How Agentic AI Transforms Real Estate Operations

How Agentic AI Is Rewiring Real Estate Operations

Real estate has never had a shortage of work. What it has had - across leasing, maintenance, reporting, collections, and acquisitions - is a shortage of coordination.

That distinction matters.

In a recent talk on AI’s role in rental housing, McKinsey senior partner Aditya Sanghvi made a practical case for why agentic AI may be more important to real estate than standard generative AI. His core argument is not that AI will replace plumbers, property managers, or investment professionals. It’s that AI can eliminate the operational gaps between people, systems, and decisions - the places where time, money, and resident trust are quietly lost.

For operators, investors, and PropTech teams, that framing is useful because it shifts the conversation away from hype and toward workflow design. The real opportunity is not "using AI" in the abstract. It is rebuilding high-friction processes around faster triage, cleaner data, better follow-up, and tighter human oversight.

This article unpacks that thesis, explains where the value actually comes from, and highlights what business leaders should do next.

Key Takeaways

• Agentic AI is different from chat-based AI because it can pursue a goal across steps, not just answer a prompt.
• In real estate, the biggest delays are often coordination delays, not labor delays - missed follow-ups, disconnected systems, and manual handoffs.
• Clean data is a prerequisite, not a nice-to-have. Even strong models fail when underlying property, resident, or operational data is inconsistent.
• The highest-value use cases are not novelty pilots. They are core operational domains like maintenance, renewals, collections, and underwriting.
• Humans still need to stay in the loop, especially when money, compliance, resident experience, or judgment calls are involved.
• Leaders should avoid two extremes: bolting AI onto broken workflows or trying to automate entire processes without redesign.
• The most durable competitive advantage may become who owns the memory - the data, feedback loops, and operational learning across a portfolio.
• A practical starting point is to pick two or three business-critical workflows, define measurable outcomes, and clean the required data inside those workflows.

The Big Shift: From Generative AI to Agentic AI

Most businesses entered AI through chat interfaces: drafting emails, summarizing documents, or generating basic content. That is useful, but in operational settings it often behaves like a smarter search tool.

Sanghvi’s distinction is more strategic:

• Generative AI responds to prompts
• Agentic AI works toward goals
• Multi-agent systems coordinate multiple specialized agents across a full process

That progression matters because real estate work rarely lives inside a single task. A lease renewal is not one document. A maintenance request is not one message. An acquisition decision is not one spreadsheet. These are linked workflows involving intake, analysis, routing, approvals, updates, and post-action learning.

In other words, the industry doesn’t just need answers. It needs orchestration.

For technical leaders, this is the architectural unlock. The question is no longer, "Can AI write this?" It becomes, "Can AI move this workflow forward safely across systems, people, and decision points?"

Why Real Estate Is a Strong Fit for Agentic AI

Real estate is often viewed as a slow adopter of technology, but operationally it may be unusually well-suited for agent-based automation.

1. Coordination is the real workload

A large share of property operations is made up of repeatable but fragmented tasks:

• Logging issues
• Routing vendors
• Sending updates
• Tracking response times
• Escalating exceptions
• Aggregating reports
• Preparing approvals

These tasks are individually simple. The problem is the delay between them.

Sanghvi described this as the hidden drag in operations: the inbox that no one owns, the callback that never happens, the follow-up that slips. That observation rings true well beyond maintenance. It applies to leasing, renewals, turns, compliance documentation, vendor management, and investor reporting.

2. Repetition creates learning potential

Rental housing generates repeated decisions at scale:

• Which renewal offer should this resident receive?
• Which maintenance issue needs immediate escalation?
• Which delinquency case requires human intervention?
• Which deal should be screened out before IC review?

This repetition is what makes agentic systems more valuable over time. Each outcome becomes feedback. Each workflow becomes training ground. Done correctly, the portfolio starts to behave less like a static set of assets and more like a system that learns.

3. Improvements can propagate across the portfolio

One of the most important ideas in the talk is that operational improvements in real estate often stay local. A single site team figures something out, but the insight never reliably spreads.

AI changes that dynamic. Once a successful workflow, logic pattern, or triage rule is captured in a system, it can be deployed more broadly. For enterprise teams, that means best practices no longer have to travel as tribal knowledge alone. They can become part of the operating layer.

The Maintenance Example: AI Doesn’t Fix the Pipe - It Fixes the Process

The maintenance scenario is the clearest illustration of the value proposition.

When a leak happens, the physical repair still belongs to a skilled human. AI does not replace that. But nearly everything around the repair can be improved:

• Intake of the resident issue
• Classification of severity
• Routing to the right vendor or technician
• Follow-up when timelines slip
• Status updates to stakeholders
• Post-resolution learning

Sanghvi described a multi-agent setup with roles such as:

• An inspection agent to triage and prioritize tickets
• A notification agent to handle coordination and updates
• An action agent to route work and validate operational requirements
• A pattern agent to identify recurring inefficiencies and broader asset-level issues

That architecture is important because it mirrors how strong operational teams already think. Different functions have different jobs. AI becomes effective when it is specialized, connected, and governed - not when it is treated as one all-purpose brain.

The reported outcome in the example was substantial: facilities teams regained significant time, and issue resolution moved from weeks to days. Whether or not every operator sees the same result, the mechanism is credible. Reducing latency in handoffs almost always improves service delivery.

For operations leaders, that’s the practical lens to use: not "How much labor can I remove?" but "How much idle time can I eliminate?"

Renewals: A Better Use of Data Than Blanket Offers

Lease renewals are another strong candidate because they combine volume, margin sensitivity, and inconsistent execution.

Traditional renewal processes often suffer from three problems:

1. They start too late
2. They rely on generic offers
3. They fail to reflect resident-specific context

The agentic alternative described in the talk uses several layers of decision support:

• Surface upcoming expirations automatically
• Build a fuller resident profile
• Estimate churn or renewal risk
• Recommend an incentive tailored to that household
• Draft the outreach for manager review
• Learn from which offers succeed over time

The operational significance here is not just personalization. It is timing plus fit.

One example from the presentation highlighted that a non-rent concession - such as a gym benefit - could outperform a blunt rent discount for certain residents while costing less. That’s a meaningful reminder for revenue teams: better retention strategy is not always about giving away more value. It is often about matching the offer to actual resident behavior.

For strategic operators, this points to a measurable opportunity:

• higher renewal rates
• improved concession discipline
• better resident experience
• less manual workload per lease cycle

But it also introduces a governance question. To make these systems trustworthy, operators need clear rules around what data can be used, how recommendations are approved, and how fairness is monitored. Those specifics were not detailed in the video, but they become essential in practice.

Investment Committees: Scaling Judgment Without Faking It

The investment committee example broadens the AI discussion beyond property operations.

In the scenario Sanghvi described, multiple agents were trained on prior committee decisions and the tendencies of key decision-makers. Deal teams could then engage that system earlier in the process rather than waiting for formal committee meetings.

That idea is compelling for two reasons.

It democratizes institutional memory

Many firms depend heavily on a handful of experienced executives. Their judgment is valuable, but it is also bottlenecked. A well-designed agent system can make patterns from prior decisions more accessible to junior and mid-level deal teams.

It improves screening and preparation

If teams can pressure-test assumptions earlier - risk, market conditions, structure, downside scenarios - they may kill weak deals sooner and sharpen strong deals before they consume senior bandwidth.

Still, this is also where caution matters most. A model can summarize historical preference, but it cannot automatically justify future conviction. If a system reaches conclusions no one can explain, it will fail the trust test.

That’s why Sanghvi’s warning against full automation is so important. In investment workflows, explainability is not optional. It is core to adoption.

Why So Many AI Initiatives Fail to Create Value

One of the strongest sections of the talk was not about capability but about failure.

Sanghvi pointed to a pattern seen across industries: AI adoption is widespread, but material business impact is rare. His explanation is especially relevant for real estate teams deciding where to invest.

1. Too much focus on generative AI

If the deployment is mostly chat, drafting, and summarization, the gain is often individual productivity - not enterprise transformation.

2. Bad data breaks good models

This is the point that should resonate most with anyone operating in property data.

The speaker’s blunt assessment was that real estate data quality is poor: disconnected systems, inconsistent records, and weak governance. That diagnosis aligns with what technical teams already know. Property management systems, CRMs, spreadsheets, service logs, and ownership records often do not reconcile cleanly.

For developers and operators alike, the implication is simple: AI cannot reliably coordinate what your systems cannot consistently identify.

If tenant names, unit records, work orders, and financial fields do not line up, the workflow will break.

3. Too many small experiments

Pilot culture feels innovative, but dozens of disconnected experiments rarely change P&L outcomes. A tool that saves a few minutes in isolation is not the same as redesigning a core process end to end.

4. CEOs delegate too much

Sanghvi argued that AI transformation cannot be left to IT alone. That is a critical point. Infrastructure teams are essential, but the business must define where value lives, what tradeoffs matter, and what outcomes justify change.

5. Companies choose the wrong implementation strategy

Two failed extremes show up repeatedly:

• dropping AI into an existing broken workflow
• trying to automate an entire process in one leap

Both approaches ignore process redesign. The right move is to rethink the workflow, assign tasks intentionally between agents and humans, and build controls around the result.

6. Change management gets treated as an afterthought

Even a strong workflow design can fail if teams do not trust it, understand it, or see it reinforced in daily operations. Adoption is not automatic.

The Most Important Constraint in Real Estate AI: Data Quality

For the BatchData audience, this is the section that deserves the most attention.

Sanghvi’s central claim is that data is the limiting factor, not model availability. Everyone can access powerful AI models. Few organizations have the data foundation to use them safely and consistently.

That creates a major divide between experimentation and execution.

What "bad data" looks like in real estate

The video highlighted familiar issues:

• systems that don’t talk to each other
• inconsistent naming and record structures
• offline spreadsheets that bypass formal workflows
• unclear data ownership
• weak definitions of accuracy and quality

In practical terms, poor data quality leads to:

• wrong routing
• duplicate work
• weak personalization
• unreliable forecasts
• compliance exposure
• low user trust

For technical architects, this is where modern data infrastructure becomes a competitive advantage. The value is not in claiming access to a massive dataset. The value is in making operational data usable, connected, and dependable at the moment a workflow needs it.

Clean data should be domain-led, not abstract

One of the most useful recommendations in the talk was to avoid giant "fix all data first" programs. Instead, clean data in the context of a specific high-value domain.

That is a strong operational principle.

If renewals are the priority, clean the data required for renewals. If collections are the priority, clean the records that drive collections. If maintenance is the priority, standardize the fields, statuses, entities, and timestamps that maintenance automation depends on.

This approach does three things:

1. It ties data work directly to ROI
2. It lets business users validate quality in context
3. It avoids years-long projects that produce little operational lift

The Strategic Question: Who Owns the Memory?

Perhaps the most forward-looking theme in the talk was not automation itself, but ownership of learning.

If AI systems improve through repeated decisions, then the long-term advantage may belong to the organization that controls:

• the cleanest underlying data
• the strongest feedback loops
• the richest operational history
• the ability to apply learning across the portfolio

That raises strategic questions across the property stack:

• Will owners control the learning layer?
• Will operators become more valuable if they own execution data?
• Will software platforms capture the memory because workflows run through them?
• Will portfolio scale become more decisive because it creates more learning volume?

The video did not resolve those questions, and it shouldn’t have. They are still open. But they are exactly the questions enterprise real estate teams should be asking now.

For PropTech companies, this also reframes product strategy. The next battleground is not just system of record. It is system of learning.

What Business Leaders Should Actually Do Next

The talk closed with a set of recommendations for CEOs and business leaders. Translated into operator-friendly language, the advice is less about buying tools and more about choosing battles.

1. Reassess where your advantage really comes from

If your differentiator has historically been local market knowledge, relationship management, or process discipline, ask how much of that remains scarce in an AI-enabled market.

Some advantages may strengthen. Others may commoditize.

2. Pick two or three operational domains that matter

Do not start with dozens of experiments. Start with a short list of workflows where:

• value is measurable
• process volume is high
• handoff friction is visible
• data can be improved with manageable effort

Maintenance, renewals, and collections are good examples because they touch revenue, cost, and resident experience.

3. Redesign the workflow before adding AI

Map:

• what the human does today
• what the system should do automatically
• where approvals belong
• what data is required at each step
• how exceptions are handled

This is where many initiatives either create real leverage or quietly stall.

4. Curate the stack instead of chasing a single vendor answer

One of Sanghvi’s clearest points was that organizations should own their data but should not try to build foundational language models themselves.

That is sensible advice. Most real estate firms will win by combining:

• internal systems
• third-party models
• domain-specific applications
• governance controls
• data pipelines that connect the whole stack

For technical teams, interoperability matters more than buzzwords.

5. Build trust with guardrails

The best path to adoption is progressive autonomy:

• start with recommendations
• move to human approval
• automate more only after teams repeatedly confirm quality

This is especially important in workflows involving payments, resident communication, compliance, or underwriting.

6. Treat change management as part of the deployment

If teams are not trained, measured, and supported, the workflow will revert to old habits. New systems need new operating norms.

A Useful Reality Check: AI Will Not Save a Broken Operating Model

The most valuable subtext in this talk is that AI amplifies operating quality; it does not replace it.

If data is inconsistent, ownership is unclear, processes are poorly defined, and leaders are not aligned on outcomes, AI simply makes the confusion faster.

But if a company knows its priorities, has enough data discipline to support a workflow, and is willing to redesign work instead of layering tools on top of dysfunction, the upside is significant.

That is the real message for real estate teams: agentic AI is not magic, but it is a serious multiplier for organizations that are ready to operationalize it.

Conclusion

The strongest insight from this discussion is that real estate’s AI opportunity is less about replacing labor and more about removing delay.

That makes agentic AI especially relevant to an industry built on fragmented systems, repeated decisions, and time-sensitive coordination. Maintenance, renewals, reporting, and investment workflows all contain the same pattern: the work itself is often manageable, but the handoffs around the work create drag.

The firms that benefit most will likely be the ones that do four things well:

• choose high-value workflows instead of scattered pilots
• clean and govern data at the domain level
• redesign work around human-plus-agent collaboration
• capture feedback so each decision improves the next one

For builders, scalers, and operators in property, that is the strategic takeaway. The next phase of AI in real estate will not be won by the company with the flashiest demo. It will be won by the company that can turn messy, manual coordination into reliable, learning-driven execution.

Source: "Winning with AI in Real Estate with Aditya Sanghvi" - Walker & Dunlop, YouTube, Aug 13, 2026 - https://www.youtube.com/watch?v=qrOAyzc0at0