The problem: after-sales overwhelmed by repetitive requests
Mobilier de France was receiving hundreds of calls and emails per week for recurring questions: delivery status, return timelines, a visible defect on a received item. Most of these requests didn't need a human agent — they needed a fast answer and structured follow-up.
What we built: the Mobilier de France SAV-Bot
The bot combines several building blocks designed specifically for the furniture sector:
- Multilingual dialogue: French, English, Arabic, Italian, German — critical for a retailer with an international customer base
- Smart product recommendations based on catalogue and purchase history
- Photo upload + visual defect analysis: the customer photographs the damaged item, the bot pre-qualifies the issue
- Voice support: speech recognition and synthesis for customers who prefer talking over typing
The priority system that changes everything
Every ticket the bot generates is automatically classified into 4 priority levels, each with a target resolution time:
- P0 — Emergency (safety, severe defect): immediate team notification
- P1 — High priority (product unusable): handled within 4h
- P2 — Normal (functional issue): handled within 24h
- P3 — Low (cosmetic, informational): handled within 72h
This automatic classification prevents a seemingly minor safety issue from sitting behind ten generic delivery-timeline questions.
Deployment numbers
The bot runs on a FastAPI + React + PostgreSQL + Redis stack, containerised and deployed on Railway. On the quality side, the project is covered by over 351 automated tests (62% backend coverage, 53% frontend) with a full CI/CD pipeline — no blind deployments.
Measured result after going live: −65% inbound after-sales calls, with most status and return requests now handled directly by the bot.
What we took away
Three lessons we now apply to every new after-sales project since this deployment:
- A ticket priority system isn't optional — without one, everything ends up feeling like an emergency
- After-sales customers want to show, not just describe: photo upload drastically cuts back-and-forth clarification
- Multilingual support isn't a nice-to-have in France — it's often the first barrier to first-contact resolution