Intelligent automation of service requests in property management

An operator responsible for infrastructure maintenance and technical services for large office and residential properties. The client was looking for a solution to automate fault handling and shorten response time without involving office staff.

The company managed thousands of square meters of office and residential space, where small defects requiring repair occurred daily, such as lighting failures, minor damage, leaks, etc. Traditionally, requests were passed on by phone or email to the office.

The process was multi-stage and each time involved several employees, who had to:

• record the content of the request and send it by email or relay it by phone to the office;
• classify it into the appropriate category (e.g. plumbing, electrical);
• find the right service company and send the request.

This process was time-consuming and prone to errors, and some defects were “lost” or forwarded too late. In addition, the subcontractor data was often not updated, causing delays in defect removal

After analyzing the client’s situation, we decided to implement a two-component system consisting of:

• a mobile application for field employees – enabling quick dictation of requests during inspections. The application recorded the description dictated by the person responsible for performing the daily property inspection.
• an AI agent – using the Whisper speech recognition model and the GPT-4o AI model to analyze and classify the request.

Now the process is as follows:

• The employee conducting the inspection uses the application to dictate requests by voice, describing all observed problems at once. No training or onboarding is required — the simplicity of the application is its strength — it only needs a request in the form of free speech. The application sends the recording to the server, where our speech processing algorithms (NLP) convert the information into further instructions. In the first stage, the Whisper model converts speech into text, then the GPT-4o model interprets the request.
• After in-depth analysis, the system classifies the request into one of the predefined — and fully configurable — categories. In case of doubt, the system sends a request to the more advanced o3 model, which helps make the final decision.
• Then the AI agent distributes the requests — sends email notifications to trusted partner companies, which are stored in an independent database, dealing with the given type of repairs, providing the exact location of the defect, a description of the problem, and the priority. The company database can be changed by a person with no programming or AI experience. By default, we use the SendGrid system to avoid email blocking or ending up in spam.

• Submission and forwarding of information to the contractor within a few minutes without involving office staff.
• Consistent documentation – each request has a uniform format and goes into the central database.
• Automatic prioritization – the system distinguishes urgent failures from minor defects.
• Greater control over building maintenance – easier analysis of which locations generate the most problems.
• Reduced risk of errors and delays – thanks to automation there is no risk of losing or misdirecting a request.
• Relief for administrative staff, who can now focus on coordinating repairs rather than handling requests.The automated service request system has revolutionized the way the company manages facility maintenance. The combination of a mobile application, speech recognition, and an AI classifier has delivered real time savings, reduced errors, and improved documentation quality — all without additional burden on the team.