PROPERTY & FM · SYSTEM

AI repair triage for managing agents and FM providers

IN SHORT

Every repair report read the moment it lands, classified by trade, urgency and lease responsibility, matched to the property and asset, and raised as a job with an SLA clock — with emergencies escalated by hard rule, not by model confidence.

The problem

Repairs are reported in whatever way the reporter finds easiest: a two-line email, a phone call typed into a note, a photograph with no address, a WhatsApp message to a manager's personal phone. Each one has to be read, understood, matched to a property and an asset, judged for urgency, checked against who is actually responsible under the lease, and raised as a job. That is five minutes of skilled attention per report, and a portfolio generates hundreds a month. The volume, not the difficulty, is what breaks the SLA.

How it works

Every inbound report is read regardless of channel. The system extracts what matters — property, unit, reporter, asset, described fault, any photograph — and classifies it by trade and urgency. It matches the address against your property records, checks lease responsibility where that is recorded, and raises a job with the right priority and target date. Anything it cannot match confidently, and anything touching a safety-critical category, goes to a person with everything it did understand already filled in.

Where the money is

Two places. The obvious one is manager hours returned from triage to actual management. The larger one is SLA performance: most breaches are not caused by contractors being slow, they are caused by jobs sitting unraised in an inbox for a day and a half. Removing that delay improves the number your client sees without changing anything about your supply chain.

What stays with your people

The judgement calls. Whether a borderline repair is recoverable through the service charge, whether a contractor should be sent to a difficult resident, whether a fault is a symptom of something structural. The system removes the sorting and the typing, not the decisions.

What you get

  • Reports read from email, portal, phone notes and photographs
  • Classified by trade, urgency and lease responsibility
  • Matched to property, unit and asset from your own records
  • Safety-critical categories escalated by hard rule, immediately
  • Job raised with target date and SLA clock, ready for dispatch

Questions we get asked

Can it read a photograph of a fault?
It can read what is visible — a leak, a broken pane, a damaged door, a fault display on a boiler — and use it to support the classification. It is treated as evidence attached to the job, not as a diagnosis, and it never overrides what the reporter said.
What happens if it misclassifies a repair?
Nothing is lost. The job is still raised, just with the wrong trade or priority, and correcting it is one click. The classifier is tuned on your own corrections over the first few weeks, and safety-critical routing never depends on it in the first place.
How long does it take to build?
Typically four to six weeks for a first version covering your highest-volume channels, then extended to the rest as it proves itself.

TRUST & CREDENTIALS

How we work, and what stands behind it

This page is maintained by Forth Systems. Everything listed below describes our own team, hosting choices and the technologies we build on — it is a statement of practice, not an independent certification or an endorsement by the vendors named.

UK team & data residency

Who builds it, and where your data sits while they do.

  • UK-based engineers

    Edinburgh-headquartered. Every system is designed, built and supported by our own UK-based engineering team — no offshore delivery pool.

  • UK / EU hosting by default

    Applications and databases are provisioned in UK or EU regions unless you require otherwise, with encryption in transit and at rest.

  • Human-in-the-loop by design

    Every AI system we ship keeps a manual override. Nothing irreversible happens without an operator able to review, edit or stop it.

  • Audit trails as standard

    Automated actions are written as first-class records — who or what did it, when, against which input — so operations stay evidence-ready.

Systems we build on

The platforms our systems run on. These are the technologies we deploy and support — not sponsorships or reseller badges.

  • OpenAI, Anthropic & Google models

    Model-agnostic delivery. We route each workload to the model that fits it on cost, latency and accuracy, and can switch without a rebuild.

  • Supabase & Postgres

    Row-level-security Postgres as the operational spine under every platform we run.

  • Stripe payments

    Payment capture, checkout and reconciliation wired directly into job and permit workflows.

  • Xero, Sage & HubSpot integrations

    Production integrations into the finance and CRM systems UK operators already run, so automation lands where the business keeps its records.

Streetworks & industry credentials

Operational domains we build in daily, with the sector systems we integrate against.

  • DfT Street Manager integration

    The permit chain we built for Kerbco creates and tracks streetworks permit applications against DfT Street Manager and local highway authority processes.

  • NRSWA-aligned workflows

    Permit, reinstatement and traffic-management workflows built to the New Roads and Street Works Act process contractors are inspected against.

  • Traffic management planning

    Roadplan produces signing and guarding layouts for council submission and for the crew installing them on site.

  • Operator-built, not consultancy-built

    Our platforms are built around real operating workflows — including the permit chain we built for our client Kerbco, a Scottish drop kerb contractor.

Insurance certificates, registration numbers and security documentation are available on request — ask the team.