Where Should AI Go First in a Real Estate Company?

Ask the heads of investments, construction and accounting where AI should go first, and you'll get three reasonable answers.

Investments wants to screen more deals. Construction wants better access to historical costs. Accounting wants the invoice package reviewed before the next funding deadline.

Each team has a real problem and a promising tool. But you have one implementation team and a limited budget. A convincing demo will not tell you which project deserves the investment.

Which one gets funded?

This is where an AI transformation starts. Someone has to decide which change is worth making, what it will take to make it work, and how the company will know whether it helped.

At Mason, these are the kinds of requests we work through with real estate teams: borrower-package review, bid comparisons, invoice checks, historical estimates and recurring reporting. They sound like separate use cases, but they often run into the same problems: information is scattered across systems, key decisions depend on a few experienced people, and outputs need checking before anyone can use them.

Before choosing what to automate, leadership needs to understand how that work happens today. AI can help gather and organize that evidence, then keep it current as the business changes.

Your roadmap reflects who you heard from

An opportunity can reach the AI team because a business-unit leader keeps asking for it. Or because an employee built a prototype. Or because a vendor gave a convincing demo.

Those are useful signals, but they leave out problems that nobody has thought to bring to the AI team.

The controller who quietly fixes the same reporting problem every month may never submit an AI request. The development manager may ask for faster document summaries because that is the capability they've seen, even though the harder problem is comparing assumptions across projects. A team may accept a slow handoff as unavoidable because it has worked that way for years.

If this is how the opportunity list gets built, leadership is choosing among the problems that managed to become visible.

Before ranking the requests, investigate the work behind them and look for problems the list has missed.

Follow a real piece of work

Consider a hypothetical funding review. A package arrives, someone checks the invoices, a reviewer resolves questions, and the request moves forward.

The obvious AI project is invoice extraction. Read the documents faster. Put the fields in a spreadsheet.

That may be valuable. But follow an actual package and you might discover that the reviewer spends more effort finding the applicable contract, checking whether an item appeared in an earlier submission, or waiting for an explanation from the project team.

Faster extraction will not fix those delays. Its value depends on how much of the team's time actually goes into copying information, rather than finding answers or waiting for approval.

This is why the first conversation should be about a recent piece of work. Ask the person to walk through what arrived, what they opened next, what they had to check, and where they got stuck. Ask to see an ordinary example and one that went badly.

Then check that account against the records the company has approved for review: timestamps, supporting files, revisions, exception records and approvals.

A final report tells you what was produced. It rarely tells you everything someone did to produce it.

Talk to the person who receives the completed work, too. If accounting completes its review sooner but the package still sits with the next approver, you need to understand that dependency before promising a shorter funding cycle.

AI can help with discovery, too

There is an opportunity to use AI earlier in this process.

AI-assisted interviews can help a team investigate recurring work in more detail. When someone calls a task “manual,” the interviewer can ask which information they copy, where it goes, and what they do when the figures disagree.

With the company’s permission, AI could also examine selected workflow records, compare how similar cases were handled and identify patterns worth investigating. At Mason, we are exploring how these systems could help teams find potential problems and gather the evidence needed to investigate them.

These findings are starting points for investigation, not conclusions. A long gap between two timestamps might be a bottleneck, an intentional control, or simply missing data. An interview can miss something too. The process owner has to help interpret what the evidence means.

The goal is to make it practical to investigate more of the business. Leadership still decides what matters, and the people doing the work still need to trust the process.

Start with the smallest set of approved information that can answer the question. Discovery should earn deeper access by showing that it produces useful findings.

Compare the changes you could make

Once the work is understood, you can make a much better investment decision.

An invoice-review project might create value through fewer corrections and less reviewer effort. An investment project might increase the number of opportunities a team can assess. A construction project might make historical comparisons practical before a budget is approved.

Not every benefit is a cash saving, and the comparison should make that clear. Faster diligence also doesn't, by itself, prove better investment returns.

Put the opportunities through the same questions while preserving the outcome that matters to each team:

• What would improve, and how much would that matter to the business?

• How often does the problem occur, and who is affected?

• What evidence supports the expected improvement?

• Can the available technology handle the actual inputs and exceptions?

• What needs to change in systems, responsibilities and daily work?

• Who owns the result after launch?

The answer should fit in a decision memo. A sponsor ought to be able to see why one project deserves attention before another, including where the business case is uncertain.

An attractive opportunity with no usable data and no process owner may need preparation before it needs a build. A less ambitious workflow with a clear owner and a testable outcome may be the better first investment.

Let the problem determine what you buy or build

A serious assessment has to leave room for several answers.

An existing product may already do the job well. A custom workflow may be justified because the firm's process or information is unusual. You may need a combination: a specialist tool for one step and an integration into the systems the team already uses.

Sometimes the right first change is an intake rule, a standard template or a clearer approval responsibility. That change should still count as progress.

For the funding-review example, a better submission checklist might prevent avoidable back-and-forth. Software can handle exact comparisons and calculations. AI can help interpret inconsistent documents and prepare an explanation. The reviewer can make the decision with the supporting evidence in front of them.

Choose the combination that improves the whole job. Include implementation, ongoing maintenance and human review when comparing the options.

The solution also needs to fit how the team works. If the reviewer works in Excel and email, design the handoff around that reality. Ask what they need to review, how they will correct it, and where the approved result belongs.

Decide what success means before the first release

For that same review, I would want to know how long a package takes from receipt to approval, how much effort reviewers put into it, how often it comes back for correction, and which exceptions cause the most trouble.

Establish those measures before changing the process. Measure them again afterward, taking account of differences in package complexity and workload.

That is what monitoring should tell you: whether the work is improving and what still needs to change. Usage and cost logs contribute to the picture. So do output quality, review effort and feedback from the people responsible for the work.

If people use the system but still repeat the old checks, investigate why. If preparation improves and approval becomes the constraint, the next improvement may be a different project. If an intervention produces too little benefit to justify its upkeep, stop expanding it.

The same evidence that helped you choose the work should help you decide whether to continue.

Start somewhere you can learn

You don't need a complete map of the company to begin. Choose one business unit or a connected set of workflows. Include the steps before and after it, so you can see whether a faster task would actually lead to a faster result.

Speak with the people doing the work, gather the evidence and compare a few possible changes. Choose one, assign someone to own the result and record how the process performs before you change it.

Then use what happened to make the next decision better.

That is what we mean by continuous AI transformation at Mason. Discovery, implementation and measurement belong in the same operating process. Every deployment should teach the company something about where to invest next.

If you're deciding where AI should go first, bring us one recurring workflow your team wants to improve. We can work through what happens today, what is worth changing, and how you would know it worked.