|
flevyblog
The Flevy Blog covers Business Strategies, Business Theories, & Business Stories.
|
Editor's Note: Take a look at our featured best practice, Cost Segregation Study: Multiple Property Class Analysis (Excel workbook). Accountants, accounting firms, and real estate property owners can all benefit from this cost segregation study (CSS) template. It makes the estimated benefit easy to figure out and has flexible assumptions for high level or detailed segregation analysis. You get a summary of cost allocations [read more]
* * * *
Real estate is moving beyond chatbots, listing copy and isolated predictions. The next phase connects AI to live property data, transaction state, customer context, permissions and human approval. That shift creates larger opportunities, but it also makes data quality, governance and accountability part of the product itself.
For several years, the real estate debate around artificial intelligence was dominated by a dramatic but imprecise question: will AI replace the real estate agent? The question was easy to understand, but it compressed a complex industry into one occupation and one outcome. It also directed attention toward the most visible outputs, including listing descriptions, chatbots, virtual staging and automated price estimates, while more consequential changes were beginning inside search platforms, brokerage systems, contracts, property operations, valuation, construction and data infrastructure.
The more useful question is how AI is redistributing work, information, control and economic value across the real estate system. The answer is neither that AI is replacing the industry nor that it is merely a writing assistant. Real estate is entering a second phase of adoption. The first phase added AI features to existing tasks. The emerging phase connects models to domain data, customer identity, workflow state, permissions, audit records and human approval.
That difference matters. A model that drafts an email is easy to reproduce. A system that understands the current transaction, retrieves the authoritative document, identifies the next permitted action, protects confidential information, records the evidence behind its output and escalates exceptions is much harder to build. The most durable value is moving away from isolated tools and toward governed workflows.
| The central shift
Real estate AI is moving from producing content to coordinating work. The competitive advantage will come less from access to a general model and more from the ability to combine trusted data, domain rules, workflow state and accountable human decision-making. |
The 2025 REALTORS® Technology Survey offers a useful picture of the adoption-impact gap. Forty-one percent of respondents reported using AI or generative AI, and 20% said they used it daily. Yet 46% reported no noticeable business impact, compared with 33% who reported a moderately positive impact and 17% who reported a significantly positive impact. ChatGPT was the most frequently named AI product.
Those results do not mean AI has failed. They show that use and transformation are different. A professional who asks a public chatbot to improve a message is an AI user, but the brokerage workflow remains essentially unchanged. The larger gains appear when AI is embedded inside the systems where decisions, documents and handoffs already occur.
The same pattern appears at the enterprise level. JLL’s 2025 Global Real Estate Technology Survey reported that 88% of investors, owners and landlords and 92% of occupiers had started AI pilots. Only 5% of occupiers said they had achieved all program goals. Most initiatives were still experimental, and organizations were pursuing several use cases at once.
This pilot gap is one of the most important facts in the market. It explains why impressive demonstrations can coexist with limited operational change. Real estate organizations often add AI on top of fragmented data, overlapping software and undocumented workarounds. The model may perform well in a controlled example, but the workflow fails when a record is missing, two sources conflict, a user lacks permission, or the situation requires professional judgment.
Three maturity levels are emerging. At the surface level, AI produces content or answers questions. At the embedded level, AI works inside a CRM, transaction platform or property-management system. At the governed level, AI can act only within defined permissions, show its evidence, preserve a complete history and return difficult cases to a qualified person. The third level is where AI begins to alter the operating model rather than simply improve an individual task.
Figure 1. Adoption has expanded faster than measurable business impact. Percentages are from the 2025 NAR technology survey.
Home search is one of the most visible areas of change. Traditional portals organized the market around location, price, bedroom count and a fixed set of filters. Conversational search allows a user to describe a more complete situation: a monthly budget, commuting needs, desired building features, accessibility requirements, space for working at home, or a preference for renting and buying options to be compared together.
In March 2026, Zillow introduced an AI mode that connects conversation with live for-sale and rental listings, affordability questions, tour scheduling and agent contact. Zillow described the product as a shift from search toward coordinated action. Its later personalized hub organized a buyer around milestones such as setting a budget, finding a home, making an offer and closing.
At first glance, this appears to be a better search box. The more consequential change is that the platform attempts to retain the customer relationship across a much larger portion of the housing journey. Search becomes the entry point to financing, collaboration, professional services and transaction management rather than a separate activity that ends when the user opens a listing.
Natural-language search itself will not remain a rare competitive advantage. Large portals can access similar foundation models, and consumers will increasingly expect conversation. Advantage will depend on which platform can combine fresh inventory, verified property facts, affordability tools, customer context, professional relationships and permitted actions without losing trust.
That creates a difficult explainability problem. A consumer may ask for a quiet neighborhood, a manageable commute and a total monthly cost below a threshold. The system must translate those preferences into data without presenting assumptions as facts. It must distinguish between a verified listing field, a calculated estimate and an inference. It must also avoid discriminatory steering and give the user a meaningful way to correct the recommendation.
The strongest future search products will therefore do more than rank listings. They will act as decision-support systems. They will make trade-offs visible, show the basis of a recommendation and identify where the available data is incomplete. Confidence without provenance will become a liability rather than a feature.
For individual agents and brokerages, the earliest mainstream use case was marketing content. Generative tools could draft listing descriptions, social posts, newsletters and follow-up messages. These tools saved time, but they were easy to copy and rarely changed the economics of a brokerage.
The stronger emerging opportunity is relationship intelligence. Brokerages already possess years of poorly organized information: past conversations, property interests, showing feedback, transaction history, family circumstances, documents and follow-up commitments. An integrated assistant can retrieve the relevant context, identify what changed and prepare a low-risk next action. It can help an agent understand why a contact may be active again without requiring the agent to reconstruct the relationship from scattered notes and email threads.
This is also where the line between helpful service and surveillance becomes thin. A consumer may not expect a professional to know that they resumed browsing late at night or repeatedly viewed a specific property. A system may misinterpret curiosity as intent, or use personal information for a purpose that was never clearly disclosed. Strong products will need purpose limitation, role-based access, meaningful controls and communications that do not manipulate the customer.
The professional role also needs to be examined at the level of individual activities rather than as one replaceable job. A task-by-task analysis of whether AI will replace real estate agents shows why some activities can be automated after setup, others require professional review, and others remain human-led because they depend on negotiation, physical observation, local context, conflict management and accountability.
That division of work is likely to define the next brokerage model. Less time will be spent locating context, repeating data entry and drafting routine communications. More time will be spent interpreting circumstances, testing assumptions, advising clients, negotiating and accepting responsibility for consequential decisions. AI changes the allocation of work even when it does not remove the professional from the transaction.
A real estate transaction produces a dense stream of agreements, amendments, conditions, notices, deposits, identity records, financing documents and deadlines. Much of this information still moves through email attachments and manual checklists. That makes transaction coordination one of the most credible near-term AI opportunities.
Platforms such as Docusign Iris are expanding agreement intelligence through term extraction, review assistance, searchable repositories and obligation tracking. The same capabilities can be adapted to real estate files. A purpose-built transaction workspace could identify dates, confirm which documents are present, detect inconsistencies, prepare status updates and route missing items to the correct participant.
The distinction between coordination and judgment is essential. Extracting a date is not the same as interpreting the legal effect of a clause. Identifying that a financing condition remains outstanding is not the same as advising whether it should be waived. A system should show the underlying source, preserve the event history and require approval before any consequential action.
The best transaction platform will not attempt to become an invisible autonomous representative. It will make the file more complete, the timeline more visible and the handoffs less fragile. Its value will come from reducing missed steps, duplicate work and uncertainty among agents, administrators, clients, lenders, inspectors and lawyers.
Over time, the transaction may become a shared data environment rather than a collection of documents. The purchase price, deposit, conditions, parties and deadlines would exist as structured data linked back to the signed agreement. Each participant would see the portion they are entitled to access. Changes would be recorded once and reflected across the workflow. AI would help interpret and coordinate that environment, but the authoritative record would remain visible.
Property management has many of the conditions that favor workflow automation: recurring requests, large portfolios, measurable service levels, distributed teams and direct links between delay and cost. Leasing inquiries, tour scheduling, payment reminders, maintenance triage, invoice processing and renewal outreach can all be defined as bounded workflows with clear escalation points.
In 2026, Yardi expanded AI agents across multifamily operations, including leasing, resident service, inspection and accounting workflows. AppFolio has similarly developed agentic functions across leasing, maintenance and resident operations. These are vendor announcements, so reported performance should not be generalized without independent evidence. They nevertheless show that the market is moving from tools that answer questions toward systems that can complete bounded work inside existing property software.
A resident request can be classified, matched to the property and lease, checked against prior work, routed to an approved vendor, scheduled and followed through to completion. An invoice can be read, matched to a purchase order and directed to a reviewer when it falls outside policy. A portfolio manager can ask a question across operating data without waiting for a custom report.
The operational opportunity is substantial because the outcome can be measured. Response time, leasing conversion, work-order completion, invoice-processing time, renewal rate and vacancy are observable. This makes it easier to determine whether AI has improved the business or merely added a new interface.
Property management also demonstrates the governance risk. Pricing and revenue-management systems can influence rents across a market. The Canadian Competition Bureau closed its investigation into RealPage and Yardi after finding that revenue-management tools had not been adopted widely enough in Canada to substantially harm competition during the period examined. The investigation still illustrates a broader point: a system can be commercially effective for one operator while raising competition, fairness or affordability concerns at the market level.
The next generation of property-operations systems will therefore need controls that extend beyond technical accuracy. They will need policies governing data sharing, pricing inputs, resident communications, service prioritization and the treatment of vulnerable tenants. Operational efficiency is not a complete measure of responsible performance.
Automated valuation models are not new. What is changing is the range of data they can process and the expectations placed on their controls. Text, images, property documents and building records can supplement structured comparable-sales data. At the same time, regulators are placing greater emphasis on testing, conflicts, manipulation and nondiscrimination.
The U.S. interagency quality-control rule for automated valuation models requires covered institutions to maintain policies and controls designed to support confidence in estimates, protect against data manipulation, avoid conflicts, test performance and comply with nondiscrimination law.
These requirements point toward a more useful product category: valuation assurance. Instead of presenting one number as objective truth, the system should show its evidence, identify the limitations of the comparable set, quantify uncertainty and explain which property characteristics could cause the estimate to fail. Unusual properties should be recognized as unusual and directed to deeper review.
This matters because real estate is local and heterogeneous. A substantially renovated house, a live-work property, a legal non-conforming use or a home in a thin market may not behave like the model’s dominant training examples. Photographs may reveal that a kitchen is updated, but they may not establish workmanship, permits, hidden conditions or the cost of reproducing the improvement. A fluent explanation cannot repair weak valuation evidence.
The strongest systems will combine different methods. Traditional statistical and machine-learning models may remain better suited to numerical estimation. Language models can help retrieve documents, organize evidence and explain context. Computer vision can support condition analysis. Human review remains necessary where the property, data or decision falls outside the model’s reliable range.
Construction and development provide another important test because the cost of a missing answer or outdated record is easy to observe. Project teams manage drawings, contracts, schedules, RFIs, submittals, changes, safety records, daily logs and site photographs. AI can retrieve the relevant record, compare versions, identify a discrepancy and prepare a review for the responsible person.
The commercial real estate market is also moving beyond individual productivity tools. JLL identified 56 AI use cases across the corporate real estate value chain, with organizations piloting applications in portfolio strategy, lease administration, facilities, workplace experience, energy management and investment analysis. The challenge is not a shortage of ideas. It is the ability to scale selected uses across inconsistent portfolios, systems and markets.
Construction exposes why evidence must remain attached to the output. Progress affects payment. A change can create a claim. A safety observation can trigger a serious response. A model that summarizes a site record is useful only when the user can inspect the underlying record and understand the limits of the analysis. The likely operating model is hybrid: AI accelerates search, comparison and prioritization, while qualified people verify physical conditions and accept responsibility for decisions.
This principle will extend to building operations. Sensors, maintenance histories, utility data and equipment manuals can help identify abnormal conditions and predict failure. But a predictive system must also account for missing sensors, incorrect equipment records, deferred maintenance and local operating context. The product is not simply the prediction. It is the workflow that turns the prediction into a verified inspection, approved work and a documented outcome.
Real estate data is valuable because it is local, current and connected to physical assets. It is also fragmented and contractually restricted. Listings, photographs, floor plans, transaction records, consumer behavior, building systems and municipal information may each have different owners, permissions and retention requirements.
The Real Estate Standards Organization’s 2026 work on hierarchical AI governance and machine-readable usage rights shows where the industry is heading. The objective is to translate contractual terms into data that software can enforce, including whether content may be displayed, analyzed, retained, shared or used for model training.
This is not an administrative detail. An AI product cannot become a trusted part of the industry if it cannot determine whether it is allowed to use the information in front of it. Permissions need to travel with the data. Identity and role need to determine which action is available. Deletion, correction and withdrawal need to propagate through connected systems.
Provenance is equally important. When a system states that a building has a certain number of stories, parking spaces or amenities, the user should be able to see whether the fact came from an official record, an MLS field, a listing description, an uploaded document or an inference from imagery. Conflicting values should not be silently resolved. They should remain visible until a reliable source or qualified reviewer confirms the value.
This creates a defensible opportunity. Organizations that preserve lineage, accumulate verified corrections and manage rights can build a learning loop that a generic model cannot easily reproduce. The interface may be copied. The trusted data history and permission structure are much harder to duplicate.
Figure 2. Durable advantage increases as AI moves from surface features into domain intelligence, coordinated workflows and governed data.
A hallucinated answer in a casual conversation may be inconvenient. The same error inside a transaction can affect a deadline, disclosure, advertisement or financial decision. Real estate systems should not allow model-generated text to become an authoritative fact merely because it sounds plausible. Source retrieval, confidence thresholds, conflict detection and approval rules must be built into the workflow.
Real estate files contain identity records, income, credit information, addresses, family circumstances, financial obligations and negotiation strategy. The Office of the Privacy Commissioner of Canada has emphasized that existing privacy obligations continue to apply to organizations developing and using generative AI. Collecting more data because a model may find it useful is not a sufficient purpose.
Systems should minimize the data they use, limit access by role, disclose material uses and provide a path for correction. A brokerage or landlord should also consider whether a customer reasonably expects a particular form of analysis. Legal permission alone does not guarantee that the use will feel fair or maintain trust.
Recommendation, screening, advertising and pricing systems can reproduce historical patterns or create new ones. A housing platform may steer users through proxies even when it does not use a protected characteristic directly. A tenant-screening system may convert incomplete or inaccurate records into a high-risk score. A pricing system may change competitive behavior across a market. These are not problems that can be solved only by removing one sensitive field. The complete system, including its data, objective, feedback loop and business incentives, needs to be tested.
Organizations sometimes claim that a person remains “in the loop” when the workflow gives that person little time, weak evidence and strong pressure to accept the output. Effective review requires authority, context and a clear duty to challenge. The system must make disagreement possible and preserve it in the record.
Generative tools can create convincing employment letters, income records, identification images, listing photographs and voices. Real estate transactions are particularly exposed because they are urgent, distributed and financially significant. Verification systems will need to compare information across documents, confirm identity through independent channels and treat metadata or visual anomalies as signals rather than definitive proof. Fraud detection itself must remain reviewable so that an innocent applicant or client can correct an error.
Canada shares many technology trends with the United States, but the structure of the market is different. Local real estate boards, provincial regulators, privacy law, mortgage practices, transaction forms and data permissions shape what can be deployed. A product designed around one U.S. market may not understand the legal responsibilities, document flow or data access rules in a Canadian province.
In June 2026, the Canadian Real Estate Association stated that AI use in real estate should be guided by transparency, accuracy and accountability. It also emphasized that adoption does not remove the professional responsibilities of REALTORS® for the information, advice and services they provide.
That position creates friction for products built around broad autonomy, but it also defines a market opportunity. Canadian systems can be designed around provincial workflows, local forms, board permissions, privacy requirements and professional review. The strongest local products may not be the most dramatic. They may be the ones that make data more reliable, transactions more complete and AI use easier to govern.
Near-term Canadian opportunities include property-data normalization with source lineage, transaction coordination, privacy-preserving CRM intelligence, fraud controls, valuation assurance and practical governance tools for brokerages and property operators. Canadian adoption may appear slower in some segments, but the underlying problems are not smaller. They are often more fragmented and jurisdiction-specific, which increases the value of domain expertise.
The most credible opportunities share three characteristics. They address a recurring workflow with measurable cost or risk. They use domain data that can be traced and governed. They keep consequential judgment and accountability visible. On that basis, several opportunity areas stand out.
The near-term opportunity is not autonomous negotiation. It is a shared workspace that converts signed documents into structured milestones, identifies missing information, tracks conditions, prepares routine updates and directs exceptions to the correct professional. The value can be measured through fewer missed steps, faster turnaround and lower administrative burden.
Real estate platforms need systems that extract facts from descriptions, photographs, floor plans and documents without presenting every extraction as confirmed. The strongest product will show the source, compare conflicting values and preserve verified corrections. This can improve search, valuation, building pages, compliance and analytics at the same time.
Leasing, maintenance, resident communication, invoice processing and renewals offer repeatable workflows and observable results. The opportunity is largest where AI is integrated into the property system of record rather than placed beside it. Governance will be especially important where the system influences rent, screening, service priority or access to housing.
Remote transactions and generative media increase the need for independent verification. Systems can help identify cross-document inconsistencies, duplicate records, unusual metadata and identity mismatches. The commercial opportunity is significant, but the product must support review and correction rather than produce an unexplained fraud score.
The better opportunity is not another unqualified estimate. It is a system that tests inputs, shows uncertainty, identifies unusual properties, monitors model performance and directs difficult cases to deeper review. This product can serve lenders, appraisers, investors, brokerages and consumers without pretending that one model can eliminate professional judgment.
Most real estate technology loses the relationship once the transaction closes. A persistent property record could organize permits, warranties, maintenance, renovations, insurance information, equipment details and future resale preparation. With user control and clear permissions, this could create a longer customer relationship and improve the quality of property data available at the next transaction.
Large companies can build internal policy, testing and audit functions. Independent brokerages, property managers and smaller developers often cannot. There is a market for practical governance tools that control approved models, protect confidential data, record material AI use, test outputs and define which actions require approval. Responsible AI can become a product category rather than a policy document.
Search and advisory tools can help consumers compare monthly cost, commute, building characteristics, property condition and long-term obligations. The opportunity is not to tell people where they should live. It is to organize verified information, expose assumptions and help them prepare better questions for the professionals involved.
The market contains many impressive demonstrations and fewer proven operating systems. Managers need a practical way to distinguish between them. Five tests provide a useful starting point.
| Test | Management question |
| Truth test | What authoritative source supports the output? How current is it, and what happens when sources disagree? |
| Authority test | Which actions may the system perform, which require approval, and who remains responsible? |
| Value test | Which measurable delay, cost, conversion loss or risk is reduced? What is the baseline? |
| Trust test | Can the affected person understand, correct or challenge the information and outcome? |
| Scale test | Will accuracy, security and compliance hold when the system moves beyond a selected pilot into different users, markets and edge cases? |
These tests change the implementation sequence. The project no longer begins by asking where AI can be added. It begins with a defined workflow, an authoritative source, known exceptions, clear duties and baseline performance. The organization then determines which steps can be automated, which require review and which should remain human-led.
This approach also makes pilots more useful. A pilot should not only demonstrate that the model can perform the happy path. It should test missing data, contradictory records, unauthorized requests, edge cases, user corrections and escalation. The product should be evaluated not only by the quality of its average answer but by the safety of its failure modes.
Artificial intelligence is not entering real estate through one dramatic substitution. It is advancing through hundreds of smaller changes to how people discover property, interpret information, maintain relationships, coordinate transactions, operate buildings, value assets and manage construction.
The first wave made AI visible. The next wave will make it operational. That transition will reward organizations that can connect intelligence with data rights, workflow state, evidence, approvals and trusted human responsibility. It will punish organizations that automate weak processes, treat model confidence as fact or use personal data without a clear purpose.
The winning question is no longer whether a model can produce an impressive answer. It is whether the complete system can perform useful work, show its basis, respect its authority and remain reliable when the transaction becomes difficult.
In real estate, the future belongs less to the autonomous chatbot than to the governed operating model around it.
You can download in-depth presentations on Real Estate and 100s of management topics from the FlevyPro Library. FlevyPro is trusted and utilized by 1000s of management consultants and corporate executives.
For even more best practices available on Flevy, have a look at our top 100 lists:
These best practices are of the same as those leveraged by top-tier management consulting firms, like McKinsey, BCG, Bain, and Accenture. Improve the growth and efficiency of your organization by utilizing these best practice frameworks, templates, and tools. Most were developed by seasoned executives and consultants with over 20+ years of experience.
