Case study § 01 · In production

AI receptionist for a gym

Product · voice · intent detection · slot booking · member record

Context

A gym receives a steady stream of member calls at the front desk: slot bookings, questions about their record, practical information on hours or classes. Every missed or misrouted call is a lost service moment and an interruption for the on-site team.

The client needed a French-language voice receptionist able to understand caller intent without a rigid decision tree, and to act directly rather than simply routing the call.

The constraint was operational: the system had to work in real conditions, during opening hours, without degrading the front-desk experience for members.

Three systems cover most requests, slot booking, member record, and practical information, with a fallback output: handoff to the on-site team.

The system

The AI receptionist understands caller intent in natural French, without a rigid decision tree, and resolves it from the first exchange.

Diagram: AI receptionist for a gymA member call is handled by the AI receptionist, which understands voice in French and detects intent. It acts on three connected systems: the booking system for time slots, the member record, and practical info. When the call exceeds what it can handle, it is escalated and transferred to the on-site team.ESCALATIONMEMBER CALLAI RECEPTIONISTVOICE · FR · INTENT DETECTIONBOOKING SYSTEMTIME SLOTSMEMBER RECORDCLIENT FILEPRACTICAL INFOON-SITE TEAMTRANSFERDiagram: AI receptionist for a gymA member call is handled by the AI receptionist, which understands voice in French and detects intent. It acts on three connected systems: the booking system for time slots, the member record, and practical info. When the call exceeds what it can handle, it is escalated and transferred to the on-site team.ESCALATIONMEMBER CALLAI RECEPTIONISTVOICE · FR · INTENT DETECTIONBOOKING SYSTEMTIME SLOTSMEMBER RECORDCLIENT FILEPRACTICAL INFOON-SITE TEAMTRANSFER
DIAGRAMIN PRODUCTION · 20–30 CALLS / DAY

A gym receives a steady stream of member calls at the front desk: slot bookings, questions about their member record, practical information on hours or classes. Every missed or misrouted call is a lost service moment and an interruption for the on-site team. The AI receptionist understands caller intent in natural French, without a rigid decision tree, and resolves it from the first exchange.

The system does not just route the call: it acts directly across three connected systems, the booking system, the member record, and practical information. When a request falls outside that scope, the call is transferred to the on-site team, an explicit output rather than a hidden failure. In production, 20 to 30 calls are handled every business day, in real front-desk conditions.

Decisions & trade-offs

Key decision

The AI receptionist connects directly to the three systems that matter to a member: slot booking, member record, practical information. It acts inside these systems, rather than simply routing the call to someone else.

Trade-off

A natural French voice paired with intent detection was chosen over a menu-driven flow, so members can speak naturally. The handoff to the on-site team remains an explicit, intentional output, not a hidden failure, when a request goes beyond what the system can handle directly.

Result

20–30 calls / dayHandled in real front-desk conditions, every business day

In production. 20 to 30 calls handled every business day, in real front-desk conditions.

What it shows

Shipping a voice system into production that acts directly inside a real client's systems, rather than simply routing the call.