AI and Automation in Community Management

The test for whether a job belongs to automation or AI is simple. Does the job have one correct answer that follows from a rule you could write down? If yes, automate it. If the job requires judgment, phrasing, or interpretation, AI can give you a first pass, and you finish it.

Notifying the right person when a plumbing request comes in is a rule. Explaining a special assessment to 240 owners without starting a fight is judgment.

Both halves of that test assume your data can answer it, and that assumption is where this usually falls apart. If every maintenance request in your system is filed under "General," no report will show you what is breaking and no model will find a pattern in it. Categories, unit profiles, and a current equipment list are unglamorous work, and they decide whether the rest is worth doing.

Triage is four jobs: take the request in, categorize it, get it to the right person, and keep the resident informed so they stop calling the desk. Vendors sell AI for the second job. The other three have been automatable for years, and they are where the hours go.

Intake and notification are the two you can fix this week. In BuildingLink, residents and staff submit maintenance requests online, and the notification settings decide who hears about each request. You choose which events trigger an email to management and which trigger an email to residents. Our support documentation recommends, at a minimum, notifying the resident when a request is opened and when it is closed.

One setting worth checking notifies residents when building staff post a comment. Turn it on and the update reaches the resident before they think to ask, which can take the "any update on my leak?" call off the desk.

Categories take longer and return more. They decide what your maintenance reports can tell you six months from now, and you can set notification options on an individual category that override your defaults. BuildingLink includes default categories such as electrical, plumbing, elevators, doors and locks, and appliances. Reports are only as useful as the categories behind them, so treat the default list as a starting point.

A general-purpose AI tool can help in two places, both small and both outside your platform. It can compress a long comment thread into a paragraph for a board update, and it can read a resident's long description and suggest a category. Remove names and unit numbers before you paste anything, and treat the suggestion as a suggestion. The category shapes your reports, so a person confirms it.

Triage settings worth checking this week:

• Turn on resident notification for request opened, request closed, and staff comment posted.
• Review your maintenance categories and add the five that describe your actual work orders.
• Set up the assigned employee list so you can assign requests when you enter them, not after.
• Write predefined maintenance responses for your three most common replies.

General-purpose AI is good at drafting, and it is the use case most managers land on first. A first draft of an outage notice, a plain-English version of a vendor's technical explanation, a meeting recap, a newsletter section that does not sound like a memo. AI is fast at all of it, and the quality is fine once you edit.

The rule is that AI drafts and a person sends. Do not wire a model directly into a resident-facing channel. The moment a written answer goes out under the community's name, it becomes the community's position, and you cannot take it back by explaining that a model wrote it.

Two hard limits apply.

• Do not paste resident personal information into a consumer AI tool. That includes names, unit numbers, account details, and medical accommodations. Write the prompt about the situation, not the person.
• Do not let AI answer questions about your governing documents. Researchers at Stanford tested three AI legal research tools built by LexisNexis and Thomson Reuters, which run on curated legal databases, and found that each hallucinated more than 17% of the time. The study was published in the Journal of Empirical Legal Studies in 2025. Those are purpose-built tools on clean data. We have not found an equivalent published test of a general chatbot reading a set of governing documents, which is a reason for caution rather than confidence.

Both limits are about what AI says. The bigger win in communications is about who hears it, and that half is automation. Filtered groups let you build rules on unit location, occupant type, and custom fields, so a notice about the north stack goes to the north stack instead of all 400 units. Fewer, more relevant sends are the part of email performance you control without buying anything. You also have announcements, email and SMS broadcasting, and emergency broadcast by voice message or SMS for the messages that cannot wait.

Here is a prompt that works:

"Write a notice to residents of a 240-unit condominium. The elevator in the north tower will be out of service for modernization from September 8 to September 26. There is one working elevator in that tower. Move-ins are suspended for the period. Tone: direct and apologetic without over-apologizing. Under 200 words. No exclamation points. Do not promise a completion date earlier than September 26."

Then edit it. The draft is a starting point, not something to send as is.

Your platform already holds the numbers. BuildingLink's reports and data section pulls reports across management, maintenance, amenity reservations, and occupants, and the analytics view shows site activity for the last 90 days, including logins and items entered. You can download unit profile data, occupant data, employee data, maintenance requests, maintenance request actions, the event log, front desk instructions, and survey results to Excel.

Nothing on that list is a number you are missing. The gap is that nobody has time to sit with a 12-month export and ask it questions, and that is a job AI does well.

Export your maintenance requests for the last 12 months. Strip the unit numbers and resident names, and check the description fields for names too. Hand the spreadsheet to a general-purpose AI tool and ask what is repeating. This is a low-risk use because the data is yours, the output is a hypothesis rather than a decision, and you can check every claim against the source file.

What comes back is the kind of pattern you half-suspected and never proved. A handful of lines producing most of the plumbing volume. A day of the week when requests sit longer before anyone touches them. A vendor whose average time to resolution is worse than the one you replaced. None of those are conclusions until you check them, and all of them are worth checking.

That is the material for a board packet. A board that sees a total request count learns nothing. A board that sees which line produced the plumbing volume, over what period, and how old the riser is can fund something.

Four questions your maintenance export can answer:

• Which categories produce the most requests, and which produce the most reopened requests? Those are different problems.
• Which units or stacks are repeat offenders? Check a repeat as a system problem before you treat it as a unit problem.
• What is your average time to close by category, and where did it get worse this year?
• How much of your volume is preventable work that should be on a schedule instead of a request? That last question is the one that pays for itself.

If you do one thing after reading our guide, do this one. The setup is a one-time configuration, and the savings are documented. The U.S. Department of Energy's Operations and Maintenance Best Practices Guide, prepared by Pacific Northwest National Laboratory (Release 3.0, 2010), says studies indicate a preventive maintenance program can save as much as 12% to 18% on average compared with running equipment to failure. That federal guide covers federal facilities, so treat that figure as directional for a residential community rather than a promise.

Preventive work splits into two jobs: the tasks you can put on a calendar, and the problems you only find by walking the building.

Recurring tasks handle the first. You enter a preventive task once and set the interval, and BuildingLink automatically generates a new maintenance request every time it comes due. Intervals can be specific days of the week or specific months of the year, so seasonal work lands when it should. A seasonal bypass stops a task from being created in months when it should not run. You can have staff members receive an email when a new task is generated, and you can have a task email an outside party such as a vendor with instructions included, so a quarterly filter contract can run off the same schedule. See how this works in our preventative maintenance software.

Link each task to an item in your equipment directory, and the requests attach to the asset, not just the calendar. A report counts maintenance requests per equipment item over a date range, which is how you show a board that the boiler needs attention more often instead of just asserting it. Record the installation date, useful life, and replacement cost on each item, and the equipment replacement report calculates replacement dates and projects replacement costs by year, which gives the reserve conversation a number to work from.

Inspections handle the second job. Build custom checklists for equipment checks, unit walkthroughs, or common area cleaning. Schedule them daily, weekly, monthly, or yearly, and assign them to a staff member or leave them open for anyone to claim. Staff complete them in the field on the GEO by BuildingLink mobile app with photos and notes. When something fails, they enter a maintenance request from the checklist item, and the note carries into the request description automatically. Completed inspections download as PDFs, and move-in and move-out inspections can capture a signature through an event type in the GEO app. Learn more about our inspections module.

That is a closed loop with no AI in it anywhere. Rules solved scheduling years ago, and waiting for a smarter version risks equipment that fails without warning.

Recurring tasks worth setting up before winter. Use the specific months option for quarterly, twice-yearly, and annual timing, and confirm every interval against your local code and vendor contracts.

Every section above had a yes in it. This one is the no.

• Skip AI that talks to residents unsupervised. A chatbot answering rules questions can give a confident wrong answer in writing. That answer is now in a resident's inbox with your community's name on it, and a board will not accept "the software said it" as an explanation.
• Skip any AI-assisted screening decision you cannot explain in plain English. This one has moving parts right now, so know where things stand. HUD withdrew its April 29, 2024 guidance on fair housing and digital advertising, effective September 17, 2025, in a notice the Federal Register published April 6, 2026. The companion guidance on screening applicants for rental housing was not on that withdrawal list. More to the point, the Fair Housing Act itself did not change, and HUD's notice states that actions that do not comply with the Act remain subject to enforcement. If a vendor's model produces a decision nobody at your community can explain in plain English, that is a question for your counsel before it is a question for your vendor.
• Skip AI as a purchase reason. The label is on everything now and means very little on its own. Ask the five questions below. A vendor who can answer all of them is worth listening to.
• Skip uploading resident data to consumer AI tools. Aggregate and anonymize before anything leaves your systems. This is a habit, not a policy document, and it is easier to build now than to retrofit after something goes out.
• Skip AI-written minutes as the official record. Transcription is fine, and a draft summary saves real time. The approved minutes are a governance document that can end up in litigation or a resale package, so a person writes the final version and the board approves it.
• Skip predictive maintenance pitches for a building with no sensors. Prediction needs instrumented history. If your equipment is not reporting anything and your maintenance records live in three categories, the model has nothing to learn from. Fix the record first. That is the same clean-data point from section one, and it is the reason to do it this year.

Five questions for any AI vendor:

1. What does the model do here that a configurable rule could not?
2. What happens when it is wrong, and how would we find out?
3. Who reviews the output before a resident sees it?
4. Where does our data go, who can see it, and is it used to train anything?
5. Can you show me the same feature working in a community our size?

Automate anything that repeats on a schedule or a trigger, because it is reliable and auditable. Use AI where a draft saves you real time and a person still signs off. Keep a person on anything that becomes the community's official position.

Our bet is that the communities getting the most value from AI will be the ones that cleaned up their data first: categories that describe the actual work, unit profiles that are current, and an equipment list that includes install dates. That is the boring part, and it decides whether any of this pays off.