AI for property management companies in 2026: where the staff hours really are

Adoption moved fast and stayed shallow

Two years ago, AI in a property management office mostly meant someone drafting a listing in ChatGPT. IREM and AppFolio reported in December 2025 that the share of property managers using AI tools rose from 21% to 45% between late 2023 and mid 2025, and that 62% of current users had started in the past year. Most rely on general tools such as ChatGPT or Microsoft Copilot rather than software built for the industry.

Buildium's 2026 industry report, produced with NARPM, puts adoption among property management companies at 58% in 2025, up from 20% in 2024, and adds that only 8% had been able to fully automate any process. The surveys ask different questions of different people, so the numbers do not line up. They agree that many teams use AI and few have changed how the work gets done.

The National Apartment Association's 2025 Performance Ecosystem survey, sponsored by AppFolio, asked which AI tools companies use most. General-purpose tools led at 53%, AI features inside the company's own property management software followed at 43%, and 24% of respondents said they use no AI tool in their role.

Where onsite time goes

The same NAA survey asked 1,984 apartment professionals how they spend the week. On average, 42% went to routine operational work such as paperwork, data entry and administration, and 24% to reactive tasks such as troubleshooting and urgent requests. Strategic work took 16%. Asked how they would like to spend their time, respondents put strategic work at 26%.

Applied to one person as an illustration, a 40-hour week at those averages holds about 16.8 hours of routine work and 9.6 hours of reactive work. Those are the hours an agent can reach. Owner relationships and judgment calls about residents stay with people.

Staff turnover makes the routine work harder to hold together. NAA, citing an RCLCO survey, reported in April 2026 that turnover among onsite property staff reached 29.2% over 12 months, against 14.2% for corporate roles. When a property manager leaves, the knowledge of which vendor needs which endorsement, and which owner wants a call before any repair over a set amount, often leaves too. Written rules that an agent follows stay behind.

The jobs worth an agent

The jobs that suit an agent share a shape. A request or document comes in, someone looks something up in the property management system and checks it against a written rule, then drafts a reply or a letter. A person reviews the result and sends it.

Maintenance deserves its own mention because it shapes how owners see you. In Buildium's survey, 38% of rental owners named maintenance as their top source of stress. An agent that turns a resident's text into a work order with the unit, the category, the priority and a vendor whose insurance is current gives the coordinator a head start on every request.

Owner reports are a smaller job with a firm rule. Every number comes from the accounting system. The agent drafts the commentary on variances and maintenance spend, and the manager edits it before the owner sees it.

Across residential, commercial and association management, these are the jobs to count first:

• Vendor and tenant certificates of insurance, read against a requirement set, with deficiency requests drafted
• Maintenance intake: classifying requests, asking for photos, drafting work orders and resident updates
• HOA violation letters and architectural request packets for association managers
• Late rent notices prepared from the ledger, the lease and the state notice rules
• Owner report commentary drafted from the general ledger and work order history

What to keep away from AI

Tenant screening decisions come first. HUD published guidance in 2024 on how the Fair Housing Act applies to tenant screening, including screening that uses machine learning and AI, and the documents now sit on HUD's archive site. The guidance said the Act applies to housing decisions regardless of what technology is used, and that a housing provider stays responsible even when a screening company does the work. The statute applies either way. An agent can prepare an adverse action letter after a person has decided, with the notice elements the Fair Credit Reporting Act requires. The decision to accept an applicant or turn one down stays with your team.

Rent pricing is another line to hold. The Justice Department and a group of states sued RealPage over algorithmic rent pricing, and several large property managers have since been parties to proposed or final judgments in that case. A small company has little to gain from a pricing model and a lot to explain if one goes wrong. We do not build them.

Leasing chatbots that answer eligibility questions are a quieter risk. The Fair Housing Act bars any notice, statement or advertisement that indicates a preference or limitation on a protected basis, and HUD's regulations extend that to words, phrases, photographs and other content. A bot that improvises answers about who qualifies can create exactly that problem. If you use AI for listings or resident messages, have it fill templates your company approved.

Last, keep resident and owner data out of consumer AI accounts. Applications and maintenance notes contain personal information, and some contain health details a resident shared when asking for an accommodation. Any AI that touches them should run under a business agreement with clear data terms, ideally inside your own Microsoft 365 or cloud tenant, with no training on your data.

The messages that always go to a person

Some messages carry legal weight whatever channel they arrive through. Under HUD's regulations, refusing a reasonable accommodation can be unlawful discrimination, so a resident asking for a reserved parking space near the door or an exception for an assistance animal needs a person's attention, even when the request arrives as a maintenance ticket. Build the routing rules before the agent reads a single message.

Repairs that affect health or safety are the other category. Texas, for example, requires a landlord to make a diligent effort to repair a condition that materially affects the physical health or safety of an ordinary tenant once a tenant who is current on rent gives notice, with a rebuttable presumption that seven days is a reasonable time. It also requires an emergency phone number. An agent can flag such a notice for a person as soon as it arrives, and your staff stay responsible for it. These go to a person every time:

• Anything that could be a reasonable accommodation or modification request
• Fair housing complaints and any mention of discrimination
• Habitability disputes and threats to withhold rent
• Mentions of military service, which can bring federal protections into play
• Letters from attorneys and anything from a court

How to measure the hours

Before buying anything, count one ordinary week. Ask each coordinator and manager to tally requests by type, with rough minutes for each. A sheet with one row per workflow and a tick per request is enough to show where the time goes.

When you pilot an agent, count review time honestly. If a work order took 12 minutes by hand and now takes 4 minutes of review, the saving is 8 minutes. Track the share of items the agent routes to a person and how often staff correct its drafts. A falling correction rate tells you when a category might be ready for lighter review.

Worked example: maintenance intake at a 40-person manager

This is an illustrative example with assumed numbers, not a client result. Take a residential manager with 40 staff and 3,000 units. Residents send about 150 maintenance requests a week through the portal, text, email and the after-hours line. Turning each one into a dispatched work order, including the first reply and the resident update, takes about 12 minutes. That is 1,800 minutes, or 30 staff hours a week.

Assume 70% of requests are routine, with a clear category and an obvious vendor. With the agent drafting the work order and the resident reply, a coordinator spends about 4 minutes reviewing each of those 105 requests, 420 minutes in all. The other 45 still take 12 minutes each, another 540 minutes. Weekly time falls to 960 minutes, or 16 hours, which gives back 14 staff hours a week.

Emergencies sit outside these numbers. They go to the on-call person by phone, as they do today. Test the routine share first, because a portfolio of older buildings produces more requests that need someone on site before anyone can classify them.

Starting small in 2026

Pick the workflow with the most weekly hours and the clearest rules. For many residential and commercial companies that is vendor certificates or maintenance intake, and for association managers it is often violation letters. Run a test on copies of real requests, with every output going to a review queue, and compare the results with what your team actually did.

Our five-day AI Workflow Sprint does that on your redacted documents for $2,500. It ends with a working prototype and a fixed quote for the build, based on the staff hours each workflow costs you, with a full refund if we do not find 10 staff hours a week. Either way, start with one workflow and a count of hours, and add the next one when the numbers support it.

This guide provides general information, not legal advice. Have counsel review your repair escalation rules and notice templates.

Common questions