Notices when the same fault keeps coming back, gathers what has already been spent on it, and puts the replacement case to the owner while it is still a choice.
An AI agent for repeat repairs watches the maintenance history of each property rather than each job in isolation: it recognizes when the same system has failed more than once inside a defined window, totals what has been spent on it, checks whether the last visit is still within a warranty period, and raises the repair-or-replace decision with the owner while it can still be made calmly. Maintenance is processed one job at a time, and that is exactly why this fails. Each individual repair is cheap and easy to approve. Nobody is summing them, so an owner approves a fourth visit to the same boiler without ever being shown that the four visits together cost more than the replacement he declined eighteen months ago.
Looks at the property's history instead of the job in front of it.
Recognizes the same system failing repeatedly within your window
Totals what has already been spent on that item across all visits
Checks whether the last repair is still under warranty before dispatching
Presents the repair-or-replace case with the history attached
Flags where the same contractor has returned to work they did
Records a refusal to replace, with a date, against the property
The pattern is familiar to everybody in property management and almost never acted on, because acting on it requires somebody to notice a sequence spread across two years of separate jobs. The financial argument is straightforward once the history is on one page: repeated visits to a failing item routinely exceed the cost of replacing it, before counting the emergency premiums that failures attract. The tenant argument is stronger and less often made. A resident whose heating has broken four times in one winter has stopped believing anything the agency tells them, and that is not recovered by fixing it a fifth time. There is also a warranty angle worth its own mention, since a surprising share of repeat visits are chargeable jobs raised on work that somebody else should have been putting right for free.
A three-step flow that runs when a new job is raised.
Before a repeat job is placed, the agent looks at what has been done to that system at that property and how recently.
It checks whether the previous repair should still be covered, and whether the frequency crosses the threshold your office has set for escalation.
Where the pattern is real, it assembles the dates, the faults and the spend, and raises repair-or-replace with the property manager and the owner.
A fourth heating failure at the same property in one winter.
Scenario: an office was processing repairs job by job with no view of history, and had several properties quietly absorbing repeat visits. A tenant reports no heating in early February. The agent checks the property before anything is placed and finds three previous heating jobs since October: a thermostat replacement, a pump repair, and a call-out in January by the same contractor who fitted the pump. That January visit falls inside the contractor's own warranty period on the pump, so the agent flags it rather than raising a fourth chargeable job. It assembles the four dates, the faults and the total spent, notes the boiler's age from the property record, and puts repair-or-replace to the property manager. She goes to the owner with a history on one page rather than a fourth quote in isolation. He approves replacement, and the tenant — who had by then complained twice about being cold — is told the same day that the boiler is being replaced rather than repaired again.
Anybody who would notice the pattern if they could see it.
You process jobs one at a time, which is precisely why patterns stay invisible.
A tenant losing patience with repeat failures becomes your problem quickly.
You approve each repair alone and never see what the item has cost in total.
Communal plant failing repeatedly affects every resident in the building.
Repeat repair data is the practical input to a capital replacement plan.
Repeat visits to the same home are a recognized driver of formal complaints.
Reads the history, checks the cover, frames the decision.
Holds the maintenance history per property, per system, with dates and cost.
Supplies what has actually been spent on the item across all visits.
Puts the repair-or-replace case to the owner with the history attached.
Holds previous engineer reports and warranty documents for the item.
Reports repeat-fault properties across the portfolio for planning.
Raises a suspected warranty case before another chargeable job is placed.
The patterns that only exist across jobs, never within one.
Questions about catching a pattern spread across separate jobs.
An AI agent for repeat repairs watches the maintenance history of each property rather than each job in isolation: it recognizes when the same system fails repeatedly, totals what has been spent, checks whether the last visit is still under warranty, and raises the repair-or-replace decision with the owner.
It depends on the system and the window. Two heating failures in one winter is a pattern; two tap washers in three years is not. Set thresholds per system type rather than one global rule, or the escalations will be either constant or useless.
No. It assembles the case — dates, faults, spend, item age — and puts it to the people whose decision it is. Replacement is capital expenditure on somebody else's asset, and the judgment involves tax position and plans for the property that no maintenance record contains.
By comparing the new fault against the previous repair's date, scope and warranty period on the job record. It flags a likely case rather than asserting one, because whether a fault is genuinely the same failure is a technical question. Even flagging catches jobs that would otherwise be paid twice.
Record it with the date and the case that was put, and carry on repairing. The record matters: when the item fails again in January, the conversation starts from a documented recommendation rather than from an argument about whether anybody ever raised it.
It needs jobs categorized well enough that the same system is recognizable across visits. Free-text descriptions alone will miss matches, because one job says boiler and another says no hot water. A system field on the work order is the single change that makes this work.
That is the most valuable output over time. Failure frequency by system and property age is real evidence for planning capital work, and it is considerably better than the usual basis, which is replacing things in the order they break most inconveniently.
Notices when the same fault keeps coming back, gathers what has already been spent on it, and puts the replacement case to the owner while it is still a choice.