KEY TAKEAWAYS
- While 92% of CRE firms are piloting AI, only 5% say they have achieved all of their AI goals, leaving most stuck in “pilot purgatory.”
- Stalled pilots are expensive. A 12-month delay costs a $1.5B multifamily operator roughly $1.8M in foregone NOI.
- The five failure modes are pilot sprawl, starting with tools instead of workflows, waiting for perfect data, lacking human review, and having no executive owner or baseline.
- Firms that escape purgatory map workflows first, run AI on top of existing systems, require human approval, and track results against a baseline.
IN THIS ARTICLE
Everyone is piloting. Almost no one is finishing.
In July 2023, fewer than 5% of real estate companies had started an AI pilot. Two years later, JLL’s 2025 Global Real Estate Technology Survey of more than 1,500 senior CRE decision-makers found that 92% of occupiers were running AI pilots, and 88% of investors, owners, and landlords had started too.
That is the fastest technology shift this industry has ever seen. But here is the number that should worry every CEO: only 5% of firms say they have achieved all of their AI goals. More than 60% of investors say they are not prepared to scale AI beyond the pilot stage.
We call this pilot purgatory. It’s the place where a firm has bought the licenses, run the demos, and asked a few enthusiastic employees to “try it out,” but nothing has changed in the P&L. The leasing team still answers inquiries the next morning. Underwriting still takes a week. The monthly investor report still eats three days of the asset manager’s time.
Pilot purgatory feels safe, because the firm can say it is “doing AI.” It isn’t safe. It costs real money every month, and the firms that escape it are building an advantage the rest will struggle to close.
This article covers why the gap matters now, the five failure modes that keep CRE pilots stuck, and what the 5% do differently.
Why a stalled pilot is expensive right now
A few years ago, a slow AI rollout was a missed opportunity. Today it’s a cost line, because the pressure on CRE operating margins is coming from three directions at once.
- Expenses are outrunning income. Since 2021, multifamily NOI has grown about 10%, while administrative and payroll costs are up about 20%, repairs and maintenance about 28%, and insurance about 55%. NAREIM data shows the median real estate investment manager already spends 42% of revenue on salaries and benefits.
- Lenders are reading your operations. Multifamily CMBS delinquency reached 7.15%, up from 1.84% two years earlier, according to Trepp. When loans come up for refinance, lenders size them off trailing operating statements. A bloated expense line now shows up directly in your loan proceeds.
- LPs can compare you. The ILPA Reporting Template 2.0 makes fees and expenses easier to compare across managers. Capital is concentrating, too: 46% of 2025 private real estate capital went to the 10 largest funds.
Against that backdrop, the cost of waiting is easy to underestimate. Our modeling of a $1.5B multifamily operator puts the cost of a 12-month delay at roughly $1.8M in foregone NOI. That’s not a technology budget question. It’s an asset management question.
The firms in the 5% aren’t spending more on AI than the firms in purgatory. They’re getting it into production, where it actually changes the numbers.
The five failure modes that keep CRE pilots stuck
After working with CRE sponsors on AI rollouts, we see the same five patterns again and again. Most stalled firms have at least two of them.
1. Pilot sprawl: five experiments, zero finished
JLL found that investors who are piloting AI run an average of five use cases at the same time. That sounds ambitious. In practice, it spreads attention so thin that nothing reaches production. A leasing chatbot, a lease abstraction tool, a market research assistant, a handful of standalone AI chat seats, and an underwriting experiment all compete for the same three people’s time.
The fix: Pick two or three high-volume workflows and take them all the way to production before starting anything new.
2. Tools instead of workflows
Many pilots start with a tool (“let’s try this AI app”) instead of a process (“our rent roll review takes 11 steps and two days”). The result is individual productivity at best. One analyst saves an hour, but the workflow, the handoffs, and the waiting stay exactly the same.
The fix: Map the workflow first: every step, every handoff, and every wait. Then decide where AI removes steps entirely, not just where it speeds up one person.
3. The data-lake trap
Some firms decide they can’t use AI until all their data is clean and centralized. So they launch a data warehouse project that runs 18 months and stalls. JLL found that more than 60% of firms have to fix basic technology issues, like duplicated or dormant systems, before they can use AI fully.
The fix: Run AI workflows on top of the systems you already have, like your property management system, CRM, and accounting platform. Clean up the data that each workflow actually touches, as you go.
4. No human review and no audit trail
A pilot can run on trust. Production can’t. The moment an AI output affects a lease, a loan covenant, or an investor report, your COO, your controller, and your compliance team will ask who approved it and how they can check it. If the pilot has no review step and no record, it never gets past that conversation.
The fix: Build human-in-the-loop approval into every workflow from day one, with a full audit trail of what the AI did and who approved it.
5. No owner, no baseline, and no team buy-in
The most common failure is organizational. The pilot belongs to one enthusiastic employee or to IT, not to an executive with a target. Nobody measured how long the work took before, so nobody can prove it got faster. And the team quietly resists, because they suspect the project is about replacing them.
The fix: Give each workflow an executive owner, measure the baseline in hours and dollars before you start, and train the team alongside the build. Tell them plainly that the goal is to remove the work they hate, and to grow through attrition and new acquisitions, not layoffs.
What the 5% do differently
Firms that escape pilot purgatory don’t have better AI models than everyone else. Everyone has access to the same models. What they have is a better operating plan. In our experience, it looks like this:
| Pilot purgatory | Production |
|---|---|
| Five tools tested at once | Two or three workflows, finished |
| Starts with a tool | Starts with a mapped workflow |
| Waits for clean data | Runs on top of existing systems |
| AI output goes straight out, or nowhere | Human approves, rejects, or modifies every output |
| No record of what happened | Full audit trail |
| “It feels faster” | Monthly scorecard vs. baseline, in hours and dollars |
| One champion, team on the sidelines | Executive owner, whole team trained |
| Timeline: “someday” | First workflows live in 30–45 days |
Camden Property Trust is a good public example of what production looks like. Camden centralized more than 90% of resident communications, moved onsite teams from a seven-day to a six-day week, and reports $4–5M a year in labor savings. That isn’t a pilot. It’s a different operating model.
You don’t have to be a REIT to get there. Middle-market sponsors often move faster, because the CEO can make the call and the workflows are easier to map.
How Sophia members get out of purgatory
At Sophia CRE, we built our approach around these five failure modes. We call it Learn → Automate → Scale.
- Learn. Your team goes through AI Workforce Training: six live, 90-minute sessions built around your own workflows. They come out with the skills, and with written AI SOPs, so the knowledge stays in the firm.
- Automate. While your team trains, we build your custom AI workflows on top of the systems you already use. Every output goes to a person in Microsoft Teams or Slack to approve, reject, or modify. Our proprietary AI Synchronization Model™ has multiple AI models cross-check each other, so the information is as strong and trustworthy as possible. Your first production workflows go live within 30–45 days of kickoff.
- Scale. The Sophia platform and Mission Control give you a live view of every workflow, a full audit trail, and a monthly scorecard against your starting baseline, in hours and dollars. Our team keeps tuning and extending your workflows as you grow.
The result is simple to state: your team learns AI, Sophia automates your slowest work, and you scale without adding headcount.
Getting out of purgatory starts with two workflows
The gap between 92% and 5% isn’t about technology. The models are good enough, and they’re cheap. The gap is about focus, measurement, review, and ownership. Those are leadership decisions, and every CEO can make them this quarter.
If your firm is still in pilot purgatory, you’re in good company. But the window where “we’re experimenting” is an acceptable answer is closing, for your lenders, your LPs, and your competitors.
Frequently asked questions
What is AI pilot purgatory?
It’s when a firm runs AI experiments that never reach daily production, so the firm pays for tools and staff time without changing its operating results.
Why do most CRE AI pilots fail?
The most common reasons are running too many pilots at once, starting with tools instead of workflows, waiting for perfect data, having no human review or audit trail, and having no executive owner or baseline.
How long should it take to get an AI workflow into production?
With a focused scope and workflows built on top of existing systems, the first production workflows can be live within 30–45 days.
Do we need to replace our property management system or CRM to use AI?
No. The fastest path is to run AI workflows on top of the systems you already use.
Will AI mean laying off our team?
It doesn’t have to. Many firms reduce headcount needs through normal attrition and use the freed capacity to take on more properties without adding staff. Read how firms do it in AI Without Layoffs.
Ready to move a workflow from pilot to production?
Book a demo with Sophia CRE, and we’ll walk through which of your workflows are the best candidates to go live in the next 30–45 days.
Learn → Automate → Scale. Your team learns AI. Sophia automates the rest.
