Phonely's Custom LLMs Handle Millions of Calls, Fool 80% as Human
Phonely optimizes voice AI agents with custom modular LLMs and data analytics, processing millions of calls/month across verticals like call centers and insurance; 80% of callers mistake it for humans, with statistical tweaks boosting outcomes 5%. Raised $16M Series A.
Data-Driven Optimization Beats Conversational Quality Alone
Phonely's platform doesn't just answer calls—it statistically improves agent performance on business outcomes like lead qualification and appointment booking. Customers in call centers, insurance, and home services use it to sift inbound leads from billboards and ads, handing off only when needed (e.g., licensed insurance sales). By analyzing millions of calls/month across hundreds of verticals, Phonely surfaces actionable insights: changing one question recently lifted a customer's outcomes by 5%. This full-stack approach—answering calls plus tools for iteration—differentiates from generic voice AI, as early-mover battle scars enable production reliability where others falter.
Modular Custom LLMs Solve Latency, Cost, and Latency Without Quality Loss
Instead of monolithic models like OpenAI's, Phonely deploys specialized smaller LLMs for voice AI components: one for storing variables like names/emails, others for tasks needing context. Running on fast inference hardware like early Groq integration, this cuts latency to 'good enough' levels (no longer a bottleneck) while matching big-model quality at lower cost. Models switch by call phase or function, not per question, enabling isolated updates. Result: 80% of callers can't tell it's AI today, rising to near 100% by year-end as conversational accuracy and edge-case handling (interruptions, garbled audio, transcription) improve. Prioritize inbound revenue calls over support; disclose AI for outbound due to ethics/regulations.
Pivoting from SMB Feedback to Enterprise Scale
Started as AI receptionist for dad's small practice (no solutions existed), iterated 4-5 months on $30-100/month SMBs for rapid feedback, then pivoted when one call center outpaid all SMBs combined. PhD roots in voice AI (pre-'voice AI' term) and athlete grit fueled custom model builds. $16M Series A from Base10 Partners (sparked by LinkedIn ultra-endurance post) funds 50M+ calls/month goal, SF sales/engineering hires. Low-ego team culture emphasizes problem-solving.
Startup Realities: Endless Fight, Create Your Own Luck
Success feels like daily war—new models/competitors loom—even at top. Outsiders see LinkedIn wins; insiders grind. Advice: Test founder fit at startups first; if compelled, roll dice daily to create luck. Telephony nuances (not just LLMs) moat against generic models.