The first Phonely customer paid $7 a month. The company says the deal was written on a pizza box. It is an amusing origin story, though the interesting part is what was being sold: the right to have a phone answered when nobody was there to pick it up. A great many businesses have spent decades treating that small gap between a ring and a reply as a staffing problem. Two AI researchers thought it could be a software problem.
- Phonely builds AI agents that answer, route and act on business calls.
- Customers range from appointment-led small businesses to large contact centers.
- Plans start free; paid self-service tiers list $50 and $150 a month.
- Its $16 million Series A included three companies already using the product.
Will Bodewes and Nisal Ranasinghe met in an AI research group at the University of Melbourne. In 2023, an after-hours demo for a couple of small businesses became Phonely. The company joined Y Combinator's Summer 2024 batch and now operates from San Francisco. Its proposition has grown from an AI receptionist into something broader: a place to build, run and improve phone conversations.
What happens after hello
A business can give Phonely its website and create a basic phone agent that answers common questions. From there, it can connect the agent to a calendar, a customer relationship system or an existing phone setup. The caller can book an appointment, ask about a service, be qualified as a lead or be transferred to a person. The outcome can land in a transcript, a summary or a CRM record. Phonely also extends some of the same workflows to SMS and web chat.
That last step is the reason to pay attention. A traditional receptionist can tell you that a caller sounded interested; a phone tree can tell you which button they pressed. Phonely can test two versions of a question and compare the outcomes. In its financing announcement, the company said a single change to a call-script question lifted conversion by 5%. This is a company-reported result, not a promise for every customer. It suggests a useful way to think about the product: the conversation itself has become an interface that can be edited.
One product, two very different desks
At one end is the small business that cannot staff a desk all night. Signpost, which works with local businesses, reported that its partners answered every call and tripled lead capture with Phonely. At the other end are contact centers whose problem is thousands of repetitive calls, not a single missed one. The company names Etech Global Services, TSA Group and Engage CX among its enterprise customers. Healthcare practices, insurers, law firms and service businesses also appear in its use cases. The jobs vary, but they share a test: did the caller get the thing they called for?
The commercial arrangement follows that divide. A free plan lists 100 minutes a month. On monthly billing, Starter is $50 and Professional is $150, with more features and usage allowances; call-minute overages can change the bill. Enterprise contracts add assisted setup, custom integrations, specialized models and negotiated rates. For a one-location practice, the question is whether recovered calls justify a subscription. For a contact center, the question is whether automation improves cost and resolution at the same time.
The callers became shareholders
In April 2026, Phonely announced a $16 million Series A led by Base10 Partners, bringing its stated total funding to $19 million. Y Combinator participated. So did Etech, TSA Group and Engage CX. A customer investing in a supplier is no substitute for audited performance data, but it is a more revealing signal than a testimonial alone: these organizations were buying the service and a piece of the company providing it.
“We have 4,500 human agents, and we didn't believe AI could match that level of conversational quality at scale.”Xander van der Westhuizen, TSA Group, quoted by Phonely
Phonely says TSA Group uses its agents for more than 15,000 calls a day and reported a 74% reduction in operating cost for the relevant workflow. Etech reported a 72% cost reduction and a 34% lift in first-call resolution. Engage CX, according to Phonely, drove more than $10 million in insurance policy sales through the platform in the first four months of 2026. These are customer cases presented by the company, and their denominators matter. A routine policy question and a distressed claim are different conversations, even when both arrive at the same number.
The pause is the product
Speech AI has an awkward physical property: the caller can hear the computer thinking. A slow reply feels like a dropped line. Phonely says it ran into this with general-purpose, closed-source models. Its published technical account describes a move to customized open models using Maitai's model platform and GroqCloud inference. In its benchmark, the ninetieth-percentile time to first token fell from 661 milliseconds to 179 milliseconds by the third model iteration. Measured accuracy across that same series rose from 81.5% after the initial switch to 99.2%. Those numbers describe Phonely's selected evaluation, not a universal score for every call.
The technical trick is not simply a faster chip. Maitai and Groq describe multi-LoRA support that lets one inference instance switch among customer-specific model adaptations without adding latency. It gives Phonely a way to serve distinct enterprise workflows while keeping a live conversation moving. Its other differentiator is less exotic and perhaps more useful: the builder, integrations, call records and experiments sit in one operating surface. Competing approaches include hiring receptionists, outsourcing calls, using old menu-based IVR and buying another voice-agent platform. Phonely's wager is that the agent and the feedback loop should be sold together.
Try a small, honest call first
There is a practical lesson in the pizza-box sale. Start with a narrow call that is easy to recognize and safe to complete: opening hours, a straightforward appointment or a lead that needs a callback. Give the agent the same knowledge a capable employee would need. Connect the calendar or CRM before asking it to promise a result. Then listen to real calls, count completed tasks and inspect transfers and failures. Test a script change against an outcome, not against whether the voice sounds impressive in a demo.
Conditions matter. Complex, emotional or regulated conversations need clear escalation to qualified people. An agent cannot book a real appointment if it cannot see the schedule, and multilingual fluency does little good if the underlying policy answer is wrong. The economics are also sensitive to call length, overage minutes and the value of a successfully resolved call. Phonely's ability to run both a tiny receptionist and a high-volume contact-center deployment is noteworthy; neither customer should assume the other's results will transfer automatically.
The phone is an unusually stubborn piece of infrastructure. People still use it when the matter is urgent, confusing or too important for a form. Phonely's story is about making that old channel measurable. The $7 customer bought an answer. The large customers are paying, and in three cases investing, for what happens after it.