Main site

Real Estate AI agent · Maintenance triage

Maintenance-Triage & Work-Order AI Agent

Take maintenance intake verbatim, triage it against the repair clock configured for that jurisdiction, and hold dispatch, closure and lead-regulated scope for the licensed property manager, who decides.

Typical delivery
4–6 weeks
Deployment
Your stack
Human dispatch
Triage only
After launch
Agent Care

What this agent does

Triages the queue, not the life-safety call

In

  1. 01

    Collect the report verbatim, with photos, access notes and pets on site, from the portal, phone and SMS channels.

  2. 02

    Collect the unit's construction-year and vendor records from the systems of record, each carried with its source.

Reason

  1. 03

    Classify urgency against the repair clock and heating-season rules configured for that property's jurisdiction.

  2. 04

    Classify any life-safety trigger straight to the on-call supervisor, without waiting for a second signal.

  3. 05

    Classify a pre-1978 or unsourced-year unit as lead-regulated scope, which only a certified firm may narrow.

Decide

  1. 06

    Dispatch against a vendor record re-checked for licence, insurance and scope class at dispatch time.

  2. 07

    Dispatch nothing a human has not approved, and leave the order open until that human closes it.

Out

  1. 08

    Retain the verbatim text, the photos, the notice sent, the vendor checks and the approving human.

  2. 09

    Execute write actions only inside the approval boundaries agreed during implementation.

Product statement

The agent triages and prepares the order; the licensed property manager decides what dispatches, and the repair duty stays the owner's.

Example workflow

One ticket, intake to dispatch

  1. 1Report receivedPortal, phone, SMS or on-site report from the tenant
  2. 2Record assembledVerbatim text, photos, unit history, construction-year source and vendor status
  3. 3Urgency proposedCategory, proposed route, the clock that applies and confidence
  4. 4Controls appliedLife-safety trigger list, lead screening, vendor re-checks, entry-notice rules and confidence threshold

No human action required

Stages 1 to 4 run unaided, and nothing is dispatched at any of them — the agent is preparing the order, and the manager's lane opens at the confidence gate.

5DecisionSplits at the confidence threshold
High confidence

Goes to the property manager to approve.

Low confidence

Adds an on-call supervisor read first.

Manager approval

The order is held with its verbatim text, its triggers and the confidence.

Approve dispatch · Reprioritise · Send to on-call

Approved — released to dispatch

  1. 6Work-order system updatedOnly where write access and approval policy allow it
  2. 7Outcome evaluatedReclassification rate, clock adherence, vendor checks and repeat reports

Reclassifications

Reclassifications made in review are counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Determining whether the RRP rule applies to a scope.
    • Dispatching paint-disturbing work in pre-1978 housing.
    • Classifying or reclassifying a life-safety condition.
    • Approving a vendor whose licence or insurance is unverified.

  • Automation boundary

    Agent acts unaided

    • ✓Collect structured intake — unit, access, pets, verbatim text, photos.
    • ✓Confirm a window a human approved with a verified vendor.
    • ✓Send status updates from the system of record.
    • ✓Prompt the tenant to confirm the work once it is marked complete.

    Every dispatch waits for the approval boundaries agreed at implementation to allow it.

    • Closing, deferring or cancelling a habitability work order.
    • Authorising entry, or invoking the emergency exception.
    • Deciding an accommodation inside a repair request.
  • Advising on repair-and-deduct or rent withholding.

Example output

One ticket in the queue, annotated

Everything the agent proposes is attached to the report it was drawn from.

Triage output · single work order

Illustrative example
Unit
Garden duplex, unit B
Reported condition
Ceiling stain spreading in the back bedroom, paint now flaking
Year built
1962
Proposed route
Certified-firm queue
Confidence
88%
Approval held for
Licensed property manager

As received

Taken from the tenant's report and the property record — nothing on this side is written by the agent.

Signals used

  • Verbatim tenant report
  • Photos attached at intake
  • Unsourced year record

Why this route

The year-built source is unsourced, so the scope routes as lead-regulated.

What the score decides

Below the threshold the ticket picks up an on-call read before the manager sees it.

Action

Approve dispatchReprioritiseSend to on-call

Value

Where AI adds value

The same four claims, placed at the point in the workflow where each one applies.

Where the value lands

Value 01 – 04

All tickets

From the intake channels

03Triage

Triage against the record

Draw on the unit's history, the construction-year source and the jurisdiction's clock.

01Approved path

Get emergencies to a person faster

Routine repairs arrive scoped, with the intake already structured.

02Human review

Send the urgent ones to a human first

Life-safety triggers and low-confidence tickets are marked, so the on-call read starts where harm concentrates.

04

Build an evidence trail

The intake text, the photos, the triage class and the dispatch record stay with the ticket.

Integrations

Typical integrations

Five system groups connect to the same agent. Which of them are in scope is decided in discovery.

  • Maintenance coordination

    • Property Meld
    • Latchel
    • Lessen
    • Lula vendor networks

  • Work orders in the PM core

    • Yardi Maintenance
    • Voyager
    • AppFolio Property Manager

  • Inspections and condition

    • HappyCo
    • zInspector
    • Unit condition records

Agent

Maintenance triage & work orders

  • Reads the report
  • Proposes the route
  • Holds for dispatch

  • Vendor compliance and AP

    • AvidXchange
    • Nexus
    • Licence and COI records

  • Observability & evaluation

    • OpenTelemetry
    • Langfuse
    • Supported monitoring/evaluation sources

Integration availability depends on the client's existing systems and API access.

Agent controls

Six layers between the model and the dispatch

Each control wraps the one inside it. What each one leaves uncaught is named in the failure-mode map below.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modeNarrow the agent to intake only when evaluation or production signals degrade.Roll back
L5Version monitoringTrack model, prompt, SLA-table and jurisdiction-configuration changes.Track
L4TraceabilityRecord the verbatim report, the triage, the checks, the notice and the approver.Record
L3Manager approvalHold orders for the named manager; it governs dispatch, not whether the triage was right.Gate
L2Policy guardrailsTest each order against the life-safety list, lead screening and vendor status; a failure returns it.Restrict
L1Confidence thresholdsRoute low-confidence tickets to an on-call read before the manager sees them.Require review
Model coreOrder prepared — category, proposed route, the clock that applies and confidence
L1 – L2
Check whether the triage may stand
L3
Puts the dispatch in a manager's hands
L4 – L5
Keep the record and the version honest
L6
Narrows the agent when signals degrade

How Nestack evaluates it

Evaluate the triage workflow — not only the finished work order.

Coverage runs the whole depth of the workflow, and every layer is cut by slice.

Surface — the repair the tenant gets

  1. E1Final-output evaluationWas the urgency the agent proposed the one a human kept?
  2. E2Step-level evaluationDid the agent read the right unit, clock and vendor status?
  3. E3Tool evaluationDid it write to the correct work order and the correct field?
  4. E4Confidence calibrationDo low-confidence tickets actually attract more reclassification?
  5. E5Slice evaluationHow does performance change across specific property cohorts?
  6. E6Business outcomeHow many tickets were reclassified or re-opened after closure?

Floor — the outcome the firm answers for

Failure modes

Where each failure originates in the agent

Seven failure modes, each shown where it starts, not where it lands.

Agent lifecycle

Direction of processing →

  1. 01 · Retrieval

    1 mode

    • WO-01

      Unsourced year-built read

      A pre-1978 unit clears as post-1978.

    Stage gathers

    Report, unit history, year-built and vendor status

  2. 02 · Reasoning

    2 modes

    • WO-02

      Life-safety signal missed

      A gas or CO report is triaged as an appliance nuisance.

    • WO-03

      Intrusion read as cosmetic

      A spreading stain is scoped as paint, not water.

    Stage proposes

    Category, proposed route, the clock and confidence

  3. 03 · Tool / write

    2 modes

    • WO-04

      Lapsed vendor record

      A cached licence or certificate expired after onboarding.

    • WO-05

      Auto-close on no access

      The statutory repair clock stops while the defect persists.

    Stage writes

    Only where write access and approval policy allow it

  4. 04 · Output

    1 mode

    • WO-06

      Notice to a stale address

      The vendor arrives without lawful notice to the occupant.

    Stage returns

    The order the manager approves and the tenant gets

  5. 05 · Change / Version

    1 mode

    • WO-07

      Stale SLA constant

      A state amends its window; the hard-coded table does not.

    Stage tracks

    Model, prompt, SLA table and jurisdiction config

  • Sev-1 · dispatched outside the boundary
  • Sev-2 · an emergency is triaged as routine
  • Sev-3 · source degrades, ticket routes to review
See the complete Real Estate failure-mode catalogue

Affected slices

One cohort can carry most of the mis-triage

Pre-1978 units with a child under six absorb most of the mis-triage long before the aggregate rate moves. Nestack reports the emergency mis-triage rate by slice, not only in total.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Pre-1978 units, child under 65.4%2.7×Review
Accommodation-adjacent4.2%2.1×Review
After-hours heating season2.4%1.2×Watch
Scattered-site single-family1.0%0.5×Normal

Bar: emergency mis-triage-rate lift vs. the all-ticket baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

A cycle is not closed until a test exists

Nothing closes until the failure is a regression case the next release has to clear. The suite it joins is what the next detection is measured against.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Emergency mis-triage rate rises in a property cohort.

  2. 02

    Diagnose

    The ticket, its photos and the record behind it are read together until one cause remains.

  3. 03

    Improve

    The fix lands against a version, with the reclassifications that prompted it attached.

  4. 04

    Verify

    Nothing releases until the affected regression cases pass again.

  5. 05

    Learn

    The case joins the permanent suite and the firm's triage rules.

  6. Learn → Detect

    The return edge. The next detection runs against a suite that is one case longer.

Typical build scope

Twelve workstreams across six weeks

The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, sources, triage workflow, evaluation, integration, then production validation and handover.

Workstream

Week 1Week 2Week 3Week 4Week 5Week 6

  1. 01Maintenance intake discovery and boundary definition.Week 1
  2. 02Work-order and vendor source assessment.Week 1, Week 2
  3. 03Jurisdiction clock, lead-screening and trigger-list mapping.Week 1, Week 2
  4. 04Ticket ingestion and verbatim-record capture.Week 2
  5. 05Triage logic and property-record binding.Week 2, Week 3
  6. 06Confidence scoring and escalation routing.Week 3
  7. 07Manager approval workflow.Week 3, Week 4
  8. 08Work-order and vendor-system integration.Week 5
  9. 09Regression cases and the evaluation suite.Week 4, Week 5
  10. 10Guardrails and dispatch controls.Week 4
  11. 11Trace instrumentation and observability.Week 5, Week 6
  12. 12Deployment, documentation and Agent Care handover.Week 6

12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel

Final scope and sequence confirmed in discovery

Engagement tiers

What each tier includes

Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.

Capability✓ in scope · — not at this tierPilotOne portfolio, one marketProductionProduction work-order systemsAdvancedScattered-site portfolios
Introduced at Pilot
Triage to your clocks and rules✓✓✓
Manager approval✓✓✓
Triage-accuracy baseline✓✓✓
Introduced at Production
Triage reporting by property type—✓✓
Dispatch approval in your systems—✓✓
Approved write-back—✓✓
Dispatch and vendor logging—✓✓
Introduced at Advanced
Multi-state clock and lead rules——✓
Multi-stage dispatch approvals——✓
High ticket volume——✓
Enterprise vendor controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build price
Confirmed after discovery based on integrations, workflow complexity, transaction volume, approval controls and deployment requirements.
Separate from build
Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.

What we need from you

What you bring, and what we build with it

Each input maps to a piece of build scope and a week in the delivery timeline.

  1. 01Your intake channels and ticket fields

    Ticket ingestion and verbatim-record capture

    Week 1

  2. 02Representative closed work orders

    Triage baseline, trigger-list extraction and record binding

    Week 2

  3. 03Your emergency list and jurisdiction clocks

    Clock, lead-screening and trigger-list mapping

    Week 1

  4. 04Access to relevant APIs, feeds or exports

    Work-order, vendor and inspection assessment, then integration setup

    Week 2

  5. 05Tickets that were reclassified after closure

    Regression cases, evaluation suite and failure-mode testing

    Week 4

  6. 06Your dispatch thresholds and write-back limits

    Confidence scoring, escalation routing, guardrails and dispatch controls

    Week 3

  7. 07Named managers and on-call supervisors

    Manager approval workflow, then pilot and production validation

    Weeks 5–6

Nothing else is required

Deployment, documentation and Agent Care handover are ours.

Delivery timeline

Four phases across six weeks

Phases are drawn over the weeks they occupy. Week 5 carries evaluation and pilot together.

Phase

W1

W2

W3

W4

W5

W6

  1. DiscoveryW1
  2. BuildW2 – W3
  3. EvaluateW4 – W5
  4. Pilot & LaunchW5 – W6

Week focus

W1Intake discovery, clock mapping and the automation boundary

W2Source integration and the triage baseline

W3Triage workflow, confidence logic and dispatch controls

W4Evaluation suite, lead-screening checks and failure-mode testing

W5Vendor-system integration, pilot tickets and targeted corrections

W6Production validation on live tickets, then Agent Care handover

Reading the band
A bar covers only the weeks its work is named in. The week 5 overlap is real, not padding.
At the end of W6
Everything verified in production, with Agent Care carrying it forward.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Real Estate AI agent

Build a maintenance-triage agent around your on-call chain.

Show us your intake channels, your vendor bench and who is on call. If your after-hours tickets wait for a human until the morning, that is where we set the automation boundary.

Typical delivery
4–6 weeks
Deployment
Your maintenance stack
Human dispatch
Triage only
After launch
Agent Care available

Nestack Agents · Maintenance triage & work orders

AGT-RE-08 · Agent Care available after launch

Build a maintenance-triage agent

Four questions so the call starts from your actual setup — then pick a time that suits you.

About you
Your queue

No obligation. We’ll map the workflow, name the automation boundary and tell you if an agent is the wrong answer.