Planning an AI workflow? Follow the data it needs.

Every week, we talk to real estate operators who want to "do something with AI." They've seen the demos. They've sat through the vendor pitches. They've watched a competitor announce an AI partnership on LinkedIn. And they come to us with the same question: Where do we start?

Start with a task worth improving. Then trace the information it depends on.

A first experiment might need only a few documents and a clear set of instructions. A workflow that draws on your property system, CRM, and underwriting models needs a closer look at access, definitions, and data quality. The investigation should fit the work you want to do.

Real estate firms often work across several systems. An analyst may combine an accounting export, an Excel model, and notes from a broker before making a recommendation. Understanding those steps helps you decide where AI could be useful and what else needs to change.

The Systems of Record That Matter

For most real estate firms, the core systems of record fall into a few categories. For the workflow you are considering, identify which system holds each input and who is responsible for it.

Property Management & Accounting Systems Yardi, RealPage, MRI Software, AppFolio. these are the operational backbone for most firms. They hold lease data, rent rolls, tenant information, maintenance records, GL entries, and budget actuals. They also hold years of historical data that most firms barely touch beyond standard reporting.

The question to ask: What data is in our PMS that we currently extract manually for analysis, reporting, or decision-making? A recurring export is worth investigating. Check whether the work needs AI, a standard integration, or a simpler reporting change.

CRM & Deal Management Salesforce, HubSpot, Juniper Square, Dealpath. wherever you track leads, investor relationships, or deal flow. These systems contain the narrative of your business development process: who you talked to, what they said, what stage a deal is in, and what fell through.

The question to ask: How much of our deal screening or investor communication is manually assembled from information that already exists in our CRM? If your team spends hours pulling together deal summaries or investor updates from data that's already in the system, trace those steps before choosing a tool.

Documents & Unstructured Data Leases, loan documents, appraisals, environmental reports, offering memorandums, partnership agreements. These live in shared drives, Dropbox, Box, SharePoint, or (more often than anyone wants to admit) in email attachments.

The question to ask: What decisions require someone to read a 200-page document and extract specific data points? AI can help extract clauses, financial covenants, or inspection findings. Test it on your documents and require source references so the team can check the result.

Spreadsheets & Models The unsung system of record in real estate. Underwriting models, budget templates, waterfall calculations, construction draw schedules. critical financial logic living in Excel files that get emailed around and versioned by filename.

The question to ask: Which spreadsheets contain institutional knowledge that would be painful to lose? These aren't just files. They're encoded decision-making processes. Understanding them is essential before you can determine whether AI can augment, automate, or replace the workflows they support.

Databases & Data Warehouses Some firms have invested in centralized data infrastructure. Snowflake, Databricks, a SQL Server database that IT set up a few years ago. Others haven't. Either way, the question is the same: Is there a single place where your operational, financial, and deal data comes together? You do not need to centralize everything before you start. Determine what the chosen workflow needs, where the authoritative source is, and how to access it.

---

How to Actually Uncover Use Cases

Make an idea specific enough to test. For example, which part of handling a maintenance request takes time, and what information does the person doing it need?

Here's the process we use:

1. Trace a specific process. Pick a core business process. say, underwriting a new acquisition. Then trace every piece of information that touches that process. Where does the rent comp data come from? Who pulls the historical financials? How does the team get from a broker email to a go/no-go decision? Every handoff between a system and a person, or between a person and a spreadsheet, is a potential intervention point.

2. Find the repeated handoffs. In almost every real estate firm, there are people whose primary job is moving data from one system to another. Pulling reports from Yardi into Excel. Copying deal terms from a PDF into a model. Reformatting investor data from Salesforce into a quarterly letter. Check where these handoffs involve simple copying and where someone applies judgment. The distinction affects how you automate the work and how you review it.

3. Check how often the task occurs. Some document types are touched constantly. lease abstracts, operating statements, offering memos. Others are pulled once and filed. Focus on the high-frequency documents first. If your team reads 50 OMs a month to screen deals, that's a concrete use case with measurable time savings. A less frequent task may still be valuable if a mistake or delay is costly.

4. Ask what breaks when someone leaves. This is a brutal but effective test. If a key analyst or property manager left tomorrow, which processes would grind to a halt? The answer usually points directly at institutional knowledge that's trapped in someone's head, their spreadsheets, or their personal workflow. Document the reasoning and exceptions with the person who knows the work before deciding what AI should handle.

5. Compare value, effort, and risk. Time spent is useful, but it is only one measure. Consider error costs, turnaround time, frequency, and the effort to build and maintain the workflow. Include the time needed to check AI output. Test a target improvement before using it to justify an investment.

---

What a Good Data Audit Looks Like

For a workflow that depends on several systems, answer these questions about the information it needs:

• What are our core systems of record, and what data lives in each?
• Who are the primary users of each system, and what do they actually use it for versus what it's capable of?
• Where does our unstructured data live, and how is it organized (or not)?
• Which processes involve manual data movement between systems?
• What reports or analyses take the most human time to produce?
• Where is our institutional knowledge concentrated. in systems, or in people?

If a needed input is missing or unreliable, investigate it before promising the workflow can be automated. That may mean a focused data audit. It does not need to delay a separate experiment that uses information your team can already access and check.

Pick one process to investigate

Choose a recurring report, analysis, or decision. Walk through it with the person who does it, using a recent example. Record the inputs, handoffs, judgment calls, and expected output.

That gives you something concrete to evaluate with your team or a potential implementation partner.

MetaProp Labs helps real estate firms plan and build AI workflows. Tell us what you are working on.