24/7 Tenant Query Agent

A 24/7 tenant query agent is an LLM system that answers resident questions about lease terms, rent and fees, and building services by reasoning against the specific lease and property records for that unit. It does not invent policy from a generic FAQ. It retrieves the clauses, dates, amounts, and service rules that apply to this tenancy, drafts a clear reply, and routes anything ambiguous or high-stakes to staff.

Property managers hear the same categories of questions at all hours: When is rent due? What does the pet clause allow? How do I book the parking garage? Is the gym open during renovation? After-hours and weekend volume often sits in email or a portal inbox until the next business day. An agent that can resolve the routine share of that queue keeps response times predictable without staffing a night desk for every portfolio.

Commercial products in this category (for example BrickwiseAI, STAN.AI, and Entrata ELI) illustrate the same pattern: conversational front end, connection to property and lease data, and escalation when the model cannot ground an answer safely. The value for a practitioner is operational, not novelty. Tenants get accurate answers faster. Staff spend time on exceptions instead of restating the same lease language.

Related reading: Move-In Condition Documentation covers how condition records support disputes later in the occupancy lifecycle. Query agents answer policy and process questions; they do not replace documented condition baselines.

Handle when the answer is fully determined by structured property data or an explicit lease clause the agent can cite:

• Rent due dates, grace periods, accepted payment methods, and posted balances as shown in the ledger
• Standard building hours, amenity booking rules, package-room procedures, and published maintenance request channels
• Pet, parking, subletting, and alteration clauses when the lease text is complete and unambiguous for that unit
• Contact paths for utilities, insurance proof, and move-out notice windows when those terms appear in the record

Escalate (or hold for human send) when the reply would change money, access, or legal position, or when grounding is incomplete:

• Rent reductions, fee waivers, payment plans, or any promise about what is owed
• Lockouts, key fobs, visitor access overrides, or temporary entry permissions
• Lease amendments, early termination, roommate substitutions, or notices that start formal timelines
• Habitability disputes, alleged discrimination, or anything that belongs with counsel or senior management
• Conflicts between the lease PDF and the PMS fields (amounts, dates, or named parties do not match)

Human-in-the-loop means the agent drafts. Staff still review and send anything that alters amounts due, physical or system access, or the legal position of landlord or tenant. For routine informational replies that only restate published terms, many portfolios allow auto-send after the agent attaches the source clause or field. Policy should say which path applies, not leave it to model confidence alone.

Grounding starts with identity. The channel (portal, authenticated chat, or verified phone/email) must map the speaker to a unit and lease. Anonymous lobby chat can only answer building-wide public facts. Unit-specific lease terms require a verified tenancy link.

Once identity is known, the agent pulls a bounded context pack: active lease PDF or structured clause extract, current ledger summary, property handbook sections tagged for that building, and open work-order or amenity statuses if those systems are connected. The model is instructed to answer only from that pack. If a needed document or field is missing, empty, or unreadable, the correct behavior is empty output for that question: no speculative paraphrase of "typical" leases, no filler apology that invents a policy. The tenant sees a clear handoff ("I do not have the lease record needed to answer this; a team member will follow up") and the ticket opens with the data gap flagged.

Reasoning should prefer citation over summary when stakes are higher. "Per Section 4.2 of your lease, rent is due on the first of each month" is safer than a free-form restatement that drifts from the clause. For ledger questions, quote the system fields (balance, due date, last payment) rather than narrating implications about collections.

Conflicts inside the context pack (handbook says one amenity hour, lease exhibits say another) are escalations, not creative merges. The agent should surface both snippets to staff instead of picking a winner.

Building-service questions sometimes overlap energy and ESG programs (thermostat guidelines, recycling rules, EV charging). Pair query handling with ESG & Energy Consumption Monitor when published building rules and meter-backed programs need a single source of truth.

Treat the agent as a queue layer on top of your PMS and document store, not as a separate knowledge base you maintain by hand.

1. Define the auto-send allowlist. Informational restatements of lease clauses and published handbook items can auto-send when grounding succeeds. Money, access, and legal drafts always sit in a human outbox.
2. Require empty output on missing records. Incomplete lease upload, expired lease without renewal attached, or unit not linked to a tenant profile must produce no substantive answer. Measure "empty/handoff" rate as a data-quality KPI, not only "deflection."
3. Staff the exception lane. Edge cases still need people who can interpret gray clauses and approve exceptions. The agent shrinks volume; it does not remove judgment.
4. Log every draft with sources. Keep the retrieved clause IDs, PMS field snapshots, and model draft for audit when a tenant later disputes what they were told.
5. Refresh after lease events. Renewals, amendments, and mid-term fee schedule changes must re-index before the agent serves that unit again.
6. Align channels. Portal chat, email auto-replies, and phone IVR handoffs should share the same grounding rules so tenants do not get conflicting answers by channel.

Space planning questions from commercial or flexible assets ("Is the lounge bookable for my team on Thursday?") sit next to residential FAQ volume in mixed portfolios. Space Utilization Analytics informs capacity and booking policy; the query agent only communicates the rules those analytics and ops teams have already published.

Quality for this outcome is whether answers match the lease and property record, not how chatty the agent sounds.

Useful checks:

• Grounded accuracy: Sampled replies match cited clause or field (spot audits weekly at first, then monthly).
• Empty-output correctness: When lease or property record is missing, the agent must not invent terms (track false substantive answers on incomplete packs).
• Escalation precision: Money, access, and legal drafts reach humans; pure FAQ items do not clog the outbox.
• Time to first substantive reply: Median minutes for allowlisted auto-sends vs human-approved sends.
• Reopen rate: Tenant comes back saying the answer conflicted with the lease PDF or a prior staff email.

Common failure modes:

• Serving a prior lease version after renewal
• Answering from a sister building's handbook
• Softening a hard clause into friendly language that changes meaning
• Auto-sending a fee waiver "clarification" that reads as approval
• Filling silence when OCR failed on a scanned lease instead of returning empty output

Mitigations are operational: versioned document IDs, building-scoped retrieval filters, mandatory citation blocks, and hard policy gates before send. Model choice matters less than refusing to answer without the right pack.

Start with one building or one asset class where leases are consistently digitized and the PMS is trustworthy. Publish the allowlist and escalation matrix to the on-site and central teams before flipping the channel on. Run a shadow period where the agent drafts and staff send everything, then expand auto-send only for categories that pass audit. Keep empty-output behavior non-negotiable: no lease, no answer. Expand to more buildings only after reopen rate and grounding audits stay acceptable, and after related occupancy systems (condition documentation, energy rules, space booking) agree on the same published facts the agent will quote.