The setup
- Multilingual agent (English/Arabic on day one; 12 languages today), female voice, answering our web-call line
- Qualifies callers: name, contact, business, what they want an agent to do
- Books intro calls straight into the founder's calendar
- Every call: recorded, transcribed, analyzed, and written to our own dashboard
Real numbers from day one
- A ~3.5-minute qualification call cost about $0.53 in voice + AI, coffee money for a captured, qualified lead
- The full loop worked unattended: call → transcript → structured analysis → lead in the inbox with name, email, and intent
- Time from 'we should dogfood this' to the agent answering live: under a day
What broke (the honest part)
Day one exposed exactly the issues a pilot should: the agent initially had no way to hang up after wrapping a call (dead air until the caller gave up); a payload limit silently ate the first real booking; and open language auto-detection misread a caller's Arabic as Hindi until we constrained recognition to English + Arabic.
Every one of those fixes is now baked into every agent we ship. That's the actual argument for dogfooding: our own line finds the sharp edges before a client ever does.
Why this is our proof
We don't have logos to borrow yet, we have something better: you can call the product right now and judge it yourself. When our first client pilots convert, their numbers will appear here with permission. Until then, the demo IS the case study.