AI agent for service request management
A technical maintenance company was receiving 40 service requests per week by email and WhatsApp. The owner answered them himself, often 48 hours later. Clients were leaving for faster competitors. We fixed that.
Representative scenario based on real engagements. Names, figures and contexts have been adapted for confidentiality reasons.
The problem
A 12-person SME specializing in commercial building maintenance. Good reputation, full order book, efficient field teams. But one invisible bottleneck: all incoming requests went through the owner.
40 requests per week on average. Email, WhatsApp, sometimes both for the same job. The owner triages, responds, schedules - usually at the end of the day after being out on site. Average response time: 36 to 48 hours.
The result: prospects who don’t call back. Clients who try a “more responsive” competitor. And a manager spending his evenings on his phone instead of running his business.
What we built
An AI agent that handles first contact on his behalf, 24/7.
Automatic triage - the agent reads every incoming message and identifies whether it’s an emergency, a quote request, a follow-up, or something else. It classifies and prioritizes in real time.
Immediate response for standard cases - for common requests, the agent responds within a minute with the relevant information and a proposed time slot. The tone matches the company’s.
Smart escalation - urgent cases are flagged via SMS to the owner with a summary. Complex cases are held with full context ready for human handling.
Daily briefing - every morning, an email summarizes the previous day’s requests, responses sent, and what’s waiting for a decision.
The unexpected challenge: a significant share of requests came via WhatsApp with photos of the fault attached. Integrating image analysis to automatically assess urgency wasn’t in the original scope - it required an additional iteration. It’s now one of the system’s strongest features.
Results
- Average response time: 48h → under 10 minutes for standard cases
- Automatic handling rate: 73% of requests without owner involvement
- Revenue grew over the two following quarters, partly from opportunities that were no longer falling through
- The owner stopped managing his inbox in the evenings
Stack
- Custom AI agent - Python, LLM pipeline
- Integration with the company’s email and WhatsApp Business
- Image analysis for visual urgency assessment
- Knowledge base covering intervention types and service areas
- SMS alerts via API