AI for Claims Handling in Real Estate
A third floor flat floods on a Friday night. The site team isolates the riser, a drying company comes in over the weekend, and the asset manager first hears of it in a Monday helpdesk ticket headed "water ingress, unit 304". By then the carpet is in a skip, the photographs are on four different phones, and nobody has opened the policy schedule.
You are the claimant, not the handler
In real estate the organisation handling claims is almost never an insurer. It is an owner, a landlord, a fund or a managing agent, and it is the party bringing the claim. You are not adjudicating coverage or reserving against a book. You are assembling a claim, proving your own loss, and chasing somebody else's process.
Nobody is employed to do this. It lands on an asset manager or a facilities lead between rent reviews and service charge budgets. The counterparty has a claims department, a diary system and a reason to wait. You have an inbox.
Most claims software is built for the other side of that table, for a handler deciding whether to pay. This work needs document assembly, clause retrieval, date discipline and chasing.
Escape of water, a latent defect and a deposit dispute, logged the same way
An asset management team treats four unrelated things as one inbox.
Insurance claims on the building come first: escape of water, storm, flood, fire, impact, malicious damage. Schedules differ by asset, deductibles differ by peril, and a portfolio programme may be placed in layers with insurers at different attachment points.
Loss of rent and business interruption sit alongside them, where the recoverable period turns on lease wording as much as policy wording.
Latent defect and warranty claims run against the contractor, the structural warranty provider or the developer after practical completion, where the limitation clock and the handover file decide whether a claim exists.
Tenant damage, dilapidations and deposit disputes are the fourth kind, and here the counterparty is the occupier.
Four counterparties, four standards of proof, four clocks. An agent can classify a matter at intake and attach the right checklist.
The clock started when the site manager saw it
Property policies require notification immediately or as soon as reasonably practicable after the insured becomes aware of a loss, and awareness attaches to the organisation, not to whoever eventually opens the file. The real first notification is often a WhatsApp message to a facilities supervisor, an out of hours callout log, or a helpdesk ticket a tenant raised and somebody closed as maintenance.
An agent can read the helpdesk history and the site correspondence and date the first notification, quoting the sentence that proves it. Late notification is an argument an insurer may run, so you want to know before they do, and decide whether to explain the sequence up front.
Tracking the clocks across open matters
Notification windows, policy time bars, defects liability expiry, warranty limitation and deposit deadlines run quietly. An agent can hold every open matter against its clock and surface what expires next.
The schedule is a spreadsheet and the building names do not match
On a multi asset portfolio the insurance position lives in a spreadsheet, a broker's schedule and a wording nobody has read since placement. The claim names "Northgate House". The schedule says "Northgate House and Annexe, Market Street", insured in the name of a subsidiary sold last year.
An agent can match the damaged asset to the correct policy, peril and deductible, and show the schedule line it relied on. It can also flag the claims not worth making, where likely repair cost sits below the deductible, a decision often taken by accident.
The sum insured was accurate four years ago
Reinstatement cost assessments age badly, and where the declared value is stale an average condition can cut a settlement proportionally. An agent can list assets whose assessment predates a given date and compare declared values against indexation, as a prompt for a valuer.
A plumber's invoice that never says what caused it
The evidence for a building claim is rarely in a document system. It is photographs on a site manager's phone, an invoice reading "attend and repair leak, three hours, materials", a drying report with no cause narrative, and an inspection report describing a different unit.
An agent can build the file properly: every document labelled with its date, asset, unit and what it proves, ordered as a chronology. More valuable is the gap list. No photograph of the failed component. No statement of pre loss condition. No invoice narrative tying the repair to the peril claimed. Those gaps are nameable within a day, while they can still be filled.
The agent then drafts the notification and chase correspondence, policy number, deductible and schedule reference already correct, for a named human to send.
Loss of rent is decided by the lease as much as the policy
Whether rent keeps arriving turns on the rent suspension clause: which risks trigger it, whether it covers service charge and insurance rent or only principal rent, how long it runs, and whether it applies when premises are partly rather than wholly unusable. The policy then has its own indemnity period.
An agent can extract both, quote them side by side, and compute the period from the rent ledger and works programme. The gap between the two is the owner's exposure, worth knowing in week one rather than at settlement.
Latent defect claims live or die on the handover file
After practical completion a defect is a contractual question before it is an insurance one. The route might be the contractor under the building contract, a collateral warranty assigned on purchase, a structural warranty policy or a professional appointment, each with its own notification requirement and limitation period.
Whether a claim exists usually depends on handover documentation: as built drawings, operation and maintenance manuals, commissioning certificates and the snagging schedule. An agent can read that bundle against the warranty and appointment documents, identify who owes which obligation, confirm whether the benefit was assigned when the asset changed hands, and state the limitation date with the clause that sets it.
Dilapidations and deposits put you opposite your own occupier
Here the counterparty pays your rent. The evidence is a schedule of condition annexed to the lease years ago, a check in report, and photographs taken by somebody who has left.
An agent can compare the schedule of condition against the exit inspection item by item, and separate what the lease obliges the tenant to do from what a surveyor would like, which produces the supportable version of a claim. The decision about what to pursue against a tenant you want to keep stays with the person who owns the relationship.
Recharge decides who you end up arguing with
Every pound of damage cost sits somewhere: with the insurer, the tenant, the service charge or the owner. The lease decides whether the deductible is recoverable, whether insurance rent covers it, and what happens with an uninsured risk. Getting this wrong does not produce a mispriced claim. It produces a service charge challenge, credit notes and an occupier who reads every demand suspiciously.
An agent can prepare the recoverability view with the clause quoted, flag where a deductible is going through the service charge without a clear contractual basis, and keep the treatment consistent across units.
A tidy repair crew ends your recovery against a third party
Many building losses have somebody else at fault: the leaseholder whose appliance hose failed, the contractor who drilled through a pipe, the neighbouring site whose works caused movement. Recovery depends on evidence surviving the repair, and the failed component goes in a skip within days.
An agent monitoring works instructions against open claims can flag where a scheduled repair is about to remove the physical evidence, and draft the hold instruction: preserve the component, photograph it in situ with a scale, obtain the contractor's written account before the site is cleared. Subrogation is lost on timing, which software handles better than a busy person.
Repairs that run ahead of the adjuster's visit
Reinstating before inspection invites an argument about scope, and replacing like for better invites a betterment deduction. An agent can hold the repair sequence against the adjuster's appointment and record in advance which elements are being improved.
The reserve nobody has updated since March
Finance is carrying an accrual against each open claim, set early on partial information and rarely revisited, while the works complete and an interim payment lands.
An agent can keep claim status and reserve figure in step, report movements with the document that caused each one, and list matters where nothing has changed for a defined period. Stale claims are not usually disputed. They are forgotten.
Where AI fails here
Settlement acceptance is not a model's decision. Whether a figure is good enough involves the broker relationship, the next renewal, the cost of arguing and the owner's appetite, none of it in the file.
Dispute escalation is the same: instructing a loss assessor or solicitor, responding to a declinature, deciding to litigate against a contractor you still employ. A named person owns each.
Anything touching a tenant relationship stays human. Deposit deductions, dilapidations negotiation and recharge conversations involve a counterparty you will still have next year. An agent prepares the position. A person has the conversation.
Policy interpretation at the margin needs a broker or a lawyer. Non standard construction, condition precedent breaches and averaging are places where retrieval helps and conclusions mislead.
Thin files defeat it. With no reinstatement assessment, no handover bundle and a one line invoice, an agent can tell you there is little to work with but cannot manufacture a cause of loss. Photographs flatter in the same way: a wet floor with no scale, no date and nothing identifying the unit will be described fluently and prove nothing.
Dating the first notification from written records has its own failure mode, because the conversation may have happened verbally on site. Treat its date as the earliest provable one, not the true one.
Where to start
Take the peril that fills your inbox, usually escape of water, and nothing else. Baseline three numbers: days from incident to notification, days from notification to a complete file, and how many open claims have no deductible analysis. Then put an agent on intake, notification dating, policy matching and file assembly for that peril alone, inside the helpdesk and document store the team already uses.
Adopt AI Partners builds exactly that, claimant side agents that date the notification, match the asset to its policy, assemble the labelled file with its gaps named and draft the chases, inside the systems your team already uses. If you have a claim type that keeps slipping, describe the situation to hello@adoptaipartners.com and we will say whether it is worth doing.