AI Phone Assistant for Property Management

Artificial Intelligence

The Challenge

A property management company looking after around 6,500 residential units repeatedly hit the limits of its phone support. Call volumes spiked at quarter-end, during the annual service-charge statement, and after weather events. More than 70% of conversations revolved around a few recurring requests: reporting damage (a dripping tap, a broken heater, a lift breakdown), submitting a meter reading, scheduling a tradesperson, notifying an address change, or checking case status. At peak times tenants and owners waited 12 to 18 minutes on hold, and many hung up in frustration. Meanwhile, experienced property managers spent much of their day answering the phone and entering data instead of actually managing the buildings. A classic IVR phone menu ('Press 1 …') was already in use, but it was perceived as rigid and cumbersome and regularly failed with dialects, follow-up questions, or multiple requests in one call.

Our Solution

Solution insights

We developed a voice-based AI phone assistant that answers and understands incoming calls in natural language. Connected to the existing phone system via SIP, it identifies the request, asks targeted follow-up questions, and guides tenants or owners through the process – without a rigid keypad menu. For information on procedures, deadlines, and the service-charge statement, the assistant uses RAG to draw on a curated knowledge base of the company's approved documents and states only verified information. Structured transactions such as damage reports or meter readings are created directly as a ticket or case in the property management software through secured interfaces, with every handover logged. For urgent cases – a burst pipe, a heating failure in winter, or someone trapped in a lift – the system recognises the urgency by clear rules and immediately routes the call to the emergency service or a human agent, including a conversation summary. Data protection was a foundation from day one: speech processing and the language model run in an EU region, conversations are processed for their purpose only, and at the start of each call the assistant transparently states that it is an AI assistant.

Results

During a four-month pilot, the AI phone assistant handled around 62% of incoming routine calls fully without human involvement. Average hotline waiting time dropped from over 12 minutes to under 90 seconds, because urgent damage and complex requests now reach property managers faster. Damage reports and meter readings are accepted around the clock – including evenings and weekends, when the office used to be closed. Staff report noticeably less routine pressure and more time for actual property management. After the learning phase, the correct-intent recognition rate was around 91%; calls where the assistant was unsure were reliably handed to humans rather than guessed.

Features

Feature overview

• –Natural-language call handling without a rigid keypad menu
• –Damage reporting with structured capture and read-back confirmation
• –Meter reading (heating/water) with value read-back confirmation
• –Scheduling with tradespeople and service providers
• –Information on service-charge statements and deadlines via RAG, based only on approved documents
• –Immediate routing to the emergency service for urgent cases (e.g. burst pipe, heating failure)
• –Seamless handover to staff including a conversation summary
• –Creation of tickets and cases directly in the property management software
• –Transparent AI notice at the start of each call
• –GDPR-compliant speech processing in an EU region
• –24/7 availability – also outside service hours
• –Analytics dashboard for requests, recognition rate, and handover rate