I built an AI receptionist for a mechanic shop
An experiment to replace a small auto shop’s missed phone calls with an AI “receptionist” raises questions about whether LLM-powered voice agents are a good fit for real customers. Commenters debate trade‑offs between cost savings and perceived service quality, highlighting reliability risks, hallucinated quotes, legal exposure around pricing, and the mismatch with “luxury” branding, while others note that many tradespeople are already too busy and might be better served by human or simpler alternatives like answering services, web forms, or smarter voicemail. Some see real promise in tightly scoped, well‑grounded agents for routine tasks like scheduling and FAQs, but emphasize that edge cases, trust, and escalation to humans remain critical.
Perception of “Luxury” and Brand Impact
- Many argue that using an AI receptionist undermines any claim to a “luxury” experience; high-end clientele expect human, high-touch service.
- Others point out the shop is really “European”/boutique, not luxury in the sense of Bentley/Rolls dealerships, so some brand‑damage concerns may be overstated.
- Several feel that for genuinely high-value customers, an AI frontline is a negative quality signal.
Customer Experience & Trust
- Numerous commenters say they hang up as soon as they detect a bot, especially for nuanced or urgent issues.
- Some report good experiences with well-implemented LLM agents (e.g., telco support, prescription refills) and see them as superior to legacy IVRs or long hold times.
- There is worry about uncanny voice, overlapping speech, and failure modes (misunderstood addresses, wrong promises, repeated questions).
Business Case vs Alternatives
- Many suggest simpler, cheaper options: voicemail, email, web booking forms, Calendly-style schedulers, or long‑standing “telephone answering services”/virtual receptionists.
- Several argue that if missed calls truly represent “thousands per month,” hiring or outsourcing a human receptionist is straightforward and more reliable.
- Others note some small tradespeople are already at capacity and actively don’t want more work, so capturing every lead may be pointless.
Technical Architecture & Reliability
- Multiple people say RAG is overkill; the shop’s info likely fits in a context window. RAG is defended as a learning exercise and for possible latency benefits.
- Strong doubt that “no hallucinations” is achievable; guardrails and “if you don’t know, say so” are seen as necessary but insufficient.
- Concerns raised about prompt injection, misquoting prices, legal/expectation issues with “estimates,” and the risk of mismanaging someone else’s livelihood.
Meta-Discussion and Reception of the Post
- Some appreciate the project as a practical experiment and source of implementation ideas.
- Others criticize it as over‑engineered, possibly AI‑written, and functioning as marketing for courses/SaaS templates.
- Several ask for real metrics, call recordings, and evidence it is actually deployed and beneficial; this remains unclear.