This Startup Built AI That 80% of Callers Think Is Human
TL;DR
Phone AI sells optimization, not just answering calls — Will Bodewes says the real product is a platform that helps businesses improve voice-agent outcomes over time, including finding prompt changes that lifted a customer’s performance by 5%.
They’re already at massive scale across hundreds of verticals — Phone AI handles millions of calls per month for customers in call centers, insurance, and home services, mostly where inbound phone leads directly affect revenue.
Their technical edge is a custom multi-model stack built for latency and control — instead of relying only on closed models, Phone AI worked with Groq and split voice tasks across smaller specialized models to reduce latency, lower cost, and update components independently.
Bodewes claims the human/AI line is already blurry on the phone — he says about 80% of callers don’t realize they’re talking to AI today, and predicts that by year-end that number could get close to 100%.
The company found product-market fit by starting embarrassingly small — they first sold a receptionist-style product to small businesses for $30 to $100 a month, then pivoted hard after one call center customer paid more than all those SMBs combined.
The founder story is pure grit: canceled NCAA race, failed first startup, PhD in Australia, then a LinkedIn post led to Bessemer — Bodewes traces Phone AI back to watching his dad struggle to answer phones, and says Bessemer’s Caroline first reached out after seeing his ultra-endurance cycling post.
The Breakdown
From “AI answering your phone” to a performance engine
Will Bodewes opens by saying Phone AI is more than an AI receptionist. The pitch is that businesses don’t just want calls answered — they want voice agents that measurably improve outcomes, and he says Phone AI can show that statistically. His favorite example: changing a single question on a call recently improved one customer’s results by 5%.
Millions of calls, and mostly where phones equal revenue
The company is already handling millions of calls every month across hundreds of verticals, with especially strong traction in call centers, insurance, and home services. Bodewes keeps bringing it back to the same point: for a lot of businesses, the phone is still the front door to revenue. Think billboard leads, appointment booking, qualification — all the stuff companies cannot afford to fumble.
The founder backstory is unusually intense
Bodewes’s path to voice AI is not the standard founder arc. He was a college cross-country skier, made the NCAA level, then had his final race canceled because of COVID — a moment he describes as leaving him with a chip on his shoulder. After a first startup didn’t work, he took a full-ride AI PhD in Australia, partly because he believed AI was going to be the next big thing.
The spark was watching his dad miss calls
The original idea came from home. Bodewes says his dad built a small practice from their garage, and when asked what AI could solve, his answer was simple: “If somebody could answer my phones, that would be amazing.” Nothing on the market really did that yet, so Bodewes started building it.
They started with tiny customers, then one call center changed everything
Phone AI’s first version was aimed at small businesses paying $30, $50, maybe $100 a month. The point wasn’t the revenue — it was tight feedback loops and fast iteration. After four or five months, they landed a call center customer that paid more than all their small-business users combined, and that made the market focus obvious.
Why they replaced off-the-shelf models with their own stack
Bodewes says most voice AI companies just plug in OpenAI or another large model, but Phone AI went a different way. Working early with Groq for fast inference, they built an architecture where smaller models handle different jobs — like storing variables or managing other call components — instead of one giant model doing everything. That gave them lower latency, lower cost, and cleaner ways to improve specific parts of the system.
Voice AI is “good enough” on latency — now the fight is quality
Latency used to be one of the biggest bottlenecks, but Bodewes says it now feels like oxygen: you still need it, but it’s mostly there. The harder problems are conversational quality, transcription, interruption handling, endpoint detection, and all the ugly phone-audio edge cases. On realism, he makes the bold claim that around 80% of customers already don’t realize they’re speaking to AI.
Regulation, inbound use cases, and the founder-as-war metaphor
On disclosure, Bodewes says outbound AI calls should probably be disclosed and expects regulation there, even though he thinks consumers may eventually prefer AI for sensitive or awkward questions. Most current adoption is inbound, where AI can qualify leads or schedule appointments, though regulated cases like insurance still require licensed humans for the final handoff. The interview ends with fundraising news — a $16 million Series A led by Bessemer, sparked by a LinkedIn post about his 300-mile-plus ultra-cycling races — and a very founder-coded closing line: starting a company is “going into war,” and you keep rolling the dice until you create your own luck.