical exercise, but as something that requires the same attention as any other operational control.
- AI Consulting: We audit your current AI setup, map the cost drivers across your team and workflows, and put a governance structure in place: model selection policies, spend alerts, agent scope controls, and a pilot framework.
- AI Development Services: We build AI into specific CRE workflows rather than layering generic tools on top of what already exists.
- Custom CRE Software Development: When the off-the-shelf options don’t match the workflow, we build from scratch. We know CRE well enough to skip the domain briefing, which cuts a significant amount of time off the delivery cycle.
- CRE Data and Integration Services: We connect CRE platforms, databases and LLMs to seamlessly share information and interact with each other.
Talk to an Ascendix AI Specialist
Whether you’re just getting started with AI in your brokerage or managing unexpected costs from a current setup, a free consultation gives you a clear picture of where you stand and what to do next.
What is tokenomics in AI, and why does it matter for CRE firms?
Tokenomics refers to how AI usage is priced and measured. AI tools charge by the token, a small unit of text. Reading input costs less; generating output costs more. For CRE firms connecting AI to CRMs, email, listing data, and document workflows, tokenomics determines whether AI is a controlled operating expense or a recurring billing surprise.
What is the ROI of AI in commercial real estate?
The ROI shows up in reduced time on data entry, faster deal research, quicker document turnaround, and more consistent client communication. But the financial return depends on how deliberately AI is deployed. Firms that connect AI to workflows without governance structures often find that costs outpace savings. The ones seeing real returns have matched tools to tasks, put basic controls in place, and used a pilot period to establish per-user cost baselines before rolling out broadly.
How Much does AI cost for a commercial real estate firm?
Per-token prices have been falling, from around $18.40 to $6.07 per million tokens between Q1 2025 and Q1 2026. But agentic workflows that process data automatically use 5 to 30 times more tokens per task than a simple conversational query, so total bills can still grow quickly.
Kateryna Mamchych
Kateryna is an experienced writer focused on topics related to Salesforce consulting and optimizing operations within the real estate industry via CRM implementation. Through her insightful articles, Kateryna aims to help organizations make the most of their Salesforce adoption.
Loading...