# AI in real estate: make property service more dependable AI-assisted editorial draft • Bill Allred review pending • Draft and sources checked 19 October 2026 ## Executive answer Start with the coordination between a service request and a verified resolution. For an operator of existing commercial buildings, AI may help assemble scattered information, identify missing details, and prepare the next action. The strategic opportunity is a more dependable service operation: managers can cover more requests with context, occupants receive clearer updates, and recurring failures become visible enough to address systematically. Lower administrative effort is one benefit, not the whole business case. This guide deliberately concerns commercial property maintenance and occupant service. It does not cover tenant eligibility, screening, housing allocation, lease enforcement, rent setting, or property-price forecasting. Those are different decisions with different consequences. ## A hypothetical operating situation Imagine an operator responsible for six office buildings. Occupants report comfort and maintenance issues through a portal and email. A coordinator translates reports into work orders, checks asset history, finds a vendor, and answers status questions. After hours, an established emergency process remains available. In this hypothetical example, three occupants report warm rooms on adjacent floors. The system proposes that the reports may concern the same air-handling asset, but one room has an ambiguous location. An AI assistant assembles the reports, links the possible asset records, highlights the uncertainty, and prepares a coordinator brief. It does not diagnose the equipment, alter controls, or suppress any request as a duplicate. The coordinator confirms the location and the relationship before assigning work. A new capability becomes possible: a portfolio-level view of unresolved service issues with consistent explanations and accountable next steps. Whether that supports portfolio growth or a differentiated service offering depends on observed demand and actual delivery capacity. It is not an assumed revenue gain. ## Segment the opportunity before selecting a workflow Different property activities need different designs: - **Occupant service:** organize requests, prepare factual acknowledgments, and identify unanswered questions. Judge usefulness by clarity and dependable follow-through. - **Maintenance coordination:** retrieve approved asset information and prior work, propose routing, and assemble vendor handoffs. Judge usefulness by complete information and correct ownership. - **Portfolio oversight:** surface repeated failures and aging exceptions for management review. Judge usefulness by whether managers can make better maintenance and resource decisions. - **Capital planning:** organize condition evidence and alternatives for specialists. Do not treat a language model's confident summary as an engineering assessment or investment recommendation. The Department of Energy's operations-and-maintenance guidance distinguishes operational, maintenance, engineering, training, and administrative responsibilities. That is a useful reminder to redesign the whole handoff, rather than make a faster inbox while leaving every downstream constraint unchanged. [DOE O&M challenges and solutions](https://www.energy.gov/cmei/femp/operations-and-maintenance-challenges-and-solutions) ## Workflow and decision map 1. **Receive and preserve.** Retain the original report, timestamp, building, location, and relevant attachments. Show the occupant the established urgent-contact route; do not make model classification the only emergency safeguard. 2. **Assemble context.** Retrieve only authorized records for that building. Distinguish reported symptoms from confirmed observations. Missing asset identity stays missing until checked. 3. **Propose the next step.** Suggest a category, likely responsible team, and necessary clarification. The coordinator owns priority and assignment; suspected hazards enter the existing emergency procedure immediately. 4. **Authorize and execute.** The responsible manager approves spending and access as required. Qualified people inspect and perform physical work. The assistant may prepare a vendor packet but must not invent access permission or a repair instruction. 5. **Update and verify.** Draft an occupant update from the actual work-order state. The technician records completion evidence; the designated owner verifies closure and any follow-up. Status semantics matter. IBM Maximo documentation distinguishes approved work, physical completion, and closure. An integration must preserve its own system's meaning rather than interpreting every positive technician message as a closed case. [IBM work-order statuses](https://www.ibm.com/docs/en/masv-and-l/maximo-manage/cd?topic=overview-work-order-statuses) ## Separate pilot review from the intended operating model The initial review process is a way to gather evidence. Later, an authorized routine status update could run without individual approval if tested controls verify the recipient, property, current work-order state, and permitted content. Conflicting records and uncertain identity should block it. Test the checks themselves with wrong inputs and plausible wrong outputs, and measure missed errors and exception workload. Spending, access, physical work, and closure authority still follow the organization's policy; automation does not grant new decision rights. ## Use the worksheet to make a pilot decision The companion [property-service worksheet](https://altiport.ai/downloads/ai-property-operations-worksheet.md) asks the team to specify one request class, evidence sources, human decisions, and acceptance tests. Fill it out with an operations manager, coordinator, technician, and system owner. Start with historical cases, then shadow the live workflow without changing it. Include easy cases, wrong building identifiers, duplicated reports, stale vendor records, unresolved safety language, and completed work that failed again. Compare the assistant's proposal with a reviewed reference answer. Test whether it exposes uncertainty and routes exceptions correctly, not merely whether its prose sounds reasonable. Track time to a useful response, time awaiting a decision, reopen rate, assignment corrections, and coordinator effort per resolved request. Count service quality and newly supportable workload separately from cash savings. Faster acknowledgment is not faster repair; reduced drafting time is not automatically reduced payroll. ## Acceptance, limits, and next step Before a bounded pilot, require source-linked summaries, property-level access tests, correct status handling, an operational fallback, and named coverage for escalations. Treat any cross-property information exposure or fabricated completion as a stop-and-investigate event. Agree numerical targets locally using the baseline, case severity, and available reviewers; there is no universal acceptable error rate. NIST identifies confidently false content as a generative-AI risk. Retrieval alone does not establish that a proposed answer is true. [NIST Generative AI Profile](https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.600-1.pdf) The next decision is modest but consequential: can this team reliably improve one service handoff while keeping physical work and accountability intact? If yes, expand by request type and property readiness. If not, fix ownership or source data before adding more automation.