When no one picks up, the ticket never exists.
An AI voice agent platform that transforms how facility management companies capture, track, and resolve resident maintenance requests — from chaotic WhatsApp threads to a structured, self-operating ticket system.
The stakes
In traditional facility management, an estimated 35–40% of maintenance requests are never logged — not because residents don't report them, but because there's no reliable system to receive them. Calls go unanswered. WhatsApp messages get buried. Spreadsheets lag by days.
For FM companies managing 10–100 properties, this isn't a workflow inconvenience — it's a direct hit to resident retention. Residents are 3× more likely to decline lease renewal after two or more unresolved maintenance requests. A facility management company came to Thirtydays Studio needing more than a ticketing tool. Their ops team was drowning in calls, WhatsApp threads, and missed handoffs.
The problem
Mid-sized FM companies fail to capture resident maintenance requests reliably because intake happens across fragmented, unmonitored channels — phone calls, WhatsApp, walk-ins — with no structured data collection, resulting in an estimated 35–40% of requests being missed or unresolved within SLA, directly increasing resident churn.
What we found
Stakeholder workshops
Ops coordinators spent 2–3 hours per day manually logging calls, following up on WhatsApp, and chasing service personnel. No structured intake meant every request depended on one person's memory or inbox.
Competitive audit
Tools like Buildium, Facilio, and ServiceMax are built for enterprise property groups — heavy ERP-style systems requiring weeks of onboarding. None solve the front-door problem: getting a request into the system in the first place.
Heuristic audit
The highest failure rate was at intake — the moment a resident tries to report something. Phone calls weren't answered. Form submissions were never seen. WhatsApp went to personal phones, not shared ops inboxes.
Options considered
Option A — Resident web form + email intake
RejectedResidents don't open web portals for maintenance. Adoption rates for tenant-facing forms in residential FM are below 20%. The channel mismatch remains — residents call, not click.
Option B — WhatsApp chatbot intake
RejectedResidents call — not text — for maintenance. WhatsApp also requires personal phone numbers, creating data ownership risk. A text bot doesn't solve after-hours or urgent call escalation.
Option C — AI voice agent + ticket CMS platform
ChosenMatches the channel residents already use (phone). Eliminates human dependency at intake. Scales to unlimited properties without adding headcount.
Tradeoffs
English-only in v1
We chose English-only over full multilingual support. Residents who prefer Arabic, Hindi, or other regional languages — an estimated 15–25% in Gulf FM markets — must communicate in English. We accepted this because the primary buyer is an English-operating FM company. Shipping a working product in one language beats shipping a broken product in five. Multilingual support is the top backlog priority.
Mandatory sequential wizard, no shortcut setup
We chose a mandatory wizard over a freeform settings interface. Technical users who want to jump to a specific setting can't bypass it on first setup — adding roughly 3 minutes to initial onboarding. We accepted this because stakeholder workshops showed admins consistently skipped critical config steps when given freeform access, particularly escalation contacts and business hours. A sequential wizard prevents a live agent with no emergency escalation path.
Local numbers only, no toll-free
We chose local number provisioning only. FM companies managing properties across multiple countries cannot use one centralised number. We accepted this because the core buyer is a single-market FM operator — and one number per property is also a trust feature. Residents see their building's number, not a corporate switchboard. Multi-country enterprise is a v2 segment.
What we built
Agent creation wizard
Six sequential steps with persistent progress state. Each step has a single clear action — no multitasking, no optional order. Step order was not arbitrary: the knowledge base comes before request handling rules because the KB informs when the agent should escalate. Changing this order in early testing caused admins to leave escalation contacts blank.
Urgency detection UI
Admins see default urgent keywords pre-loaded with a clear input to add property-specific terms. The system logic — bypass queue, immediate escalation — is surfaced in plain language, not buried in settings. The single most catastrophic failure mode is a flooding call that gets ticketed normally and waits in a queue. Making the urgency system visible was a direct response to that risk.
Ticket queue with status filters
Six ticket states map exactly to the operational lifecycle. Status badges are colour-coded and filterable by property, priority, and date. Every status has a direct owner — Voice Agent, Admin, or Service Person. This removes the ambiguity that caused tickets to stall. "Who's responsible right now?" is always visually answerable.
Service person mobile journey
A simplified 4-state flow: receive → acknowledge → complete → submit proof. Photo upload is the primary action on the completion screen — not buried inside a form. The entire mobile journey is optimised for a technician in the field with one hand occupied. No unnecessary taps.
Outcome
The ~40% improvement is estimated against the industry benchmark of 35–40% miss rate in non-digital FM intake versus near-zero miss rate for AI-handled inbound calls at 99.9% agent uptime. No post-launch data was available at the time of writing — these figures represent expected directional improvement and will be validated with post-launch instrumentation.
What's next
Multilingual support
Arabic and Hindi as v2 languages. The STT/TTS pipeline already supports multilingual models. Accent preference data from wizard setup will indicate demand by market before full development investment.
Predictive ticket tagging
As ticket volume grows, LLM-based classification can surface patterns — units with recurring plumbing complaints, buildings with seasonal AC request spikes. This shifts the product from reactive to proactive. The infrastructure is already in the data model.
Resident satisfaction loop
A post-closure SMS survey — one question, 10-second response — tied to ticket close events. CSAT per service person, per property, per request type is more valuable to FM operators than any internal dashboard metric.