Data architecture plays a central role in whether enterprise AI moves beyond the pilot stage. Buying an AI tool is often easier than preparing the systems it needs to work with.
A pilot can run on a clean dataset assembled for a demonstration. A production application may need to reconcile records from finance, operations, and customer-facing platforms while respecting access rules and existing workflows.
In real estate, property, lease, financial, and building data often sit in systems built for different purposes. The key questions for leadership are which decision an AI application should improve and whether its data can reach it reliably.
The Readiness Gap in Numbers
Gartner’s February 2025 research shows the scale of the readiness gap. In a survey of 1,203 data management leaders, 63% of organisations said they either lacked the right data management practices for AI or were unsure whether they had them. For property businesses, putting those practices in place often depends on real estate data integration, since quality checks and access controls are hard to apply consistently to disconnected records.
In the same announcement, Gartner forecast that organisations would abandon 60% of AI projects unsupported by AI-ready data through 2026. The figure is a forecast and applies only to that group of projects.
JLL’s 2025 Global Real Estate Technology Survey points to a similar gap in real estate. Among investors, owners, and landlords, 88% had started piloting AI, pursuing an average of five use cases at once. In a separate group, only 5% of corporate real estate occupiers reported achieving all their AI programme goals. The two figures come from different respondent groups and describe separate measures. Read side by side, they illustrate the distance between widespread experimentation and fully achieved programme goals.
Why AI Initiatives Often Begin With the Model
An AI product is often easier to evaluate than the data work around it. Teams can test a model on a prepared dataset before they know how it performs with live records, exceptions, and existing processes.
Three factors tend to reinforce this pattern:
- Demonstrations may hide integration work. A curated dataset shows what a model can do. It does not test how the application handles missing fields, conflicting records, or changing source data.
- Accountability is split across teams. The AI product may have one budget owner, while data access, definitions, and quality sit with several departments.
- Delivery pressure usually favours visible progress. A working demo is easier to present than the ongoing work of maintaining pipelines, permissions, and monitoring.
Once a pilot meets live data, deferred data questions can become a significant source of rework, alongside model reliability, security, and adoption.
What Fragmented Real Estate Data Looks Like
A typical real estate business draws on several systems, each designed for a different operational task:
| Source | What it holds | Common data challenge |
| MLS and listing feeds | Listings, prices, property attributes | Field names, formats, units, and definitions can differ by provider |
| CRM | Leads, contacts, deal history | Duplicate contacts and incomplete activity records |
| Property management system (PMS) | Leases, tenants, rent rolls, work orders | Access and export options depend on the platform, configuration, and contract |
| Public records | Ownership, tax, permits, transactions | Recording delays, differing address formats, and inconsistent identifiers |
| Building systems and IoT | Energy use, occupancy, equipment status | Readings may need timestamp alignment and mapping to asset identifiers |
| Third-party APIs | Market data, demographics, valuations | Refresh schedules, permitted uses, and redistribution rights vary |
The challenge emerges when a use case needs to reconcile records across these systems. The same property may carry different identifiers or address formats, and unit counts can differ because platforms model spaces differently. Records also update on different schedules.
Resolving this involves two separate tasks. Standardisation aligns formats and field definitions. Entity resolution determines which records refer to the same real-world asset, unit, or lease.
A practical first step is to map how properties, units, leases, and tenants are represented across source systems. This can reveal mismatched identifiers, missing relationships, and access constraints before development begins.
JLL’s survey analysis notes that fragmented and inconsistent data has historically limited portfolio-level insight in real estate. For AI, this can mean models working with a partial view of a portfolio.
The Capabilities Needed for AI-Ready Data
Connected data may be unsuitable for a particular decision, so AI readiness needs to be assessed against each intended use.
Gartner describes AI-ready data as representative of its use case, including the patterns, errors, and outliers a model needs to handle. A valuation model and a lease-processing tool, for example, need different inputs and validation criteria.
Making data usable for a specific decision, often described as real estate data enablement, relies on five capabilities:
- Data integration: connects source systems and maps how their records relate.
- APIs and access: provide approved interfaces to data, within each platform’s permissions, rate limits, and update schedules.
- Governance: sets ownership, access rights, permitted use, and retention for each dataset.
- Data quality: tracks the dimensions that matter for the use case, such as completeness, validity, and freshness.
- Scalable architecture: lets new sources and use cases be added with manageable changes.
Five Real Estate Use Cases and Their Data Dependencies
| AI use case | Data it depends on | What may go wrong when data is fragmented |
| Pricing and valuation support | Listings, transactions, property attributes, market data | Incorrect property matches can introduce unsuitable comparables |
| Predictive maintenance | Sensor data, work orders, equipment records | Unmapped sensor IDs make alerts harder to interpret against maintenance history |
| Lead scoring and tenant matching | CRM activity, listings, enquiry history | Duplicate or incomplete records can affect scores and hide enquiry history |
| Portfolio performance analytics | Rent rolls, leases, operating costs, occupancy | Different definitions or update schedules produce discrepancies that need reconciliation |
| Lease document processing | Lease files, PMS records, accounting data | Extracted terms may conflict with existing records and need validation |
The same work also helps non-AI operations: consistent definitions of occupancy, rental income, and costs reduce discrepancies between portfolio and finance reports.
What Changes Between Pilot and Production
Moving to production changes how the system is used and supported:
- Live data: the application runs on operational sources and needs a plan for delayed, missing, or failed feeds.
- Wider use: more people rely on the outputs, so access and support need to match their roles.
- Error handling: acceptance thresholds and escalation paths decide when an output goes to human review.
- Ongoing monitoring: model performance and business outcomes are tracked after the test period ends.
Legacy systems
Established property management and accounting platforms may restrict data access through limited APIs or vendor terms. In production, the integration also has to respect source-system load, latency, and security.
Operational workflows
A maintenance prediction creates more value when it reaches the team that acts on it, through a work order, alert, or dashboard. A pricing recommendation may need asset manager review before it informs a listing.
A Six-Step Roadmap for AI Adoption
- Define the business problem. Choose a process where improvement matters, such as lease processing time. Set a baseline, a target, and a measurement method before choosing a solution.
- Map the data dependencies. Identify which systems hold the required data, who owns it, and which source is authoritative for each critical value. Record the gaps.
- Connect the required sources. Agree on an integration method for each system and test each connection before a workflow depends on it.
- Resolve definitions and identities. Align field definitions and decide how records for the same property, unit, lease, or tenant are matched. Document how conflicting values are resolved.
- Validate against operational reality. Compare outputs with known outcomes and expert assessments, including how the system behaves with missing or stale data.
- Scale incrementally. Expand once the first use case meets agreed criteria, and review what each new use case needs before reusing existing components.
The Investment Decision: Start With the Use Case
For business leaders, the practical decision concerns scope. Each AI use case needs a specific set of capabilities, and its readiness gaps can be identified before significant spending begins.
A targeted approach funds the foundations one business outcome needs, tests the result under realistic conditions, and expands when the evidence supports it.
In real estate, that might mean starting with one portfolio question, such as lease exposure, and the few systems that hold its data. Once an integration has been tested in production, the next project can assess whether it meets its needs or requires additional data, access, or validation work.
Building that kind of foundation is part of how ORIL develops real estate and PropTech software, from core platforms to AI enablement for PropTech.
