An AI pilot can look successful for all the wrong reasons. It runs on curated data, connects to a handful of systems, and operates inside a controlled workflow. Then comes the decision to scale it across the business, and that is where the story often changes.
Deloitte’s 2026 commercial real estate outlook, based on a survey of more than 850 C-level executives at real estate owner and investor organizations, found that 27% of respondents were experiencing challenges with AI implementation, including technical issues, lack of expertise, or resistance to change.
Pilots stall for a specific reason. At scale, they meet production data, multiple systems and real operational stakes that a sandbox never had to handle. Scaling AI is, in large part, an infrastructure decision. Before committing to one, it helps to know which questions to ask.
8 Infrastructure Questions Leaders Must Ask
These questions don’t require deep technical expertise, but the answers usually determine how smoothly a scale-up goes. Most of them surface issues in data, systems, or governance, the same territory a controlled pilot rarely tests.
| # | Question | Why it matters |
| 1 | Is the required data available? | AI needs relevant, sufficiently detailed data. If it cannot be extracted, the initiative may need a different starting point. |
| 2 | Can systems exchange this data? | Data trapped in disconnected systems limits what an AI application can access. |
| 3 | Is data quality sufficient? | Inconsistent, outdated, or duplicate records can undermine AI outputs. |
| 4 | Can the existing architecture scale? | A successful pilot may behave differently at production volume. |
| 5 | What security and access requirements exist? | Real estate systems can contain financial, tenant, ownership, lease, and operational information that may be subject to privacy, contractual, or industry-specific requirements. |
| 6 | What human oversight is required? | Higher-stakes decisions, such as underwriting or tenant screening, usually need a defined point of human review. |
| 7 | Who owns the data and the AI outcome? | Someone needs to be accountable for source data, access approval, integration maintenance, and what happens when an output is wrong. |
| 8 | How will ROI be measured? | A clear metric, tied to the original workflow, makes it possible to determine whether the initiative delivered value. |
Working through this list before committing budget does not guarantee a smooth rollout, but it tends to surface the biggest risks while they are still cheap to address. Three areas come up more often than the others: data fragmentation, legacy system constraints, and unclear ownership of the workflow being automated.
Why These Questions Expose Data Fragmentation
Questions 1 through 3, in particular, tend to surface how fragmented real estate data actually is. From the outside, real estate data can look like a single asset: property records, tenant information, transaction history. In practice, it is spread across many independent systems, each with its own structure and update cycle.
A typical portfolio might draw on:
- MLS and listing platforms
- CRM platforms
- Property management systems (PMS)
- ERP and financial systems
- Public property records
- Third-party data providers and APIs
- Internal spreadsheets and legacy databases
Consider something as basic as a property address. The same property might appear as “123 Main St” in one system, “123 Main Street” in another, an internal property ID in a third, and an MLS or PMS identifier in a fourth. Unless these are reconciled, an AI application working across systems may treat one property as several, or miss connections a human reviewer would catch immediately.
A useful way to think about the underlying work is as a pipeline:
- Ingestion: can data be brought together reliably?
- Standardization: can systems use consistent structures and definitions?
- Entity resolution: can records referring to the same property, tenant, or transaction be matched?
- Data management: can the data be stored, updated, and governed for operational and AI use?
- Secure access: can authorized users and applications retrieve the right data?
Underinvesting in any one stage tends to surface later as inaccurate AI output, or as reports that different teams interpret differently. Beyond model accuracy, this can slow reporting cycles, create conflicting KPIs across departments, and add back the manual reconciliation work AI was meant to remove.
This is not only a real estate problem. Research from McKinsey found that only 7% of surveyed companies have reached the level of data readiness required to scale advanced AI, even though 64% report they have already moved beyond pilots into production for at least some functions. Real estate’s system landscape, spread across MLS platforms, property management tools, ERP systems and third-party data providers, tends to make that gap harder to close than in many other sectors.
This is exactly the layer ORIL helps PropTech and real estate technology companies get right with AI enablement for PropTech, so the model becomes the easy part of the project instead of the reason it stalls.
Legacy Systems and Integration Constraints
Questions 2 and 4 usually point to a second constraint: legacy systems.
Many real estate organizations operate across a mix of modern and legacy platforms. Some of these legacy systems handle critical functions such as lease accounting, property management or transaction processing, and cannot simply be replaced overnight.
Legacy platforms become a bottleneck when they cannot expose the required data reliably, securely, or at the speed a workflow requires. A full replacement can address this, but replacements carry real business continuity risk, along with significant migration cost and complexity.
In practice, many organizations modernize incrementally rather than through a single large migration. This can involve:
- Building API layers on top of legacy systems to expose their data without replacing the underlying platform.
- Introducing an integration layer that connects systems that were never designed to work together.
- Moving toward modular architecture, where components can be upgraded or replaced independently over time.
- Sequencing gradual modernization around business priorities, rather than treating it as a single technology project.
This is rarely a choice between keeping the old system or replacing everything. It is about creating a path where legacy systems can still participate in an AI-ready environment while the organization modernizes at a pace it can manage.
Start With the Workflow and a Measurable Outcome
Questions 5 through 8 are less about data and more about how the initiative is framed in the first place. The strongest AI initiatives usually begin with a business problem, not a model.
The central question is simple: what decision or workflow are we trying to improve? In real estate, that could mean:
- Property operations and facilities management
- Maintenance scheduling and response times
- Leasing and tenant communication
- Underwriting and investment analysis
- Customer support and resident services
- Portfolio-level data analysis and reporting
Framing the initiative around a specific workflow does two things. It gives the technical team a clear target for what “good” looks like, and it gives leadership a way to measure whether the investment paid off. AI applied to a well-defined, measurable business problem is easier to justify, easier to scope, and easier to improve over time than AI introduced as a general capability.
Sequencing the AI Investment Decision
None of this argues for waiting. It argues for sequencing: knowing what the eight questions above will surface before committing budget to the model everyone wants to talk about first.
For real estate and PropTech leaders, that reframes the investment decision. The question is not only which AI use case to pursue, but what the technology environment needs to look like to support it, this year and over the next several years.
In practice, that means treating data infrastructure and system integration as core product development services, not something to backfill once the AI model is already live. It’s the approach ORIL builds into its real estate and PropTech engagements from the outset, because initiatives that treat the foundation as an afterthought rarely make it past the pilot stage.
